1. Privacy at a Glance
General Information
The following information provides a simple overview of what happens to your personal data when you use this website. Personal data is any data by which you can be personally identified.
2. Hosting and Content Delivery
Vercel
We host our website with Vercel Inc., 340 S Lemon Ave #4133, Walnut, CA 91789, USA. Vercel is certified under the EU-US Data Privacy Framework.
Supabase (Database)
Application data is stored with Supabase Inc. Our database server is located in Frankfurt am Main (eu-central-1). Supabase is GDPR-compliant and has entered into the corresponding data processing agreements. Uploaded documents and files are also processed via this infrastructure in access-protected storage areas.
finAPI (optional banking integration)
For the optional banking integration, we use finAPI GmbH, Adams-Lehmann-Str. 44, 80797 München, Germany. finAPI provides the open-banking/PSD2 services required to connect bank accounts, in particular hosted web forms for bank selection, authentication, strong customer authentication (SCA), and management of consents on the bank side.
Bank access data, TAN data, and other SCA data are generally entered directly in the interface provided by finAPI or with the connected bank and are not stored in struct-i-vio as regular access credentials.
AWS S3 (separate file and database backups)
For additional backup copies of uploaded files and encrypted backup copies of the production application database, we use separate storage provided by Amazon Web Services EMEA SARL ("AWS"). These backups are stored exclusively in the EU region Stockholm (eu-north-1). Transmission is encrypted; the backup copies are subject to strict technical access restrictions, are not part of regular user access, and serve exclusively to ensure resilience, emergency preparedness, and recovery.
3. Controller
The controller responsible for data processing on this website is:
Tigran Varazhian (structivio)
Viktoriastraße 22
80803 München
Email: tigran.varazhian@struct-i-vio.com
4. Your Rights
You have the right at any time to:
- obtain access to your stored data (Art. 15 GDPR)
- request the rectification of inaccurate data (Art. 16 GDPR)
- request the erasure of your data (Art. 17 GDPR)
- request the restriction of processing (Art. 18 GDPR)
- data portability – export of your data (Art. 20 GDPR)
- object to processing (Art. 21 GDPR)
To exercise your rights, please contact us at: tigran.varazhian@struct-i-vio.com
5. Data Collection on This Website
Cookies
We use technically necessary cookies for authentication and session management. The selected display language is stored in a cookie so the interface can be shown in that language. This cookie is not used for tracking.
Local drafts for ideas
When you start drafting a feature request using the ideas button, the application stores the title you enter, the description, the displayed page context and up to three screenshots or other permitted files you select exclusively in your browser's local storage (IndexedDB) on this device. These data are not transmitted to us before you explicitly submit the request. The draft is assigned to the signed-in person and the currently selected tenant so that it is not mixed between accounts or tenants.
A local draft last edited more than seven days ago is detected and removed from the browser the next time the feature is used. It is also removed from the browser when explicitly discarded and on sign-out. It is not backed up on the server. On shared devices, a local draft may be visible to other people using the same browser profile. Please therefore only add files that may lawfully be shared as part of your feature request.
Server Log Files
The provider of the website automatically collects and stores information in server log files that your browser automatically transmits to us (IP address, browser type, operating system, referrer URL, time of the server request).
6. Data processing in the application
Which data is processed?
- Master data: name, e-mail address, profile picture; personal display-language preference for each user and tenant
- Project data: projects, tasks, documents, time tracking
- Financial data: Budgets, expenses and invoices entered by the customer or submitted using its own tools, and invoice analyses saved from that data
- Bank and account data: optional bank connections, account descriptions, IBAN/account numbers, balances and other account metadata where provided by the connected bank or finAPI
- Transaction data from banking integrations: imported booking data including booking date, amount, payment reference, counterparty and counterparty account IBAN/BIC where provided by the connected bank or finAPI
- Banking and consent metadata: technical connection identifiers, bank connection statuses, consent expiry dates, web form/callback IDs, and sync and error logs
- File and supporting document data: uploaded documents, attachments, supporting documents and profile pictures
- Ideas and feature requests: titles and descriptions entered by you, the page context transmitted upon submission, and screenshots and other permitted files attached by you. This data is stored as a feature request only upon submission and is then processed according to the access, backup and deletion rules described in this policy. Submission may trigger an in-app confirmation and an e-mail to you, as well as in-app and e-mail notifications to currently active and authorised platform Superadmins. The receipt notifications contain the limited title and, for notifications to platform Superadmins, also the submitter’s and tenant’s names; attached files are not sent. As the submitting person, you may also be informed by e-mail about changes to the processing status visible to you. This status e-mail contains only the limited current title, the current processing status visible to you and a link to your idea that requires login and current access rights. Internal processing stages, reasons, internal notes, message histories and attachments are not included. The status e-mail describes the current state, not the history of earlier decisions. Delivery is subject to the respective recipients’ notification settings. This does not grant company administrators additional access to private ideas or conversations.
- Usage data: logins, activities for audit logs, including changes to the personal display-language preference
- AI billing and usage metadata: assignment of a customer-related AI operation (e.g. chat, lead assessment, quote creation, OCR, speech or document processing) to a user and tenant, measured usage and processing components, credit charge, version of the billing basis and, where applicable, reservation, refund, adjustment or correction entries. The credit receipt visible to the customer or tenant generally contains neither the content of the AI processing nor internal calculation bases. Separate internal billing records with restricted access may contain technical parameters relevant to billing, but no AI content.
- Internal messages (messenger): messages, file attachments, reactions, mentions and chat memberships between employees of the same company and — only if activated by the company — activity indicators; details in the section “Internal messenger” below
Legal basis
Where struct-i-vio processes the data described above for its own purposes, it does so on the basis of Art. 6(1)(b) GDPR if processing is necessary for the performance of a contract with you, and on the basis of Art. 6(1)(f) GDPR for providing a functioning application and handling the direct communication requested by you. For ideas and feature requests, struct-i-vio itself is the controller for this direct communication. This includes preparing a draft initiated by you, handling a feature request expressly submitted by you and providing updates on its current processing status visible to you. In contrast, struct-i-vio processes customers’ business data on behalf of the respective customer under the DPA and their documented instructions; the operator notices described here do not expand that mandate.
Feedback on AI answers: when you rate an AI answer, we process the rating you expressly submit, any information you provide about the cause or expected result, and technical association and context data relating to the answer concerned, in order to investigate reported problems and improve the reliability of the service. struct-i-vio itself is the controller for this quality and support work for its own purposes; the legal basis is Art. 6(1)(f) GDPR, our legitimate interest in investigating reported errors and providing a reliable application. Currently active and authorised platform Superadmins may receive an e-mail about a currently negative rating. It contains only a general notice of the negative rating, any reason selected by you as a fixed label, and a link to the access-protected review. Your free text, conversation and answer content, attachments, names of people or companies, and information about pages visited, models and tools are not included. No conversations or messages are retrieved for this e-mail content. Any separate choice you make to share the conversation history for analysis remains an independent decision and does not permit the history to be sent by e-mail. The delivery, retention and recipient-copy limits described for these notifications, as well as recipients’ applicable notification settings, remain binding.
Optional banking integration (finAPI)
If you activate the optional banking integration, we process the bank and transaction data required for it, and technical consent and status metadata, to connect bank accounts, import bookings, renew consents and trace errors.
The actual bank connection takes place through hosted finAPI web forms. Bank login, TAN and other authentication data is generally processed directly by finAPI or the selected bank. struct-i-vio does not store this bank login data as regular product data.
We limit the data transmitted to finAPI to the information required for the respective banking function. For regular use, ongoing transmission of individual users’ personal e-mail addresses to finAPI is not provided, unless exceptionally mandatory for technical or contractual reasons.
To prevent duplicate CRM notes and the recreation of deleted notes through repeated requests, we store necessary note, user, tenant and lead identifiers and creation and deletion timestamps with restricted access. While the note exists, a hash of the original request may also be stored. This hash is removed when the note is deleted. The remaining prevention record contains no note text or file names and is removed when the associated lead or tenant is deleted, or when the associated user relationship is deleted in the erasure process. Repeated requests do not extend retention. The described backup retention periods apply to database backups.
Time-limited preparation of document packages
When you request a document package, we store the job with its link to your user account and tenant, the selected scope, progress, and the data and file references needed for the Excel overview and original documents. To allow the job to be resumed, protected intermediate copies and ZIP parts may be created in the Supabase storage described above. We do not store login tokens or durable download links in the job. Processing is solely for the export you requested; original records and files are not changed.
Only the person who requested the job can resume or download it while signed in and in the same tenant. Before every further preparation step and download, we recheck your current source permissions, record visibility, and file lifecycle. If access is withdrawn or a source is deleted or moved to the trash, the prepared part can no longer be downloaded. An administrator role alone does not grant access to another person’s export jobs. Closing the export dialog or losing the sign-in pauses further preparation; a bounded step already in progress may finish. There is no autonomous background export.
Export jobs and their prepared files can be accessed for no more than 24 hours from job creation. Preparation has a fixed deadline of no more than two hours from job creation. Resuming, retrying, or downloading does not extend these periods. Cancellation, expiry, or withdrawal of access blocks further access. The cleanup process removes generated intermediate copies, ZIP parts, financial data, file references, and the job’s selection configuration. Before the first file deletion, it waits for the permitted runtime of writes already started to end, plus a 45-second safety margin; further checks are permitted no earlier than 60 seconds later. The scheduled cleanup process normally runs every five minutes; actual check and deletion times depend on successful runs and any backlog. Additional limited checks may also run on authenticated export requests. These checks also capture writes that finish late. Failed cleanup attempts are monitored and retried; service issues or a backlog may delay actual removal, while access remains blocked. Generated export files are ultimately removed as temporary processing caches without deleting the original files or moving them to the trash.
To verify cleanup, a minimal technical record is retained for seven days after the normal end of the 24-hour access period, that is, until eight days after job creation; early cancellation does not shorten this fixed retention period. It contains only user, tenant, job, and reservation identifiers; the random private path of the generated file; technical byte and checksum details; and cleanup status, deadline, access-block, and absence records. It does not retain financial records, financial totals or counters, source-file references, selection filters, original filenames, file contents, download links, or sign-in tokens. After the retention period, this record is removed once the generated file’s current absence has been confirmed. Failed deletions and late writes remain monitored until cleanup is confirmed; this may delay actual removal of the minimal record. If, after a restore or a late write, a generated orphan file is found without its associated job record, a separate minimal cleanup record is created solely for the private generated-export storage area. It contains the random generated-object path, a technical identifier, and observation, deadline, access-fence, status, and absence details; it contains no financial records, original filenames, source-file references, or credentials. This separate record’s seven-day period begins when the orphan is discovered. Removal requires fresh confirmation that the generated file is absent and may be delayed by cleanup failures.
Generated intermediate copies and ZIP files are not mirrored to file backups or saved as recoverable export files. However, job metadata, minimal cleanup records, and required export data may remain in existing encrypted database backups for the stated period of up to twelve months. A restore must not make expired, cancelled, or no-longer-authorized jobs accessible again. The 24-hour access period and the cleanup-record retention period do not promise deletion of every copy from backups. Original files continue to follow their own trash, deletion, and backup rules.
Internal messenger
With the internal messenger, employees of the same company (tenant) send each other direct messages or communicate in group chats, project chats and the company chat “General”. They can reply to messages, mention people, react with a fixed set of symbols and attach files and photos. Chats with people outside one’s own company are not possible. The controller for this content is the company that uses struct-i-vio (usually your employer); struct-i-vio processes it as a processor under § 3g of the data processing agreement.
Data processed: message text, sender and time, the markers “edited” or “deleted”, the reference to a message replied to, mentioned people and reactions; file attachments with file name, file type, size and, for images, the image dimensions; chat details (type, name, description, linked project, members and their role in the chat) and your own chat settings (e.g. muted or pinned); in addition, for each member, a continuously overwritten read position that shows you unread messages. Earlier versions of edited messages are not retained.
Who sees content: The content and attachments of a chat can only be accessed by its current members. A direct chat exists exclusively between the two people involved. The members of a project chat are the current members of the project team; anyone removed from the team immediately loses access to the chat and its attachments. All active employees belong to the company chat “General”. Anyone added to a group, project or company chat can also read its previous history. Administrators of your company and struct-i-vio superadministrators do not obtain read access through the application to chats of which they are not members and cannot see who exchanges direct messages with whom. Administrators with the “Manage company chat” right can delete group and project chats and set the company chat “General” to read-only; this does not give them the content of chats of which they are not members or an overview of such chats. Section 10a describes which administrative operations are recorded in the change log.
Editing and deleting: You can edit your own messages; they are then marked as “edited”. If you delete a message, its text and any attached files are removed for all members immediately; it is replaced by the note “Message deleted”, and reply previews no longer show the text either. Deleted messages and their files cannot be restored by anyone in the application, including you and administrators; the same applies to the messages and files of a deleted chat. The files are technically kept in the system recycle bin for up to 30 days and are then finally deleted (for details, see section 10).
Notifications: struct-i-vio informs you about new direct messages and mentions only within the application (bell and counter); no e-mails or push notifications to mobile devices are sent for this. A notification contains the sender and a reference to the chat, but no message text, and is only displayed as long as you are a member of the chat. Muted chats do not generate notifications. The messenger does not generate desktop notifications via the browser or your device’s operating system.
Typing indicator: While you are writing in a chat, other members who currently have this chat open see the transient notice “typing …”. It is not stored.
Activity indicators (optional, switched off by default): Only if an administrator of your company with the “Manage app settings” right switches on the activity indicators does the messenger show other employees of your company, within the messenger, whether you currently have the messenger open (“online”; transient, not stored) and on which day you last had it open (“last online”; day only, e.g. “today”, “yesterday” or “3 days ago”, without a time; for this, a single, continuously overwritten date is stored per person), and show senders in shared chats whether you have read their messages (in direct chats a marker without a time; in group, project and company chats only the number of readers, without names). The online indicator and the typing indicator are based on signals sent by the devices of the people involved; they are not proof of presence, working time or activity and may in individual cases differ from actual use. You can hide each of these indicators for yourself; you will then not see the corresponding indicator for others either. As long as the indicators are switched off or you have hidden an indicator, the corresponding information is not collected; stored day information is deleted when the indicators are switched off or hidden. There is no history, no export, no evaluation, no filtering or sorting by this information and no display outside the messenger. Independently of this, your own read position serves only your own unread display.
Transfer to tasks and no AI access: If you explicitly create a task from a message, the text is copied into the task form; once saved, it is part of the task and visible to everyone who may see this task. The AI functions of struct-i-vio, external AI clients and the programming interface have no access to messenger content (see section 12).
Legal basis and responsibility of your company: Your company, as controller, determines the legal basis for internal communication (for example Art. 6(1)(b) or (f) GDPR). Your company decides whether the messenger may also be used privately; the application does not distinguish between business and private messages. The use of the messenger and in particular the activation of the activity indicators may be subject to co-determination by the works council (§ 87(1) no. 6 BetrVG); your company is responsible for this.
Storage period: Messages and attachments are not deleted automatically after a set period. They remain until the person who wrote them deletes them, the chat is deleted or the company deletes its tenant. Anyone who leaves the company is removed from all chats; their messages remain visible to the other members and are marked as messages from a former member; the history of a direct chat remains available to the other person. When a user account is deleted, the link between the messages and the person is anonymised. In-app notifications already delivered to other members may continue to contain the sender’s name as displayed at the time of delivery, even after their messages or user account have been deleted, as with other notifications in the application; they contain no message text and are shown only to the recipient, as long as the recipient is a member of the chat. A chat with no remaining members is deleted. For the recycle bin and backup copies, see section 10.
Access and export: You can view your own chats in the application at any time. There is no export of other people’s chats by administrators. Please address requests under Articles 15, 17 or 20 GDPR concerning messenger content to your company; struct-i-vio supports it as processor without giving administrators insight into other people’s chats.
6a. Banking: separate agreement with finAPI and further processing
For the optional account information service, end users enter into a separate user agreement with finAPI GmbH. finAPI provides this service as an independent controller and is not a subprocessor of struct-i-vio for it. struct-i-vio processes the account and transaction data transmitted by finAPI in the tenant as the customer’s processor for its operational financial management, in particular to import, store, analyze, categorize and allocate postings.
The finAPI user agreement and finAPI privacy notice apply to the account information service. Consent to that service and a company instruction to struct-i-vio do not replace any required legal basis for processing personal data of employees, payment counterparties or other persons.
Account and transaction data stored in struct-i-vio may also be retrieved via the API and external AI clients within the permissions granted and edited within struct-i-vio. Retrieved data is sent to the connected system and possibly its provider. The API and MCP provide different functions and data: the API also enables authorized persons to export CSV files and retrieve attachment contents; MCP does not provide attachment contents or CSV exports of bank transactions. Sections 12c and 12e explain the scope, free-text risks and responsibility for external recipients. Changes in struct-i-vio do not change postings at the bank or initiate payments.
7. Data Processing Agreement
For business customers, we enter into a data processing agreement (DPA) in accordance with Art. 28 GDPR. You can obtain it at: DPA
8. Subprocessors (subcontractors)
We use the following subcontractors:
| Provider | Purpose | Location |
|---|---|---|
| Supabase Inc. | Database, authentication, storage | Frankfurt (eu-central-1) |
| Amazon Web Services EMEA SARL (AWS) | Separate file and database backups for resilience and recovery; user-requested temporary PPTX/PDF processing by the deck renderer; after a separate migration, transactional platform emails via Amazon SES (notifications, invitations, public request confirmations, demo access links and authentication). | Backups and SES-configured sending/feedback resources: Stockholm (eu-north-1); database, Supabase Edge processing configured for this mail path, and Vercel mail callers: Frankfurt. Verify the renderer’s production EU location before activation. Regional configuration does not guarantee that all ingress, provider and recipient paths are processed exclusively in Europe. |
| Vercel Inc. | Hosting, CDN | EU (Edge Network) |
| Plus Five Five, Inc. (Resend) | Transactional platform emails not yet migrated to SES, including notifications, invitations, public request confirmations, demo access links and authentication; no bulk marketing. Migration is channel by channel; Resend remains involved for messages actually still sent through it. | USA: including message content, delivery logs, webhook and account data. A European sending region does not change the storage location. The provider DPA includes EU Standard Contractual Clauses for required transfers to third countries; a valid EU–US Data Privacy Framework mechanism may also apply. |
| Stripe Technology Europe, Limited | Payment infrastructure (backend-side; customers do not enter into a self-service subscription via Stripe) | EU (Ireland) / USA (DPF) |
| Microsoft Deutschland GmbH | AI assistance via Azure OpenAI Service, speech recognition/dictation via Azure AI Speech and transient realtime audio transport via Azure Container Apps (each optional and used only when the respective AI function is used) | Registered office: Germany (Munich); data location: EU (Sweden Central, Microsoft EU Data Boundary) |
Transactional platform emails and operational notifications
If your email channel for a notification is enabled, the message may by default contain the necessary current event information, such as a title and bounded description, the specific comment, names and roles of those directly involved, authorized counts, status and deadlines or, with the corresponding financial access, amount and currency. This requires verified current source and field permissions and the processing basis applicable to your company. No additional confirmation for each event family is envisaged. Links still require login and current permissions; later permission changes do not remove content already delivered. The accompanying limits on free text, safeguards, providers, transport and retention continue to apply. Private Ideas and support communication directly with the operator retain their own basis; company administration provides no access to them. For your own personal activity routine, you may also choose to receive a copy of the single completed activity report at your current confirmed account address. This choice is initially off for each routine; the report remains in chat too. Other recipients, CC/BCC, new report data sources, or later emails for past runs are not provided. The personal delivery request is not GDPR consent or acceptance of a company contract. The processing instruction actually in force remains a prerequisite; the choice does not broaden narrower agreements. If a source used in the report has changed, been deleted, is no longer fully readable, or is unclear, the entire report email is withheld. Harmless changes can also cause this. Turning it off stops sends that have not started; your own email delivery can still be turned off after loss of general module or administration rights. Turning it on again does not send past runs. Email retries do not create a new AI report or charge. CRM monthly reports and personal CRM day plans keep their own rules.
Task descriptions and comments may contain free text; reliable automatic detection or removal of all sensitive information entered there is not promised. These messages do not retrieve attachments, file contents, full conversation histories, credentials or tokens, or protected source fields concerning salaries, individual pay rates, full bank access/account details, health, diagnoses, reasons for absence or other specially protected personnel information. Your company organizes lawful input, necessary content and suitable recipient mailboxes. Particularly sensitive content still requires suitable protected procedures and, where necessary, additional purpose-specific grounds and safeguards. The optional activity report copy is one already generated, limited report based on the existing sources of aggregated audit activity and authorized invoice details, not a full conversation history. Generated text and the routine name may also contain personal, sensitive, or inaccurate information; comprehensive detection or removal is not promised. Additional private provenance and verification metadata contain necessary references, time windows, source verification values, delivery generation, and billing mode, but no second report copy, prompts, or raw audit values. These attributable identifiers and verification values are not anonymous; the send queue receives no report content.
Delivery creates a copy in the recipient’s mail service and possibly on devices, in backups or after forwarding. Their location, retention and deletion depend on mailbox use. Later deletion or permission changes in struct-i-vio do not remove text already delivered.
Migrated transactional platform delivery uses Amazon Web Services EMEA SARL, with Amazon SES and its delivery/feedback resources in Stockholm (eu-north-1). The Supabase database/Edge processing configured for this purpose and the relevant Vercel email callers are in Frankfurt. Applicable provider agreements and any necessary transfer grounds, including applicable EU Standard Contractual Clauses, apply. For migrated SES paths, we require TLS to the recipient mail server; an unencrypted downgrade is not envisaged. This is not end-to-end encryption or a promise of exclusively European processing or storage at recipients, global ingress or necessary provider transfers.
Notification delivery jobs expire 24 hours after the event. The additional sending queue contains technical references rather than email text. Such identifiers remain linkable personal metadata. Completed attempt/delivery metadata is selected for cleanup 30 days after last update; unresolved operations after 90 days. Actual removal depends on successful monitored runs. Minimal final suppression records prevent resending and follow deletion of the original notification. Raw SES/SNS delivery events may contain addresses; SNS may redeliver them for up to around 23 days. Normalized feedback data then has separate one-day queue and seven-day error-queue periods; these are not a maximum total period for the raw path. Provider suppression lists, operations logs, backups and recipient copies follow separate rules. For your own activity report copy, the 24 hours run from the original successful run; retries do not extend the period. The chat source follows its own deletion path. Additional private provenance and verification metadata are selected for limited monitored cleanup from 30 days for terminal cases and 90 days for unresolved cases, measured from the relevant later source or outcome time; incidents can delay removal. The original job remains permanently blocked before evidence is removed. Permitted deletion of a message, conversation, routine, run, account, or tenant is not blocked and must not leave an orphaned sendable job. Recovery must not resend, regenerate, or charge expired, deleted, opted-out, accepted, or unresolved jobs. External transfers already started cannot safely be withdrawn. Copies already delivered to mailboxes or devices, backups, or forwarding destinations fall outside this app deletion path.
9. Data Security
We implement the following technical and organisational measures (TOMs):
- Encrypted transmission between your device and our application (struct-i-vio.com) via HTTPS using TLS 1.2 or higher; the outdated protocol versions TLS 1.0 and 1.1 are refused; for connections to services of the providers we use, the supported TLS versions depend on the respective provider
- Encryption of data at rest (AES-256)
- Multi-tenant isolation through Row Level Security (RLS)
- Role-based access controls
- Files are processed in separate, access-protected storage areas and are provided only following authentication and authorisation verification
- Additional file backups are retained separately from production storage at a separate EU storage location (AWS S3, Stockholm, eu-north-1)
- Additional encrypted database backups are retained separately from the production system at a separate EU storage location (AWS S3, Stockholm, eu-north-1)
- The AWS deck renderer processes PPTX inputs and generated PDF outputs only for each requested job in a non-executable temporary file system; files are cleaned up on success, failure or cancellation, and contents are neither logged nor stored permanently; production operation requires a verified EU region and private ingress
- Weekly automated security checks (including known vulnerabilities in software dependencies, inadvertently stored credentials, and TLS and security-header configuration), a quarterly internal security review and security reviews of security-relevant changes within the development process; findings, such as required security updates, are assessed and addressed according to risk
- Daily automated database backups at the database provider (Supabase, retained for 7 days), supplemented by the additional encrypted database backups mentioned above (retained for no more than 12 months); the state of an available backup can be restored; recovery to any arbitrary point in time (point-in-time recovery) is not currently used
Storage of Uploaded Documents
Uploaded documents and other files are generally stored together with metadata relating to the respective company, project, or transaction context. This enables the application to link access to files to login, tenant separation, and the respective roles and approvals granted.
For resilience purposes, uploaded files may additionally be retained in separate backup storage within the EU (AWS S3, Region Stockholm, eu-north-1). These backup copies are not part of regular user access, remain within Europe, and are used only for recovery or technical emergencies.
Database Backups
To ensure the recoverability of the application, encrypted backup copies of the production application database are additionally retained in separate backup storage within the EU (AWS S3, Region Stockholm, eu-north-1). These backups are not part of regular user access and are used exclusively for emergency and recovery purposes.
10. Data Export and Erasure
Data Export
You can request a complete export of your data at any time. The data will be provided in common formats (CSV, JSON).
Erasure During Ongoing Operations
Where regular working documents are deleted within the application, they are generally not permanently removed immediately. Files are first moved to a recycle bin and may be restored there by authorised users. Final removal from production operations generally takes place only after up to 30 days, provided that no restoration occurs and no statutory retention obligations prevent this.
Backup copies of uploaded files are retained in separate backup storage within the EU (AWS S3, Region Stockholm, eu-north-1) for as long as the respective file is still present on the platform. If a file is permanently removed from the system recycle bin, the corresponding backup copy is automatically deleted from backup storage no later than after 1 year (365 days) (backup versioning purge). These copies are not visible in the regular product and are used exclusively for recovery purposes.
In addition, encrypted backup copies of the application database may be retained for purposes of resilience and recovery for up to 12 months in separate backup storage within the EU (AWS S3, Region Stockholm, eu-north-1). These copies are not visible in the regular product and are used exclusively for emergency or recovery purposes.
For native PowerPoint presentations, at most the last 50 automatically created immutable recovery versions per document remain active. A regularly scheduled, bounded cleanup run moves versions to the trash once they are 30 days old; the actual timing depends on successful runs and any backlog. The regular trash retention period of up to 30 days and the file and database backup periods described above then apply. Renderer previews are not permanently stored or mirrored to the S3 file backup storage.
Incomplete file uploads: If a file is transferred via an upload link but the associated transaction is not saved (for example because the window was closed, the form was cancelled or saving failed), the file is not assigned to any transaction and does not appear in the application. If no record refers to such a file, it is picked up for final deletion, without the recycle bin, in a regular cleanup run approximately 26 hours after the upload link was issued; the actual timing depends on the run being successful. Restoration is not possible afterwards. As long as such a file has been neither assigned to a transaction nor released for processing, it is not transferred to the file backup storage. This also applies to incomplete uploads in the AI chat. Saved documents are not affected. This provision does not apply to files uploaded before this cleanup was introduced. For the quota check and the cleanup, a technical entry is kept for each upload link, containing the tenant and user identifier, storage area and storage path (which may contain a sanitized file name), purpose, declared size and timestamps. It is picked up for deletion 48 hours after the upload is completed and may thereafter remain in encrypted database backups for up to 12 months. The ordinary technical upload reservation entry is separate from AI/MCP transfer intents. Their actor and file metadata is selected for deletion in the next successful maintenance run seven days after a successful save or completed cleanup. Pending intents retain the minimum data needed for quota and cleanup throughout the transfer capability window; failed cleanup is retried. Saved documents follow their own trash and backup lifecycle.
Receipts for corrected transaction entries
When an explicitly instructed correction removes transaction entries created in error, the original receipt may be retained independently of the removed payment entry. The original file is neither altered nor marked as cancelled or paid. It remains accessible to people with the tenant’s current finance and receipt permissions and continues to follow the stated trash, deletion and backup rules. To explain the correction, the original record and external reference, date, amount and currency, description and creation time, as well as correction time and reason, are initially stored. These details may contain personal data. Acting persons are not additionally copied into this correction context; existing class-specific audit log retention periods apply to their records.
When the receipt is permanently deleted from the production environment, the copied original transaction content and free-text correction reason are removed completely. An earlier necessary removal of personal contextual details may be carried out by the operator on a verified, documented instruction without altering the original file or blocking authorized access. Removed details must not be made accessible again through restoration or repetition. Your company remains responsible for any required tax or commercial records and their archiving.
Separate access-restricted correction and suppression records may remain after receipt deletion to prevent removed erroneous entries or cleared contextual details from reappearing through old imports or backup restores and to complete file backup operations correctly. They contain only necessary tenant, record and operation identifiers, integrity checksums and time and status information; no copied transaction content, free-text reasons, names or identifiers of acting persons, or API keys. Such identifiers and checksums may still be attributable and are not anonymous. Operation identifiers and integrity checksums are not shown in the receipt overview or its API and AI outputs. Their necessity is assessed during regular backup reviews. Controlled removal requires permanent deletion of the receipt, completion of dependent file backup work and documented evidence that no older backup still eligible for restoration can reintroduce the removed data. There is no fixed automatic expiry for these records. Upon permanent tenant deletion, tenant-bound records are removed; existing backups remain subject to the periods already described. This is not additional statutory financial archiving.
Specific features of the internal messenger
Messages in the internal messenger are not deleted after a set period (see section 6, “Internal messenger”). When a message is deleted, its text and any attached files are removed from ongoing operations for all members immediately; deleted messages and their files cannot be restored in the application. The files are technically kept in the system recycle bin for up to 30 days and are then finally deleted; if a person’s content is deleted at their request under Art. 17 GDPR, this period does not apply and the files are finally deleted in the next regular clean-up run. Only sent attachments are included in the file backups; files that were never sent are not. Deleted messages may remain in the encrypted database backups, and deleted attachments in the file backups, until the periods stated expire. If a single company is restored from a backup, messenger data is not restored, so that deleted messages, erasures under Art. 17 GDPR and ended chat memberships are not revived. Stored “last online” day information is deleted as soon as your company switches off the activity indicators or the person concerned hides this indicator for themselves.
Specific Features of the Optional Banking Integration
When a finAPI bank connection is disconnected, we remove the associated connection and account metadata from live operations. When the last finAPI bank connection of a tenant is removed, we also arrange for deletion of the associated technical finAPI user at finAPI, unless a technical or legal exception applies. finAPI’s statutory retention obligations for its own contract and payment service records remain unaffected.
Transaction, posting, or allocation data already transferred to struct-i-vio may continue to be retained within the customer's sphere of responsibility if they have already been used in project, billing, export, or tax-relevant processes or are subject to statutory retention obligations.
Technical logs and backup copies may furthermore remain in existence until the respective backup and recovery retention periods expire and are subsequently deleted or overwritten.
Erasure upon Termination of the Contract
After the contractual relationship ends, personal data in user accounts is deleted or anonymized after review, without undue delay unless statutory retention duties prevent it; access is terminated. Technically necessary backups follow their separately described deletion cycles; those cycles do not delay deletion of data no longer needed in the active production environment. Tenant- and project-related data may be retained for a reasonable period where required for data exports or compliance with audit-log retention periods (see section 10a). Users may request account deletion; if the in-app function is unavailable, the request must be sent by email. A request is not submitted automatically when an in-app deletion attempt fails. After the ordinary deletion of a financial planning draft or saved plan version, the authorized removal of personal ancillary details from it, an access restriction on such planning data, or the end of the contractual relationship, separate access-restricted technical deletion and suppression records, as well as records of ancillary-detail removal, may remain. They contain only the necessary tenant, plan-version, and operation identifiers and minimal action, revision, time, status, and control details; they contain no names, contact details, financial values, notes, reasons, request or result content, or checksums of deleted content. Their sole purpose is to prevent deleted, access-restricted, or ancillary-detail-redacted content from becoming accessible again through earlier retry operations or backup restoration. Necessity is reviewed regularly; the records are removed once the affected content is no longer present in any backup still admitted for restoration and the related earlier operations can no longer be accepted. This does not establish a statutory retention period for planning data. Business contact and company prices are stored for calculations, explicitly selected price comparisons, and deliberate use in supplier bills of quantities. The catalog is not intended for salaries or personal employee compensation. Archiving remains reversible and is not data deletion. Authorized deletion of the related contact or company removes its catalog entries. Contract values and dated performance records already expressly adopted remain independent; required personal accompanying details and textual source references may be minimized through a protected cleanup process without changing financial amounts. Provenance records do not contain an additional contact profile with email, telephone number or address. Earlier and, where required, subsequent entry values are kept access-protected for undoing catalog changes and may be used for restoration for no more than seven days. Successful undo, a final determination that restoration is blocked, or authorized deletion or minimization removes the affected restoration values earlier. Expired values and associated structured AI copies are removed by a regularly scheduled, bounded cleanup run; the actual time depends on successful runs and any backlog. This is not complete deletion of independently authored conversation content and does not establish a new statutory retention period for price quotations.
struct-i-vio is a data capture system (see General Terms and Conditions § 2 para. 3). Responsibility for compliance with commercial and tax-law retention obligations (in particular HGB/AO) lies with the customer. You can export your data completely at any time. Upon request, you will receive confirmation of erasure.
10a. Change Log (Audit Log)
The platform maintains an automatic change log of security- and compliance-relevant actions (e.g. logins, file uploads, data changes, deletion operations). The retention period depends on the data class:
| Data class | Retention period | Justification |
|---|---|---|
| Operational business data | 90 days | Operational transparency |
| Master and contract data | 365 days | Limitation period, BGB § 195 |
| Email content | 30 days | GDPR data minimisation requirement |
| AI provenance | 30 days | GDPR data minimisation requirement |
| Financial data | 365 days (1 year) | Operational traceability |
| Banking/PSD2 | 365 days (1 year) | Operational traceability |
Users and customer administrators can neither edit the content of entries in the change log nor delete them (protection against tampering); customer administrators holding the permission to erase personal log data can, to fulfil requests under Art. 17 GDPR, redact only the content fields of affected entries, while the time, acting person, action and affected record are retained and the redaction itself is logged. On the system side, entries are changed or removed only after the retention period has expired or, where required under data protection law, for deletion, redaction or anonymisation. Data subjects may request, pursuant to Art. 17 GDPR, the redaction of personal content or the anonymisation of personal references within entries (rather than complete deletion of the entry, as this would destroy its evidential character).
For the internal messenger, only administrative operations on group, project and company chats are logged — creating and deleting a chat, adding and removing members, role changes, changes of name and description (only as a field name, without the wording) and of the rules on who may write and add members, and switching the company chat “General” to read-only and reopening it —, in each case without message content (class “Operational business data”). Switching the activity indicators on and off is logged like other changes to the app settings. Messages, their editing or deletion, reactions, read positions, the creation of direct chats and activity indicators are not recorded in the change log, so that the writing and reading behaviour of individual persons cannot be traced from it.
Legal basis: Art. 6(1)(f) GDPR (legitimate interest in operational security, protection against tampering and evidence of compliance).
11. Optional email integration (IMAP/SMTP)
struct-i-vio offers an optional integration for mailboxes accessible via the open IMAP/SMTP protocol (e.g. Hetzner Mail, Hostinger, IONOS, Strato or proprietary mail servers). Use of this integration is voluntary and may be disconnected at any time.
Categories of data processed
- Mailbox access credentials: Email address and mailbox password, as well as IMAP and SMTP server parameters (host, port, encryption type). The password is stored exclusively in encrypted form (AES-256-GCM, random IV for each encryption). Decryption takes place only server-side during the sync process; the plaintext password does not leave this process and is never logged.
- Email content and headers: Sender and recipient addresses, subject, Message-ID, email text, and attachments of synchronised messages. The information about reusable personal signature images applies regardless of connection type, both to Microsoft 365 (Microsoft Graph) and to IMAP/SMTP. Users can upload images or paste them from the clipboard for their own email signature. Stored data consists of re-encoded PNG/JPEG files and the technical information needed for ownership, tenant, usage, and deletion. Reusable personal image files can be retrieved only by their owner in the current tenant; the template management permission to read signature text does not extend to these image files. When sent, the images used are transmitted as embedded attachments through the connected mailbox to the selected recipients. The existing mailbox and lead access rights apply to the sent message.
- Imported lead attachments: Attachments from genuine incoming or outgoing emails are automatically stored in the associated lead's special ‘Email attachments’ folder once the conversation is assigned to that lead — provided that the corresponding CRM setting, which can be disabled per tenant, is enabled (default: enabled). This also applies to personal mailboxes. The message, mailbox, attachment and import time are retained as evidence of origin. Image contents identified as signatures that recur with identical content in at least three independent conversations with at least three different subject lines, small embedded images under 20 KB and GIF images are not imported automatically; other embedded content — such as photos or screenshots in which people may be identifiable — is treated as regular attachments. Copies with identical content are stored only once per lead; additional filenames for the same content are recorded as additional labels. Users with lead editing permission can manually save individual files that were not imported automatically at any time. The files are visible to people with lead reading permission. For protected source fields, current field read rights are also checked; files without a reliable content mapping remain blocked where source-field rights are missing.
- Sync metadata: Connection status, timestamp of the last synchronisation, technical error messages (without plaintext passwords).
Notice regarding full access and password storage
Unlike OAuth-based integrations (e.g. Microsoft 365), IMAP/SMTP requires storage of the full mailbox password. This grants the system full access to the mailbox — restriction to individual folders or operations is technically not possible. Where supported by the email provider, it is recommended to use a dedicated app password instead of the main password in order to limit the impact in the event of a compromise.
Legal basis
The processing of mailbox access credentials and the user's own emails is based on Art. 6(1)(b) GDPR (performance of the contract for the provision of the integration function).
Where synchronized emails contain personal data of third parties (senders, recipients and content from correspondents who are not themselves struct-i-vio users), processing of these third-party data is based on Art. 6(1)(f) GDPR (legitimate interests). The legitimate interest is to provide a functional business communication and project management environment that the user has activated as controller in relation to their own correspondents. The affected third parties reasonably expect their business correspondence to be managed by the recipient in professional tools. This interest concerns regular mailbox processing in a business context. If the optional semantic CRM email search is enabled and approved for the individual mailbox, the separate information in Section 12 applies: minimized content is transmitted to Azure OpenAI and derived search vectors are stored in the tenant.
Internal visibility within the tenant
Viewing a connected mailbox (inbox, folders, search) is restricted to the mailbox owner and people explicitly granted reading permission for that mailbox. An exception applies to individual email conversations linked to a lead in the CRM module, either manually by users or automatically (e.g. when a reply is detected or through a mailbox configured as a lead source): these conversations, including their attachments, are additionally visible to everyone in the tenant who may view that lead (general reading permission or permission restricted to assigned leads), even without their own mailbox reading permission. The remaining mailbox contents are unaffected. Removing the link also removes this additional visibility. Attachments from that conversation already imported automatically into the lead, and project copies created from them, are moved to the trash for 30 days rather than deleted immediately; if the same conversation is linked to the same lead again within that period, they are restored automatically, otherwise they are permanently deleted after the period expires. Until then, they remain viewable and retrievable by people with general trash permission, regardless of reading permission for the lead or mailbox. If the mailbox owner disconnects the integration or reauthorization is required, previously linked history remains viewable by authorized lead viewers within the retention period stated below. For a confidential pipeline phase, access to that phase is additionally required; level 0 users and platform administrators acting for support do not need this access.
Attachments transferred to the lead remain visible and retrievable to people with general trash permission only when they may view the lead; resulting project copies remain retrievable by people with that trash permission.
For project-related procurement inquiries, authorized users can prepare personal inquiry batches to known business partners from selected bill of quantities (LV) items. The selected service details, sender, recipients, message texts and signature details, as well as the respective sending status, are stored. In normal access to this function, the private inquiry batch and its sending statuses are accessible only to the person acting. Only after reviewing all recipients, content and the current signature does a separate human send step start transmission through the authorized connected mailbox. No additional shared project mail archive is created. The ordinary mailbox storage and its existing access permissions remain separate; acceptance by the mail provider does not prove delivery to the recipient.
Sending emails through a connected personal mailbox always independently requires sending authority (mailbox owner or explicitly granted reply/sending permission); reading permission alone is insufficient. The same applies to shared mailboxes once the tenant has enabled advanced mailbox access controls; if those controls are not enabled, any active tenant member with the permission required for the respective sending action may instead send through such a mailbox. Read access for support purposes by people entrusted with platform administration does not itself confer authority to send on an employee's behalf; a platform-wide support role never confers such sending authority.
A confirmed change of the sign-in email automatically triggers reauthorization for mailbox connections assigned to that authentication identity. The local connection is disabled, and mailbox secrets stored in struct-i-vio and local OAuth and synchronization data are deleted immediately. Previously synchronized messages, documents, attachments, assignments and permissions are retained in accordance with the deletion and retention rules below. This local disconnection does not necessarily revoke existing OAuth consent or a provider-side subscription at the respective email provider immediately; expiry or revocation depends on that provider's functions and periods.
Storage period
- Mailbox access credentials: Until the user actively disconnects the integration. Upon disconnection, access credentials (encrypted ciphertext in
mailbox_secrets) are deleted immediately. - Synchronised email content: Until the respective email is deleted in the application or until the tenant account is deleted, whichever occurs first. The information about reusable personal signature images applies regardless of connection type, both to Microsoft 365 (Microsoft Graph) and to IMAP/SMTP. Reusable personal signature images are stored separately from message attachments. Disconnecting a mailbox does not delete these images. Images removed from a successfully saved signature and no longer in use are moved to the personal trash. The 30-day period is for restoration; permanent deletion is scheduled after it ends and may be delayed while backup and deletion confirmations are pending. Such delays do not extend the restoration period. Uncommitted uploads are moved to the personal trash by a successful cleanup run once they are at least 24 hours old. Restoring an image does not automatically add it back to the signature; if it remains unused, its 24-hour grace period starts again at restoration. The deletion procedures described above and separate backup retention periods apply to user and tenant account deletion and backup copies. Removing a signature image does not recall emails already sent or copies held by recipients.
- Email attachments stored in the lead: Imported attachments remain with the lead for as long as needed for the business transaction or until the controller instructs their deletion. Disabling automatic import does not delete previously stored files. Previously stored signature, icon or GIF images and multiple copies with identical content are identified once based on their file content and moved to the trash. Signature and icon images can be restored there within the applicable period; a copy moved as a duplicate can, however, only be restored while the file with the same content retained in the lead is no longer active. On conversion to a project, project copies remain linked to their source and are subject to its deletion rules.
The contents of the private inquiry batch expire 30 days after the last deliberate content change, regardless of sending status. Reading, approval, sending and retries do not extend this period. After expiry, access to the content and further sending are blocked; the bounded daily cleanup removes expired content. Backlogs or failed runs may delay actual removal. The delete function removes the private live content. This period and batch deletion do not automatically apply to additional copies created in AI chat, confirmation proposals or exports; those follow their respective existing access and deletion rules. Verified access and deletion requests through the described contact channel include the content and sending evidence that can still be attributed to a person. After content deletion and until tenant deletion, only access-restricted random operation identifiers, timestamps and states remain to prevent resending, without addresses or message texts; these remnants are inventoried only at tenant level. In a full database restore, all restored private inquiry content is discarded as a precaution; the function remains blocked until a separately reviewed reactivation. Tenant restores do not include this data. Existing database backups are subject to the stated retention periods of up to twelve months. Deleting the private batch does not recall copies already sent to the mailbox or recipients.
Data security of access credentials
Mailbox access credentials are stored in a separate, particularly secured database table (mailbox_secrets), to which the following safeguards apply:
- All database permissions for authenticated and anonymous roles are fully revoked (
REVOKE ALL). Read access exists exclusively for the server-side sync process via a privileged database function. - Tenant users cannot directly query the encrypted secret data at any time, including through their own SQL queries.
- Decryption takes place exclusively in the working memory of the sync worker and is not logged.
Customer's mail server is not a sub-processor
The mail server to which struct-i-vio connects via IMAP/SMTP is operated by the user itself or its provider (e.g. Hetzner, Hostinger). This mail server is not a sub-processor of struct-i-vio within the meaning of Art. 28 GDPR, as struct-i-vio merely accesses the mail server as an external data source. As controller, the user is responsible for ensuring that it is authorised to access the emails contained in the connected mailbox and to synchronise them in struct-i-vio.
12. AI functions (optional)
struct-i-vio offers optional AI-supported functions for supporting and automating workflows. These functions are voluntary and may be used or not used by the user at any time.
This Section 12 describes the AI functions integrated into struct-i-vio. The optional connection of external AI clients (e.g. Claude, ChatGPT, Cursor) via the MCP interface is described separately in Section 12e; the statements in this Section 12 regarding consent, data location, processing on behalf by Microsoft, exclusion from training, and confirmation of write actions do not apply to these external clients.
For explicitly requested new CRM email drafts, you can save personal instructions about tone and content for each user account and company. This setting is accessible only to you through normal application access; you can read, export, replace, and delete it. Saving and managing it do not start an AI call. Only on your explicit request are your instructions, your details about the purpose, and the currently accessible Lead title and required recipient name and company processed through the existing Azure OpenAI service. Your current personal AI consent, permissions, and available AI balance remain prerequisites; processing is charged under the described credit rules. Suggestions are first shown for preview and are not saved in a separate draft store. Applying and sending are separate user actions. Saved instructions are removed from the active data set when replaced or deleted, and when you permanently leave the company or the account or tenant is deleted. Explicit deletion by you removes the saved text from the active data set immediately; replacing it creates no content history. Access ends immediately when your membership permanently ends or expires. During normal operation, the daily cleanup is scheduled to remove any remaining instructions within 24 hours, including when saving and departure happen at the same time. Delayed or failed cleanup runs may postpone actual removal. Deleting the account or company also removes the associated instructions. No content history is kept; change records contain only necessary identifiers, timestamps, revision, and status. The described database backup periods of up to twelve months apply to backups; deletion does not immediately remove copies already present in backups.
Consent (Art. 6(1)(a) GDPR)
Use of the AI functions requires your express consent. Before you use the AI assistant for the first time, you will be informed about the data processing and asked for your consent. Without this consent, no data will be transmitted to the AI service provider (Microsoft Azure) via these functions.
Your consent is documented in an audit-proof manner in our system (time, version of the privacy notices, user ID).
Withdrawal of consent
You may withdraw your consent at any time with effect for the future. Withdrawal may be made informally by email to tigran.varazhian@struct-i-vio.com. This does not affect the lawfulness of data processing carried out prior to withdrawal (Art. 7(3) GDPR).
Scope of data processed
When using the AI function, the following data are transmitted to the AI service provider:
- Identity data: Your name, your email address, and your position in the company
- Chat messages: All messages that you send to the AI assistant
- Project and task data: Names, descriptions, statuses, deadlines (to the extent visible)
- Employee data: Names, roles, working hours, absences (to the extent visible)
- Financial data: Budgets, invoices, expenses (to the extent visible)
- CRM email search data (optional): where semantic CRM search is enabled, the minimized subject line and sanitized message text, as well as lead title, description and tags
- Lead email excerpts loaded on request (optional): For an expressly requested AI task, limited excerpts of already stored messages from a specific lead conversation may be processed. This requires that you are permitted to view the lead and have current read authorization for the associated mailbox; a lead link alone does not replace mailbox authorization. Only the messages requested for the task within a technical scope of no more than 50 messages and 5,000 input tokens are transmitted; attachments, raw headers and messages not requested are not included. No automatic background retrieval or additional permanent storage takes place.
- Document and attachment data (optional): content, file name and technical metadata of a PDF file provided in the AI chat or a PPTX presentation selected for native editing, in particular the slide and object data required for this purpose
- Company Hub references (optional): content and file metadata from entries and attached DOCX, PDF or PPTX/POTX files in Company Hub areas that the user is permitted to read and that the user uses as a reference for the respective AI task
The scope of the data transmitted corresponds to the respective user's visible area and is limited by the authorization system (Row Level Security).
Content and metadata of the internal messenger (section 6) — messages, attachments, reactions, chat memberships, read positions, activity indicators and messenger notifications — are excluded from the AI functions. The AI assistant can neither read, search or summarise them nor send messages or manage chats; on the messenger pages it does not create a screenshot of the view. The same applies to external AI clients (section 12e) and the programming interface (section 12c). If you yourself insert content from the messenger into a message to the AI assistant, it is processed like any other chat input.
Data for the Azure OpenAI features described below are transmitted via the Azure OpenAI Service to Microsoft Deutschland GmbH (Walter-Gropius-Straße 5, 80807 Munich, Germany) as a processor. The separately described AI intake screening is governed by the information on Microsoft-Decision-1 in Microsoft Foundry; it does not use an Azure OpenAI model call.
For the optional analysis of a PDF uploaded in the AI chat, the local PDF text layer and the expressly requested pages are used first. Only if a visual page analysis is necessary may the page images required for this purpose be transmitted to Azure OpenAI. No additional OCR or document provider is used for this chat PDF workflow. The AI creates no more than a proposal from this; an invoice and a permanent copy of the document are created only after the user's express confirmation.
The same applies accordingly to the optional AI pre-population of forms, for example when creating an invoice or expense. The file selected for this purpose is held only temporarily in a protected processing step that has not yet been saved and is processed to determine field suggestions. The locally extracted text layer and — where the specific processing path requires this — only the page images required for visual review are preferably transmitted. Processing takes place via Azure OpenAI using a Luna or Terra deployment approved for this purpose. The analysis provides exclusively a non-binding proposal. No automatic permanent storage in invoices, expenses or other business data takes place: after the analysis, the recognized values can be transferred into the form that has not yet been saved so that the user can review and amend them. Values entered or touched by the user are not overwritten; untouched system defaults may be replaced as a related financial group if the analysis provides consistent values for this purpose. The final saving process remains an express user action. Separate technical deletion periods apply to unconfirmed proposals and temporarily held files: the staging ticket is valid for 10 minutes; the associated signed upload URL may be valid for no more than 2 hours. Following analysis, cancellation or reset, the staging file is deleted on a best-effort basis. Staging files not successfully removed are picked up again for deletion in a final sweep after 26 hours; actual removal depends on successful execution. Associated technical metadata is picked up for deletion 24 hours after this final sweep; the additional technical reservation entry of the upload is picked up for deletion 48 hours after its completion. The structured analysis result may additionally be retrievable as a short-lived technical receipt cache, including a SHA-256 hash of the document content read by the server, tenant, project and user identifiers, and the structured result, for no more than 15 minutes from completion. After this period expires, it is no longer made available for display in the product; physical deletion takes place in the next successful regular cleanup run, which normally occurs every five minutes but may be delayed in the event of a failed or delayed run. For unresolved provider-started or billing-pending transactions, technical receipt metadata may be retained until the start of the regularly running cleanup process after seven days; the actual deletion time depends on successful execution. Content-free Wallet billing evidence may remain for reconciliation in accordance with the applicable retention periods. Encrypted database backups may still contain these temporary data for up to 365 days (no more than 12 months) thereafter; they are not accessible through the production system and are deleted in accordance with the applicable backup purge rule. A permanent copy is created only through the user's normal saving process.
Irrespective of this, the raw chat attachment is permanently deleted in the next regular cleanup run after the expiry of the seven-day retention period and is not mirrored to the file backup storage. If it is adopted as an attachment to an invoice upon the user's explicit confirmation, a separate copy is created. The regular recycle bin, recovery, and backup rules for business documents apply to this invoice copy.
Pre-population may be enabled by a tenant administrator as an optional tenant-wide function. If it is enabled, a supported invoice or expense may be automatically processed for analysis after upload. The prerequisite for each individual user remains the prior personal consent to AI data processing that has been given and documented. Tenant-wide enablement does not replace this personal consent. If the function is disabled in the tenant or the user withdraws their consent, no new files from this pre-population will be sent to Azure OpenAI in the future; requests already started may still be technically completed; a call already transmitted to Azure OpenAI cannot be guaranteed to be subsequently cancelled. However, its result is discarded after server-side verification. The AI may transfer proposals into the form that has not yet been saved. Values entered or touched by the user are not overwritten; untouched system defaults may be replaced as a related financial group. No automatic permanent saving of the invoice or expense takes place.
Company Hub as a reference and template source
At the user's express instruction, the AI may use readable entries and attached files from the Company Hub as a reference for a new document or presentation. The selection remains limited to areas and content for which the user has read authorization in the current tenant. No automatic full indexing of the hub, background retrieval of all files or automatic disclosure to additional recipients takes place. The read content is treated as untrusted reference data; embedded instructions from a source file change neither the user's request nor authorizations or security rules.
DOCX and PDF files are read for text and content references. A native copy retaining slide and object data is possible exclusively for PPTX or POTX files expressly suitable as templates; the original in the Company Hub remains unchanged. Text extraction does not confirm an identical visual representation.
Native PowerPoint presentations
When optionally editing a native PowerPoint file (PPTX), the content of the presentation selected by the user and the slide, text, image and object data required for the respective request are processed. If the user requests an AI change, the excerpt required for this purpose, the user's input and the generated result may be transmitted to Azure OpenAI. By default, the AI change is initially displayed as a proposal and is incorporated into a saved presentation version only after express confirmation. Automatic incorporation takes place only if an authorized user has previously expressly selected the action policy “always execute automatically” for this AI action; server-side authorization, tenant and conflict checks continue to apply.
For a slide preview or an export requested by the user, the AWS deck renderer operated by structivio processes the respective PPTX file and, where applicable, the PDF file generated from it. Processing takes place per request on a non-executable temporary file system. Input, output and the temporary LibreOffice profile are cleaned up after success, error or cancellation. The renderer does not retain permanent presentation copies or preview caches, does not mirror these data in S3 file backup storage, and does not log presentation content. Before productive activation, the specific AWS operating location must demonstrably be restricted to an EU region and private access to the service must be verified.
For recovery purposes, immutable presentation states with a timestamp, assigned user, and a brief description of the change may be automatically stored in the product. No more than the last 50 automatic states per document remain active. A regularly scheduled, limited cleanup run moves states to the recycle bin after they have reached an age of 30 days; the actual timing depends on successful runs and any backlog. The regular recovery period of up to 30 days applies there. File backups may continue to exist for up to 365 days after final deletion, and encrypted database backups for up to 12 months; they are not part of the regular user view.
No training on customer data
Microsoft does not use data transmitted via the Azure OpenAI Service to train AI models. This is contractually stipulated in the Microsoft Products and Services Data Protection Addendum (DPA).
Retention period for Azure OpenAI
Microsoft describes abuse monitoring for Azure OpenAI in which requests are automatically evaluated and content identified as suspicious may be retained for potential review. The current Microsoft documentation does not specify a fixed retention period applicable to all resources and deployment types. The applicable terms are the Microsoft contractual terms and the settings of the specific Azure resource used; no zero-retention commitment is made here.
Storage in struct-i-vio
Your AI chat histories are stored in the struct-i-vio database to enable you to access previous conversations. You can delete individual conversations at any time. When your user account is deleted, all AI chat histories are also deleted.
To undo AI-assisted changes to draft invoices, we store only the previous values of the fields actually changed in the access-protected action log. Restoration is possible for seven days. These values are removed when restoration succeeds or the invoice is deleted; otherwise, they are removed after the retention period by a regularly scheduled, limited cleanup run. The actual deletion time depends on successful runs and any backlog.
To undo AI-assisted changes to the supplier bill of quantities, we store the earlier supplier items and contract values required for restoration with access protection. Undo is available for at most seven days and may end earlier if intervening changes occur or personal data must be minimized. The stored restoration values are removed after a successful undo, a final determination that restoration is blocked, deletion of the associated award, or such minimization. After the seven-day period, any remaining restoration values are removed in the next successful, regularly scheduled and bounded cleanup run; disruptions or backlog may delay actual removal. An action receipt limited to required identifiers, time and status details, and labels of affected fields, without those earlier contents, remains subject to the other applicable deletion and retention rules. Recorded performance progress is not undone by this; corrections are recorded as additional dated entries.
Usage-based credit billing
For each customer-related AI function, we additionally collect the usage and posting metadata required for billing. This includes, in particular, the assignment to the user and tenant, the AI function or operation, the measured processing, the resulting credit charge, the version of the billing basis used and, where applicable, a traceable reservation, refund or adjustment entry. This information serves the performance of the contract, the preparation and review of the agreed billing, as well as protection against erroneous and duplicate entries (Art. 6(1)(b) and (f) GDPR). It is considered separately from the respective AI content. The credit record visible to the customer or tenant generally contains neither input and output data nor internal calculation bases. Separate internal billing records with restricted access may contain the technical parameters required for determination, but not the AI content. The specific credit charge is displayed and documented in the usage account as soon as the respective billing has been finalized. Until then, a neutral pending billing status may be displayed. It may vary depending on the model, context and tools used. Subsequent processing steps assigned to the same chat request may be added as additional usage entries once they have each been finalized; they do not constitute a correction of an already finalized record.
Direct user, employee and conversation references are removed no later than after 1,095 days. The remaining billing evidence is deleted no later than after 3,650 days, unless a statutory retention obligation prevents this. Until the direct references are removed, the evidence is processed only in pseudonymized form; the remaining tenant assignment (tenant_id) may be retained as technical financial and allocation information. billing_pending processes that are still open, disputed or cannot be conclusively assigned are exempt from automatic deletion until they have been clarified and closed. Thereafter, the regular periods apply, with data minimization taking place at the earliest possible time in each case. Deleting the chat history does not automatically delete the record required for billing; any necessary correction of an already finalized record is made through a separate adjustment entry.
Data location and EU Data Boundary
Processing takes place on Microsoft infrastructure within the Microsoft EU Data Boundary (primarily: Sweden Central region). No transfer to third countries outside the EEA takes place for the regular AI assistant. The DPA between structivio and Microsoft is available at https://www.microsoft.com/licensing/docs/view/Microsoft-Products-and-Services-Data-Protection-Addendum-DPA.
No automated decisions
The AI functions integrated into struct-i-vio described in this section do not make automated decisions within the meaning of Art. 22 GDPR. AI outputs are to be understood as recommendations. Write actions are performed by default only after express confirmation. To the extent that an authorized user has previously expressly configured “always execute automatically” for a specific AI action, that action may be performed without renewed confirmation; authorization, tenant, and conflict checks remain effective.
When external AI clients are connected, actions are performed without additional confirmation by struct-i-vio (see Section 12e). In that case as well, struct-i-vio does not make its own decisions about individuals; responsibility for decisions that a user or the customer makes or has made through an external client lies with the customer.
Speech recognition / dictation (optional)
Anywhere in the application where a text field offers a microphone button for dictation — currently, in particular, in the AI chat, in the AI instruction fields for quotation creation, and in CRM notes — you can enter your input by microphone dictation instead of typing. The function is activated exclusively by deliberately clicking the microphone button; without this click, your microphone is not accessed at any time. The recognized text is always inserted only into the respective text field and is never automatically sent or confirmed; an action that would otherwise trigger submission of the field (e.g., a status or phase change) is not triggered by the dictation itself.
For speech recognition, we use Azure AI Speech from Microsoft Deutschland GmbH, operated on a resource set up specifically for struct-i-vio in the Sweden Central (EU) region. Speech recognition is generally processed within the region of the assigned resource; your spoken data does not leave the EU in this process.
Your browser records the audio and, as soon as you end the dictation, transmits it once and encrypted (TLS) to our server. There, the recording is retained exclusively for the duration of this individual request, forwarded unchanged to Azure AI Speech for recognition, and deleted again within the same request — regardless of whether recognition was successful. We ourselves store no audio data. For internal cost control, we log exclusively the measured recording duration and the recognition language used for each dictation process; neither the audio content nor the recognized text is stored. Logging or retention of audio data is also disabled on Microsoft's side for this service; your spoken data is not used to train AI models.
Processing takes place only while you actively use the microphone button, subject to a strict maximum limit of 5 minutes per dictation process. The recognition language is always taken from the currently configured interface language of the application, which is in each case one of the five languages supported by struct-i-vio (German, English, Spanish, Russian, Ukrainian). No automatic recognition of the spoken language takes place.
The same consent and the same authorization to use the AI assistant apply to the dictation function as to its other functions (see the “Consent” section above). Without your consent or without this authorization, the dictation function is not provided.
Interactive real-time voice dialogue (optional)
The optional real-time voice dialogue enables an ongoing conversation with the AI assistant. When activated, the audio you speak and the text portions of the voice output generated by the service are processed. Input transcription (ASR) is not activated in the currently documented path. Transmission takes place via the Azure OpenAI Realtime deployment gpt-realtime-2.1 configured for struct-i-vio in Azure Sweden Central as a Data Zone deployment within the EEA. The audio is transported through the Azure Container App Gateway used by structivio in the same Azure region. The gateway retains audio and transcript content only transiently for the ongoing connection and does not write it to its operational logs; only limited, content-free technical runtime data is stored. The Azure resource and the gateway are deployed in Sweden Central; however, in the case of a Data Zone deployment, prompts and responses for inference may be processed in any EU Member State within the EU Data Zone defined by Microsoft. Sweden Central therefore denotes the resource and gateway location, not a guarantee of every individual inference processing location. This location information applies to the currently documented primary path. It does not constitute a general commitment to a specific provider retention period; in particular, separate Microsoft rules may apply to abuse monitoring, security reviews, or optional diagnostic functions.
For the real-time gateway, only content-free runtime measurements (e.g., a coarse phase and latency classification) are logged in Azure Monitor/Log Analytics. The retention period configured for this is currently 30 days. Spoken audio data, transcript content, and skill results are not stored there as gateway logs; permanently stored session, status, and task data remain in the respective product paths described.
Separate, versioned consent is required before the first real-time voice dialogue. This is documented per user, tenant, and consent version and, once successfully granted, does not need to be confirmed again for every new conversation. You may refuse this consent or withdraw it at any time with effect for the future. Without current consent, neither the microphone is activated nor a real-time connection established. The technical microphone permission of the browser or operating system is separate from this: the browser may additionally require its own permission and a deliberate user action. The microphone is used only after this activation; raw audio is not permanently stored by struct-i-vio by default. Withdrawal terminates ongoing real-time sessions and prevents new sessions; processing already lawfully carried out remains unaffected.
Ending the audio connection terminates the real-time transport. A delegated research or background task that has already been authorized and started may continue to run within its original token, credit, and time limit; its result will later be assigned to the authorized chat. Withdrawal of real-time consent, logout, or deactivation by the tenant blocks new executions and new independent result deliveries. A root response that has already been published, as well as attachments, sidecar additions, or technical completion data previously authorized and bound solely to it, may still be completed within the original limits and assigned to the authorized chat. A provider request that has already been technically authorized at the time of withdrawal may still be completed; no new authorizations or new independent follow-up actions will be granted.
When switching from an existing text chat, the full history is not transferred automatically. Before the chargeable start, a limited context from the current conversation is prepared and the anticipated charge is displayed. If necessary, individual older conversation information or results may later be loaded specifically. The real-time session, the stored text portions of the voice output, context transfer, and downstream tasks remain assigned to the respective authorized user and tenant. If you trigger an app task during the conversation, the existing assistant receives a limited interpretation of your spoken intent, not a purportedly verbatim-verified speech transcript.
For brief, read-only app queries, the real-time assistant may directly read limited data after loading a skill enabled for your permissions. These read accesses are checked again server-side with respect to tenant, role, and project; the results are truncated in accordance with the limits of the enabled read tool. Write actions, internet research, and longer or more complex processing are delegated to the regular AI assistant or its designated research path. The existing confirmation, authorization, logging, and consent rules continue to apply there; the real-time assistant receives only the task and status required for processing.
The live display distinguishes between an ongoing estimate and the credit charge confirmed following provider feedback. The estimate is not a fixed price. Pure context preparation uses the existing limited context and does not generate an additional paid summary. Audio input and output as well as downstream research, document, or other AI tasks may trigger their own usage charges. A server-enforced credit limit applies to the entire realtime session; if there is insufficient credit or the limit is reached, no further chargeable processing steps will be initiated.
Realtime billing records do not contain raw audio recordings, prompts, full transcript texts, or tool arguments. The handover context temporarily stored for the audio connection and the connection ticket are removed when the session ends or expires. Tasks and results stored in the chat follow the lifecycle of the conversation. When the conversation is deleted, associated realtime content is removed and references to the conversation are minimized. Separate required billing records are retained; unclear billing items remain open until clarified.
Web search by the AI (internet research)
If a controller (tenant administrator) activates the optional “Internet research” function for lead pre-qualification, selected lead data—typically company name, email address and freely entered description fields—are transmitted to the Azure OpenAI Service of Microsoft Deutschland GmbH when this function is used. Azure OpenAI uses the integrated web search tool for this purpose (Bing-supported, operated by Microsoft). Search queries may involve Microsoft Bing infrastructure that is not consistently located within the Microsoft EU Data Boundary; Microsoft is responsible for this sub-processing as the operator of the Bing service and is bound by its DPA.
Leaving the struct-i-vio infrastructure: As part of this function, the lead data leave the struct-i-vio systems and the Supabase database. From the time of transmission, the data are subject to the data protection rules of the Microsoft DPA. This qualitatively distinguishes AI web search from the regular AI assistant, in which data remain exclusively within the Microsoft EU Data Boundary.
Legal basis and consent: Activation by the tenant administrator requires separate, documented consent pursuant to Art. 6(1)(a) GDPR. Without this consent, internet research is not performed.
Processing on behalf: Microsoft Deutschland GmbH acts as a sub-processor for this processing within the meaning of Art. 28 GDPR. The Microsoft Products and Services Data Protection Addendum also applies to this processing. Microsoft does not use Azure OpenAI API data to train AI models.
Data categories and responsibility: Users, as controllers, are responsible for ensuring that they do not enter special categories of personal data pursuant to Art. 9 GDPR or other confidential information worthy of protection in lead description fields that have been enabled for internet research.
Automated lead processing (pre-qualification)
If a tenant administrator activates the optional “Automatic lead assessment” function, every newly received pre-qualification lead is automatically transmitted to the Azure OpenAI Service of Microsoft Deutschland GmbH without manual intervention as soon as it is recorded in the system. Based on the transmitted lead data (in particular the name, email address, company and description fields of the contact person), Azure OpenAI prepares an initial assessment and generates field suggestions.
Suggestions only — no automatic data entry: The assessments and field suggestions generated by the AI are merely submitted to the responsible user for review. They are not automatically transferred to the lead master data; every change requires explicit human confirmation.
No application of Art. 22(1) GDPR: The process does not constitute automated decision-making within the meaning of Art. 22(1) GDPR. The AI generates only preliminary assessments and suggestions; legally or economically significant decisions are not made solely through automated processing, but are always confirmed or rejected by a human. The AI output therefore serves as a supporting indication, rather than a legally binding decision.
Continuous processing: Unlike manually triggered individual research, this processing runs continuously for every newly received lead as long as the setting is active. The administrator may deactivate the function at any time.
Legal basis and consent: Activation by the tenant administrator requires separate, documented consent pursuant to Art. 6(1)(a) GDPR. For the processing of personal data of lead contact persons, the tenant, as controller, relies on its legitimate interest in the efficient pre-qualification of business inquiries (Art. 6(1)(f) GDPR).
Data location and processing on behalf: The same principles apply as for the regular AI assistant: processing within the Microsoft EU Data Boundary (Sweden Central); processing on behalf by Microsoft Deutschland GmbH pursuant to Art. 28 GDPR on the basis of the Microsoft DPA; no model training on API data. The Microsoft rules of the specifically used resource apply to abuse monitoring and possible storage of suspicious content; no general fixed retention period is promised.
AI intake screening before lead creation (optional)
This feature is disabled by default. It screens only new incoming conversations from expressly authorized lead-source mailboxes that have not yet been assigned to a lead under the existing assignment and suppression rules. Existing or historical correspondence is not retrospectively processed for this purpose. Assessment and field completion for a lead that has already been created are separate features.
Activation requires acceptance of the current Data Processing Agreement, the activating person’s current personal AI consent and a separate documented instruction. That instruction records the purpose, automatic consequences, exact mailboxes, text version, time and activating person. The required administrative authority is checked for each mailbox; personal mailboxes may only be authorized by the person who connected them. New mailboxes or mailboxes subsequently marked as lead sources are not automatically authorized.
Only the limited subject line and sanitized current message text, together with the short configured business profile and the company’s criteria and exclusions, are sent to Microsoft. Email addresses, signatures and quoted message histories are removed to the extent that they are technically detected. Attachments, raw headers, recipient lists and any compilation of project, financial or team history are excluded. The message text is the subject of screening and cannot instruct additional actions.
A sufficiently clear relevant result may automatically create a lead in the intended initial workflow. Clearly irrelevant messages are archived reversibly within the CRM only. Unclear results, processing errors or insufficient credit remain subject to manual review. The mailbox at the email provider is not changed; no automatic rejection message is sent. Authorized persons can find archived messages, restore them and manually create a lead. The applicable mailbox and lead access rights remain decisive.
Screening uses Microsoft-Decision-1 in Microsoft Foundry on the EU DataZone deployment authorized for this feature. Inference may be distributed across several EU Member States; exclusive processing in Sweden is not promised. A Global deployment or a switch to another provider is not envisaged for this feature. Data processing, abuse monitoring and possible storage are governed by the contractual terms and settings of the specific Microsoft resource used. No general fixed Microsoft retention period or zero-retention commitment is made. Without our separate permission or instruction, the transmitted data are not used to train AI foundation models or to improve Microsoft or third-party products.
The feature and individual mailbox authorizations can be disabled or revoked with effect for the future. Before authorizing transmission and before applying results, tenant status, instruction, mailbox, current ownership assignment and the relevant consents and permissions are checked again. After revocation, no further jobs are authorized for transmission; a job authorized for transmission immediately beforehand may still be sent to Microsoft or completed there concurrently with revocation. A result returned thereafter is not applied. Disabling the feature does not automatically delete leads already created or original messages.
Processing jobs contain source references rather than another copy of the full email text. Limited structured decision and status data are kept separate from usage and billing records that contain no message content. For completed jobs, jobs handed over for manual processing or cancelled jobs, decision records and error details are removed after 30 days in the next successful bounded cleanup run; disruptions or backlogs may delay actual cleanup. A minimal status record linked to the source is retained to prevent retransmission for as long as the original conversation or message exists; permanent deletion of the source also removes the associated job. Unresolved billing cases remain separately protected until clarified. Required billing records that contain no message content are subject to their own described retention rules and are not deleted by this queue cleanup. Original messages and created leads retain their regular access and deletion rules; reversible CRM archiving is not deletion. Job metadata may remain in the described encrypted database backups for up to twelve months. The separate records of the mailbox instruction and its revocation contain mailbox, owner and activating-user identifiers, ownership generation, text version and authorization and revocation timestamps, but no message content. They serve as evidence of permission to process, are retained for the lifetime of the tenant and are removed upon its permanent deletion. Deleting a message or mailbox does not automatically remove these separate records; the described database backup retention periods also apply to them.
Every provider call actually executed is charged against the company’s prepaid AI balance according to measured usage, including mail classified as irrelevant or spam and unclear classifications. Charges are based on actual provider costs plus 30 percent, converted to euros using a documented exchange rate and the applicable wallet rounding. Credit is reserved before transmission. An error without an executed provider call incurs no provider usage charge. If it is unclear whether a call was executed or how much it consumed, the reservation remains open until billing is clarified; no further chargeable request is made automatically.
The tenant remains responsible for the legal basis for processing its correspondence and for informing data subjects; technical activation does not replace senders’ consent. The legal requirements for permitted or tolerated private use must be examined separately. The feature serves organizational pre-sorting. It must not be used to derive solely automated decisions that have legal or similarly significant effects on individuals. Authorized persons must be able to review and correct possible misclassifications.
Semantic CRM email search (optional, disabled by default)
A tenant administrator can enable the function for the tenant. In addition, the respective mailbox owner or a mailbox manager expressly authorized for this purpose must authorize processing for the relevant mailbox. Without both authorizations, no email data will be transmitted to Azure for this function. The documented individual consent to AI data processing required under this section must also be in place beforehand.
The sole purpose is to suggest suitable “top matches” in the CRM dialogue for the manual assignment of a conversation to a lead. For this purpose, existing and future synchronized messages from the authorized mailbox are processed asynchronously and in minimized form: subject line and cleaned message text, as well as lead title, description, and tags. Technically detectable email addresses are removed from the text; other headers are not sent to Azure. Attachments, quoted message threads, and technically detectable signatures are excluded from vectorization. The search does not make any automatic assignment decision; the user reviews and confirms the assignment.
Microsoft Deutschland GmbH's Azure OpenAI Service is used as a processor for the creation of embeddings. The documented standard path uses the resource configured for struct-i-vio in Sweden Central within the Microsoft EU Data Boundary. The general Microsoft documentation does not specify a uniform fixed retention period for all resources and deployment types (see above).
If the function is disabled or an authorization is withdrawn, tasks for which authorization for transmission has not yet been granted will no longer be sent to Azure. A task authorized for transmission immediately before the withdrawal may still trigger an Azure request simultaneously with the withdrawal or be completed there. Any embedding returned thereafter is discarded and not stored in struct-i-vio; pending jobs and already stored derived vectors are deleted. If a source message is changed, its vector is recreated; if it is deleted, its vector is also deleted. The source message itself continues to be subject to the deletion rules described in Section 11.
The tenant controller must ensure the lawfulness of the processing and the required information of its correspondents. Mailboxes with permitted or tolerated private use may only be processed if the controller meets the legal requirements necessary for this purpose; information on this is available under Private use of business mailboxes.
AI-assisted company research in the CRM (optional)
The optional “Company research” function for the CRM module is enabled for your tenant by default; a tenant administrator can disable it entirely. When active, an authorised user can initiate AI-assisted internet research for each company; estimated costs are displayed before starting. The tenant administrator can additionally enable automatic research for companies newly linked to a lead (disabled by default, with a daily limit per tenant).
Scope of data processed: only the company’s master data is transmitted to the Azure OpenAI Service — name, legal form, postcode, city, country, website/domain, registration number and VAT ID. Contacts, notes, e-mail contents or other free-text fields of the company are not transmitted. The function uses the same integrated Bing-powered web search tool as the web search for lead pre-qualification described above; the information provided there about search sub-processing partially leaving the Microsoft EU Data Boundary applies correspondingly.
Approval and consent: company research uses the same tenant-wide approval for AI internet research as lead pre-qualification (see “Web search by AI” above); separate approval is not required. A manually initiated run additionally requires your own documented consent to AI data processing (see the “Consent” section above). An automatic run for a newly linked company has no acting user; in accordance with “Automated lead processing” above, documented activation by the tenant administrator is sufficient, and personal consent is not required for this.
Management information: where publicly available, research results may include names and functions of people holding official corporate positions (e.g. managing directors, executive board). The legal basis for processing this information about third parties is the tenant’s legitimate interest in establishing business relationships on an informed basis (Art. 6(1)(f) GDPR) — in line with the processing of lead contacts already described above. Telephone numbers, e-mail addresses and internet addresses are removed from these name details on the server side. Each displayed item of management information additionally carries the notice: “Management information comes from public sources as of the stated date and may be outdated or incorrect. Please verify before use.”
Blocking and objection (Art. 21 GDPR): an authorised user can block all research for each company individually, block only automatic research, or specifically block information about people. Blocking personal information removes existing management details from all stored research results for that company and ensures that future runs no longer request such information. If a person named there objects to processing of their data, the tenant deletes the relevant research results and then activates this personal-information block for the company.
No automatic import; retention: research results are exclusively for viewing by authorised users; they are not automatically imported into company master data. For each company, at most the three latest completed runs and, separately, the three latest failed or skipped runs are retained (by count, not by time); older runs are removed. Research results are also deleted when an authorised user deletes them or when the associated company is deleted.
Monthly report with AI interpretation (optional)
The automatic monthly report for the CRM module is enabled for your tenant by default; a tenant administrator can disable it at any time. When active, a report with team metrics for the previous month is generated on the first day of the month (e.g. new and contacted leads, won and lost leads with values, open pipeline) and made available in the application and as a PDF file. The metrics themselves are determined exclusively by database query and are free of charge; no AI processing takes place for this.
Optional AI interpretation: if the additional “AI interpretation” setting is enabled, the AI writes a short explanatory interpretation of the team metrics already calculated. Only the numerical or categorical team metrics themselves are transmitted to the Azure OpenAI Service (e.g. counts and breakdowns by source or currency); free text such as lead titles, descriptions or notes is not transmitted, since the AI component does not technically receive that information at all. A server-side check rejects the AI response if it contains a number not originating from the transmitted metrics. If AI interpretation fails or there is insufficient AI credit, the report is delivered without the AI part.
Sections per employee (§ 87(1) no. 6 BetrVG): the report may contain additional sections with metrics per employee. These are generated only if both the relevant telephone prospecting setting and the designated team permission are present at the time of creation; in stored metrics, employees are identified exclusively by their internal identifier, never by name — the identifier is resolved to a name only when viewed with authorisation. The optional AI interpretation structurally receives no employee-related data; the AI instruction additionally expressly prohibits naming individual employees or making them identifiable.
Recipients and delivery: recipients of the report are all active tenant members with the designated permission. Notification takes place by default as an in-app notice and additionally by e-mail; the e-mail contains only neutral notification text and a link to the report, no PDF attachment. Access through this link requires an active login and rechecks the right to receive the report at the time of access — an e-mail forwarded to third parties therefore grants no access without their own authorisation.
Retention: reports are retained indefinitely so that changes in metrics over time remain traceable; they contain no free text and, where present, employee-related sections use only internal identifiers instead of names. The tenant administrator can disable the monthly report and AI interpretation at any time.
Personal daily plan (optional)
The “Personal daily plan” function for the CRM module is enabled for your tenant by default; a tenant administrator can disable it for the tenant. Each person must additionally activate the function for themselves; it is never automatically active for other people. When activated, the AI creates a short, prioritised overview on working days at a personally chosen time of that person’s upcoming CRM tasks — such as leads due today, overdue leads and leads assessed as particularly promising — with direct links into the application.
Scope of data processed: the basis consists exclusively of leads and tasks visible to the respective person, limited by the permission system (Row Level Security); no additional data categories are processed beyond those already described under “Scope of data processed” above. AI explanations in the daily plan require the same consent as the AI assistant’s other functions (see the “Consent” section above).
Strictly personal: a person’s daily plan can be viewed exclusively by that person themselves — neither a tenant administrator nor a manager has access to it. At database level, read access is restricted to the owner of the respective row.
Retention: exactly one row per person is stored; it is overwritten each working-day morning by the new daily plan. No history of earlier daily plans is maintained. The row is deleted if the person disables the function for themselves or their account is deleted.
Delivery and costs: notification takes place by default as an in-app notice; an additional e-mail notification is disabled by default and can be enabled by the respective person themselves. Creation is billed as automated AI usage; the associated entry appears in the usage account in accordance with the “Usage-based credit billing” section described above.
12a. Email open and click tracking (Cold Leads module)
Where a tenant administrator uses the optional Cold Leads module to contact outbound contacts (cold leads) by email, struct-i-vio uses open tracking (1×1 pixel) and click tracking (link rewriting) in the marketing emails sent, provided that the relevant cold lead has given confirmed marketing consent (double opt-in).
The DOI confirmation email and the unsubscribe landing page are excluded from this tracking — these messages are never tracked.
Data processed
- Open event: Timestamp of retrieval of the tracking pixel, IP address, user agent (browser and device information). Note: Open rates are approximate and may be inflated because various email clients (e.g. Apple Mail with Mail Privacy Protection) automatically retrieve images in advance without the recipient actually having opened the message.
- Click event: Timestamp of the click on a link, destination URL, IP address, user agent. Click events are reliable.
The events are stored in the email_tracking_events table, which is assigned to the respective tenant. No tracking provider is integrated; tracking takes place exclusively on struct-i-vio's own infrastructure (Supabase, Frankfurt).
Purpose
Measurement of engagement with outgoing marketing emails (deliverability, opening and clicking behaviour) in order to enable the controller (tenant) to assess the effectiveness of its outbound communications.
Legal basis
Tracking takes place exclusively for cold leads that have given confirmed marketing consent (Art. 6(1)(a) GDPR) through the double opt-in procedure. This consent covers both contact by email and the engagement tracking described. No tracking is carried out without confirmed consent.
As data protection controller, the tenant is responsible for the lawfulness of the consent and of the processing based on it.
Withdrawal
The recipient of a marketing email may withdraw their consent at any time by using the unsubscribe link included in every email or by contacting the sender (tenant) directly. Following withdrawal (unsubscribe), no further tracking events are recorded, and the email address is added to the suppression list.
Retention
Tracking events are classified as operational business data and retained in accordance with the retention period for this class set out in section 10a (currently 90 days).
12b. Campaigns to contacts and CRM leads
Customers may optionally use the platform to send campaigns to their contacts and CRM leads. In particular, recipient address, recipient type, selected filters, campaign content, delivery status, and the structured legal basis confirmed by the customer are processed. The customer is responsible for the lawfulness of the recipient selection, documentation of marketing consent or a verified existing-customer exemption, and fulfilment of the information obligations.
The platform maintains an address-specific suppression list for each tenant. Unsubscribes and objections are stored there regardless of recipient type and prevent further campaign and outbound sending to the same address. The block remains effective even if the address is subsequently assigned to another contact, CRM lead or cold lead.
Entries in this suppression list are generally retained permanently so that a subsequent assignment of the same address does not render the unsubscribe ineffective. Deletion or early removal takes place only upon documented instruction from the controller; the controller then bears responsibility for the lawfulness of renewed contact. In all other respects, processing and controlled deletion are governed by the data processing agreement and the deletion and retention rules described therein.
For contacts and CRM leads, open and click tracking is technically deactivated and cannot be activated in the sending path. The separate tracking in the Cold Leads module is limited to cold leads with confirmed double-opt-in marketing consent and is described in section 12a.
12c. Personal API keys (optional programming interface)
struct-i-vio optionally provides a programming interface (API) through which employees with active access to a tenant can create a personal API key, revocable at any time, to connect their own tools or tools selected by the controller (e.g. n8n, Zapier, custom scripts) to the platform. Use of this function is voluntary.
Categories of data processed
- Key metadata: the name assigned by the employee, a public prefix for recognition and a cryptographic hash (HMAC) of the key. The key itself is shown only once when created and is subsequently never stored in plain text anywhere.
- Timestamps and status: creation and expiry time, revocation time and reason, and an approximate time of last use (updated hourly, displayed to the day; no IP address).
- Technical control data: Data for rate limiting and preventing accidental duplicate processing (idempotency) — in existing API procedures, the latter is stored only as a hash, without request content, for no more than 24 hours. For submitted external invoice data, separate ingestion receipts store tenant, source account, user and batch identifiers, the technical authorization association, request checksum, timestamp and original ingestion status and version identifiers, without the full request text or financial amounts. These receipts are no longer accessible after 24 hours; expired receipts are removed in bounded hourly cleanup runs. Failed runs or backlogs may delay actual removal. When a source account is deleted as described below, the receipts and their checksums are also removed.
- Technical API runtime logs: limited operational data such as request ID, HTTP method, technical route identifier, status, duration, the technical identifier of the API key used (not the key itself), a shortened tenant reference and standardized error codes. The application does not write request or response content, API keys, tokens, headers, query values or signed URLs to these logs. As part of its platform logs, the hosting provider may additionally process network and connection metadata, such as IP address, time and user agent, according to the retention period applicable to the Pro or Observability Plus plan actually booked. This provider-side retention is separate from the application logs we control; the latter are retained for this processing for at most 30 days.
- Business data processed through the API: Business data read or written using an API key, such as CRM data (e.g. leads and lead documents) and bank transactions, corresponds in substance to data the same employee could process through the regular user interface; this creates an additional access channel rather than a new data category. Authorized persons may also submit data on already issued outgoing invoices from customer-owned systems. Limited source account, document and version identifiers, invoice number and date, document type and status, correction references, currency and signed net, tax and gross amounts with tax groups are stored. Customer snapshots, names, addresses, contact details, free-text line item descriptions, credentials, PDF files and other attachments are outside this submission procedure. Identifiers and invoice numbers may nevertheless be personal or confidential. struct-i-vio processes this business data on the customer’s behalf under the existing data processing agreement and the customer’s documented instructions. The customer operates the source system connection and automation using its own tools; struct-i-vio does not retrieve data from the invoice provider or store credentials for that provider for this purpose.
- Additional business data and saved invoice analyses: Through the same API, authorized persons may manage contacts and companies, read limited project master data, record or read their own working time entries, and read or create CRM lead comments. Project budgets, financial postings created in struct-i-vio, absences, sickness records and additional file functions are not made accessible by the external invoice submission procedure. Separate access to bank transactions is described in the next item. For submitted outgoing invoices, monthly reports in EUR based on invoice date may be saved and later opened and exported with their underlying invoice versions. They contain the selected document versions, amounts, tax groups, validation notes and information about the scope of submitted data. Later submissions do not overwrite earlier reports; an update creates a new report version. Source month completeness is initially unknown; reconciled customer declarations are identified as customer-attested, not verified with the provider. This internal analysis is not a binding tax calculation, tax return or statutory invoice archive and remains separate from the existing payment-based bank analysis.
- Bank transactions and receipts: Persons who, in addition to their functional permissions, hold the separate API permission for transactions can use the API to read and filter bank transactions and export them as a CSV file, create individual transactions, assign them to projects, profit and loss positions or the company fund, split them and treat them as normal or pass-through (transit), and upload, download and link receipts and move them to the recycle bin. Persons at the highest permission level (level 0) always hold this separate permission; other levels hold it only if the customer grants it to them in its permission management. The data processed are transaction, booking and value dates, amount, currency, tax details, purpose line, counterparty name, account name, cost center, the assignment to projects, positions and the company fund, splits with reasons, receipt status, treatment type, and the file name and content of receipts. In addition, invoices that may match a transaction can be retrieved with invoice number, date, amounts, status and project, as well as summarized assignment suggestions from earlier transactions.
- Information not output through the API: The data fields for IBAN, BIC and account number, raw bank data, internal notes and the record of which user assigned or verified a transaction are not output through the API. IBAN, BIC and account number are also not accepted through the API. Account names are not output if they recognizably contain an IBAN, BIC or longer account number; this check does not recognize every bank identifier, and it does not filter out the names of account holders.
- Free text and receipts: Purpose lines, counterparty names, reasons, file names and receipts originate from banks, business partners or users and are not filtered for content. Despite the restriction above, they may therefore contain account identifiers and information about natural persons, for example on salaries, expenses or payments by employees, customers and suppliers. In individual cases they may also reveal special categories of personal data under Art. 9 GDPR, for example indications of health, religious affiliation or trade union membership. If such content is retrieved through the API, it reaches the connected system.
- Evidence of DPA acceptance: UUID of the accepting user account and associated employee identity, tenant UUID, accepted DPA version, server-side timestamp and established basis of authority (owner or tenant administrator). The evidence is retained together with the tenant's contractual lifecycle and deleted by cascading deletion when the tenant is deleted, unless statutory or legally ordered retention requirements prevent this.
Legal basis
Key metadata are processed on the basis of Art. 6(1)(b) GDPR (performance of the contract, provision of the optional function) and, for rate limiting, logging creation and revocation, and protection against abuse, on the basis of Art. 6(1)(f) GDPR (legitimate interest in platform security and traceability of access).
For bank transactions and receipts processed through the API, the customer is the controller; struct-i-vio processes them as a processor under the data processing agreement. The customer determines the legal basis for this processing and for disclosure to connected tools and reviews it under its own responsibility.
Storage period
A key is valid for at most 366 days and may be revoked at any earlier time. Revocation and any permanent loss of access permission (e.g. termination of employee assignment or deletion of the user account) take effect from the next authenticated API request using the key; API requests already in progress still complete (max. 60 seconds). The associated database row does not contain the key in plain text and cannot be used for authentication after revocation or expiry. No additional automatic deletion after a fixed period is currently configured for expired or revoked key metadata. Deletion follows applicable deletion rules and legitimate deletion requests.
Download links from the public lead document API are bound for 60 seconds to the original API key and user; every retrieval checks the Authorization header and current rights again. Project files are delivered through authenticated routes with a renewed rights check. Signed storage links issued before this change remain usable until their former expiry (up to 60 seconds for the public lead document API and up to one hour for general file routes).
Bank transactions and receipts created or changed through the API are subject to the same retention, export and deletion rules as the corresponding data from the user interface. Through the API, receipts can at most be moved to the recycle bin; final removal follows the recycle bin periods described in section 10. Only transactions previously created through the API, an external AI client or the integrated AI assistant that have not yet been assigned or verified, have no receipt and are not otherwise protected can be permanently deleted through the API. Imported bank transactions cannot be deleted through the API.
Larger receipts are provided through a signed download link valid for at most 60 seconds. During that time anyone who knows the link can retrieve the receipt; no API key is required.
Key data security
An API key is stored exclusively as a cryptographic hash (HMAC); the plain-text key does not leave the creation process again and is never logged. Creating a new key is possible only through a regular, recent sign-in (no older than 15 minutes) in the user interface, rather than through an API key itself. Every request using an API key is checked individually; access permission is strictly bound to the tenant specified at creation and reconfirmed server-side for each request.
Extended API functions
The API does not extend the employee’s permissions. Visibility and write permissions are checked for every call against the current user, tenant, module, project and action permissions. Lead comments created through the API remain regular lead data. An activity note with the channel api and a preview of no more than 80 characters is also created. This operational activity row remains as a business data row until the lead or activity is deleted. The full comment text is additionally retained in the audit log for 90 days; the separate metadata-only audit copy of the activity row contains neither the comment text nor the 80-character preview. For external invoice data, submission permissions and report reading permissions are checked separately. Submission permission alone allows neither reading saved financial values nor exporting them. Opening, viewing individual details and exporting a saved report continue to require current permissions in the associated tenant. Deleting or removing the submitting person does not automatically delete company invoices and reports; personal associations that are no longer required are removed or anonymized under the applicable deletion rules.
The technical API runtime logs we control are used solely for operations, troubleshooting and abuse prevention. Their limited retention ends no later than after 30 days; longer product-side retention is not configured. Provider-side network and connection metadata is governed separately by the Pro or Observability Plus plan actually booked. The separately described audit evidence and regular business data remain unaffected. Submitted external invoice versions and saved monthly reports are retained as company data for the duration of the existing processing engagement. On an appropriately scoped, documented instruction from the customer acting as controller, support can delete an entire source account. This removes all associated invoice versions, reports, completeness declarations, ingestion receipts, request checksums and derived financial copies from the live data; other source accounts and financial data recorded in struct-i-vio remain unaffected. Automatic deletion of individual invoices is not promised. More narrowly scoped data subject requests are assessed through the existing procedure with the controller. Following source account deletion, only necessary source account and blocking identifiers and minimal protected evidence of the deletion and its instruction remain for the lifetime of the tenant, without financial values, deleted content or checksums of that content. These prevent delayed retry requests from refilling the source account; renewed authorization requires a separate explicit customer instruction. When the tenant is deactivated for deletion, external invoice data and reports are removed immediately; any otherwise longer deletion schedule for the tenant does not extend this retention. Export or return takes place before this deletion instruction. Final tenant deletion also removes the remaining source account and blocking identifiers. Encrypted database backups may still contain previously stored data for up to twelve months. After a restore, this invoice functionality remains blocked for app and API access; backup copies do not constitute current authorization, and automatic reopening or restoration of old reports is not promised. Subsequent authorization requires a separately reviewed restoration decision. Exports already downloaded are independent customer copies and cannot be recalled by deleting data in struct-i-vio.
Changes to bank transactions, splits and receipts triggered by an API key are recorded in the change log with the acting user and a reference to the API key used. These entries may contain the changed record including fields that are not output through the API, such as IBAN, raw bank data and notes. Under section 10a the retention period is 365 days for entries on transactions (financial data) and 90 days for entries on splits, receipts and receipt links (operational business data). Read access and exports are not individually recorded in the change log, only in the technical runtime logs described above. Search terms are part of the request URL. The application does not write them to its logs; however, they may appear together with the request path in the hosting provider’s platform logs mentioned above.
Connected customer tools are not subprocessors
Tools, services or scripts connected using an API key (e.g. n8n, Zapier, custom scripts) are systems selected, configured and secured by the controller. They are not subprocessors of struct-i-vio within the meaning of Art. 28 GDPR. struct-i-vio has no insight into their processing and accepts no responsibility for it.
This also applies to bank transactions, receipts and CSV exports. On retrieval, this data leaves struct-i-vio and reaches the connected system. The customer is responsible for its storage, disclosure and deletion there, for any agreement required under Art. 28 GDPR with the tool’s provider, and for transfers to third countries.
Information for employees about the administrator function
Authorized administrators of the respective tenant may revoke all keys of a named employee (e.g. where compromise is suspected) and for this purpose see only the number of that employee's active keys, without key names or usage data.
12d. Telephone cold calling (Cold Leads module)
When a tenant administrator uses the call function in the optional Cold Leads module to contact outbound contacts (cold leads) by telephone, struct-i-vio processes the call-related data described below in addition to the data described in section 12a. struct-i-vio operates no telephone system of its own for this purpose: the call itself takes place through the telephone or telephone system of the person making the call. No call content (audio) is recorded, transmitted or stored; the platform records exclusively the time, result and context of the call attempt entered by the caller.
Data processed
- Call attempts: caller, time (start, end, duration, where recorded), linked cold lead, contact and company, and the immediate result (not reached, busy, voicemail, conversation).
- Conversation result and note: where a conversation took place, the result configured by the tenant (e.g. “not interested”, “callback requested”) and an optional free-text note entered by the caller (up to 2,000 characters) about the conversation context.
- Follow-ups: a date or date and time for renewed contact depending on the result; this creates a regular task with the same content as the note, assigned to the caller.
- Warning for “not interested”: before calling a company or contact again that was previously assigned a result configured as “not interested”, the platform displays a warning with the time and person responsible for the previous classification. The caller can acknowledge this warning and still make the call; the acknowledgement (override) is logged.
- Objection to further contact (Art. 21 GDPR): if a contacted person states during the call that they do not want further promotional contact, the caller records this using a separate, non-configurable function, apart from the regular conversation results. The platform then permanently stores that this person — and, if additionally selected by the caller, the associated company — must not be called or contacted by e-mail in future.
- Metrics: metrics derived from call attempts (including number of attempts, results, success rate) for the individual themselves and as team totals. A breakdown by individual employees is disabled by default (see “Metrics per employee” below).
Purpose
Documentation and management of telephone sales outreach, avoidance of repeated unwanted calls and implementation of objections to further contact.
Legal basis
Internal documentation of calls (logging, metrics) takes place on the basis of Art. 6(1)(f) GDPR (the tenant’s legitimate interest in traceable sales management). The tenant alone, as controller under data protection law, is responsible for the lawfulness of the call itself — in particular the existence of presumed consent for business customers or prior express consent for consumers (§ 7 UWG), and fulfilment of the information obligations under Art. 13 / 14 GDPR towards the person called. struct-i-vio checks neither the existence nor evidence of such consent or legal basis; the free-text note may serve as the tenant’s own documentation basis, but struct-i-vio does not analyse its content.
Metrics per employee
A breakdown of call metrics by individual employees (rather than only as team totals) is a technical system for monitoring performance and conduct within the meaning of § 87(1) no. 6 BetrVG and is therefore disabled by default. When disabled, the team view displays totals only when there are at least three different callers in the period being considered, so that even small or narrowly filtered analyses cannot in practice be attributed to an individual. Regardless of this setting, each person can always see their own figures, and the optional AI assistant always provides only their own figures or team totals on request, never values for individual employees. Activation is the responsibility of the tenant as employer, who must observe existing co-determination rights.
This setting does not affect the case-specific call history of a cold lead or company: it shows everyone with access to the call list (“View cold leads” permission) who called when and with what result — regardless of whether the above metric breakdown is enabled. This history serves to trace individual contact attempts (e.g. to avoid duplicate or contradictory contact) and is not a tenant-wide metric breakdown in the above sense.
Objection
An objection to further contact declared during a call takes effect immediately, permanently and across channels (telephone and e-mail) and cannot be overridden by anyone — including the tenant itself; it applies regardless of the cold lead’s other status. The data subject can also declare an objection directly to the tenant at any time.
Retention
Call attempts and conversation results are classified as “Operational business data” within the meaning of section 10a (change log: 90 days); the records themselves remain as long as the associated cold lead is present on the platform and are removed when it is deleted. An objection to further contact is retained independently of the individual cold lead and remains effective even after its deletion, archiving or conversion into a lead (change log class “Master and contractual data”, 365 days; see section 10a). If the associated contact or company is completely deleted, the telephone number recorded when the objection was captured remains as a separate warning record; the associated free-text note is deleted. A renewed call to the same number then triggers a warning which the caller can override only after expressly accepting responsibility — unlike the non-overridable block described above where the contact or company still exists, because after deletion a telephone number alone can no longer be unambiguously attributed to the same person (e.g. shared numbers or reassigned numbers). The stored number cannot be viewed by anyone through the user interface and serves exclusively for technical matching on future calls. For contacts with a stored e-mail address, the contact block additionally remains permanently effective without an override option through the address-based blocklist described in section 12b. Removal of this telephone warning record is not provided through the platform; it is reserved for a separately documented measure by struct-i-vio as processor, reviewed in the individual case.
12e. Connection of external AI clients (optional MCP interface)
struct-i-vio provides the Model Context Protocol (MCP) interface by default for existing and new tenants unless it has been disabled. Employees of a customer with active access and the required permission can use it to connect external AI clients of their own choice — third-party AI applications such as Claude, ChatGPT, Cursor, Visual Studio Code, Claude Code or Codex, or other compatible applications — to their user account. The client can then read platform data and execute actions in the employee's name and with their permissions. Default provision alone does not connect a client or send business data to it. Before use on this basis, the owner or an administrator of the highest permission level (hereinafter ‘administrators’; administrators of lower permission levels are excluded) must expressly accept the DPA and Terms § 5f and issue the MCP instruction. Previously documented express activations remain valid under their existing conditions. The employee authorizes each individual connection personally. Use is voluntary. Deactivation and existing permissions remain in place; default provision grants employees no additional permissions.
This section must be read separately from section 12, which describes the AI functions integrated into struct-i-vio: its statements regarding consent, data location, processing by Microsoft on behalf of the controller, exclusion from training, and confirmation of write actions do not apply to external AI clients. The contractual provisions are set out in § 5f of the General Terms and Conditions and in § 3f of the DPA.
Roles and responsibility
The customer is the controller for the data in the tenant; struct-i-vio processes it as a processor. The providers of the external AI clients are recipients within the customer's sphere of responsibility and not sub-processors of struct-i-vio within the meaning of Art. 28 GDPR: struct-i-vio has no contract with them, does not select them, does not recommend them, and has no insight into their processing. Depending on the plan and contract between the customer and the client provider, the latter acts for the customer as a processor or as an independent controller. The customer decides and verifies whether and which clients and plans employees may connect and whether a legal basis, an agreement pursuant to Art. 28 GDPR, appropriate training settings and — in the case of transfers to third countries — a basis pursuant to Chapter V GDPR exist for this purpose. The same applies to the entry in its record of processing activities, informing employees, and safeguarding co-determination rights.
Data processed
- Business data retrieved by the client: When a function is called by the client, the data transmitted are the data that the employee is personally permitted to view in struct-i-vio and that are returned by the called function (e.g. leads, contacts, projects, times, tasks, bank transactions with purpose line and counterparty name, and file names of receipts). As with the integrated AI assistant, the results are reduced to the extent necessary in each case and limited in quantity. They may contain personal data of employees, business partners, payment counterparties, leads and correspondence partners and thus reach the client and its provider. The content of receipts and CSV exports of bank transactions are not provided through this interface. No new data category is created, but there is an additional recipient.
- Information about the person and the tenant: in particular the employee's name, tenant, and authorizations relevant to the use of the functions, insofar as the client requests them.
- Actions performed by the client: The data created or modified as a result are the customer's regular business data and are subject to the same rules as entries through the user interface. This also applies to bank transactions: with the separate permission for transactions via API and AI clients and their functional permissions, employees can use a client to create, assign and split transactions, remove splits, link transactions to the company fund or unlink them, delete unassigned transactions without a receipt that were created through the API, a client or the built-in AI assistant, change receipt links and move receipts to the recycle bin. struct-i-vio cannot tell whether the client asked for confirmation beforehand.
- Connection data: allocation of the connection to the employee and tenant, the name and redirect address of the client as specified by the client itself (not verified), the times of granting, last use, expiry and revocation, the version of the notice text displayed upon confirmation, and — exclusively as a hash value — the access and refresh tokens. The client's name is visible to the employee in their own connection overview and appears in the security notice for a new connection (unless this has been switched off in the notification settings); Administrators see it only in the client type list without allocation to individual persons (see below). When activated by the customer, the acting person, time, selected scope and text version are additionally stored as tenant-wide evidence; this evidence appears in the change log. For the separate company acceptance and instruction, the accepting person, their basis of authority, time, DPA version and text versions are also stored. A system default is identified separately as such and is not presented as customer acceptance. Minimized contractual and instruction evidence is retained for the duration of the authorization based on it and removed as part of termination and deletion of the tenant; it contains no access tokens or business data.
- Action log (“AI activity”): For each write action performed through a client, the time, tool, called function, data sent (without access secrets), result, undo information and the client's name are logged. This log can be viewed only by the acting employee themselves. The client's name is recorded exclusively there for individual actions.
- Change log (audit log): Changes made by a client on behalf of an employee appear in the change log like the employee's own changes (section 10a). It only indicates there that the change was made through an external AI client (channel identifier “external AI client”); the client's name does not appear there. A technical connection identifier is additionally stored internally with the entry and is not displayed in the application. Establishing and disconnecting a connection are not recorded there as connection events for each person; the connection data are not part of the change log. Activation, change of scope and deactivation of the function appear there as tenant-wide evidence (who, when, scope, text version).
- Technical control data: Data for rate limiting (including a hash value of the IP address, not the IP address itself) and for preventing duplicate execution (idempotency: tool, retry identifier, hash value of the arguments, status and the truncated response of the action; stored for no more than 25 hours).
- No transfer to struct-i-vio AI services by the interface itself: The search for work instructions for clients is carried out purely lexically on struct-i-vio systems. If a client calls a function that itself uses the AI services described in section 12 (e.g. AI-assisted company research), it is performed only if the requirements applicable to that function — in particular the personal AI consent pursuant to section 12 — are met; section 12 then applies to that function.
Excluded data and functions
Data and functions that are not accessible to the AI assistant integrated in struct-i-vio remain blocked for external AI clients as well. This includes, in particular, special categories of personal data (Art. 9 GDPR, e.g. health information in absence documents), performance and behavioural metrics of individual employees, the setting for breaking down call metrics by employee (section 12d), and settings through which consents or approvals for AI processing are granted. Free text is, however, not filtered for content: purpose lines, counterparty names, file names and other free-form entries may in individual cases contain information revealing special categories of personal data, for example indications of health, religious affiliation or trade union membership. If a client retrieves such content, it reaches the client and its provider. Functions that can be operated only in the user interface are likewise not available (e.g. uploading files in the AI chat), nor are functions for managing the user account, API keys and connections, and administering the tenant.
Immediate execution and decisions concerning persons
Actions are performed as soon as the client calls them; unlike with the integrated AI assistant, struct-i-vio does not obtain any additional confirmation. The employee can provide for a request for confirmation only in their client's settings. The personal AI setting “With confirmation” applies only to the integrated AI assistant; the setting “Never”, by contrast, also blocks an action for external clients. Some actions cannot be undone; deletions may be final. To the extent that an action can technically be undone, the personal “AI activity” overview provides this option.
struct-i-vio does not make decisions concerning persons. If an employee triggers through an external AI client an action that prepares or implements a decision concerning other persons (e.g. approving or rejecting absence or time applications of colleagues), this is an act of the employee and therefore of the customer. The customer as controller must ensure that such decisions meet the requirements of Art. 22 GDPR and other applicable law, in particular that they are not based solely on automated processing unless an exception applies. Such actions are identified as separate tools in the MCP interface so that clients can ask before execution, and may be restricted by the customer to “Max. confirmation” or “Blocked” in the tenant-wide AI action policies (“AI Action Floors”); the MCP interface will then not perform them. The same applies to any other individual action.
Legal basis
The processing of connection, log and control data is carried out on the basis of Art. 6(1)(b) GDPR (provision of the optional function) and — for rate limiting, abuse protection, security and traceability of access — on the basis of Art. 6(1)(f) GDPR (legitimate interest in the security of the platform). The provision of business data to the client connected by the employee and the performance of actions called by the client are carried out as a processor in accordance with the customer's documented instruction (activation by the authorized administrator, connection by the employee). The customer as controller must verify the legal basis for transferring the data to the client provider.
Recipients and third-country transfers
The recipients of the data retrieved via the interface are the client selected by the employee and its provider. struct-i-vio transmits data only when an employee has expressly connected a client and that client retrieves it; struct-i-vio does not transmit data to client providers on its own initiative. Client providers may have their registered office or processing locations outside the EU or the EEA (in particular in the USA). The commitments in Section 12 regarding data location (Microsoft EU Data Boundary), processing on behalf, and exclusion from training apply only to the Azure services used by struct-i-vio itself and do not apply to client providers. From the point of transfer to the client, the client provider's terms apply; struct-i-vio can neither retrieve nor delete data already transmitted there. Retention, use for training purposes, and transfers to third countries are governed by the plan and settings of the customer or employee with the client provider.
Storage period and termination
A connection ends after 30 days without use, at the latest after 365 days, by disconnection (by the employee in the account menu under “API keys” in the “Connected AI clients” section or in the client; by administrators for all connections of a specifically named employee), by switching off the function for the tenant, and upon the permanent loss of access authorization (e.g. termination of the employee assignment, deletion of the user account). The termination takes effect from the next request via the connection; processes already in progress at that time will still be completed, and this may take up to five minutes for longer-running functions (e.g. research). Access and refresh tokens can no longer be used after the connection has ended. No fixed automatic deletion period is promised for connection data and the action log; they are removed upon deletion of the tenant or in accordance with the general deletion and anonymization rules upon deletion of the user account (Section 10). Section 10a applies to the change log.
Data security
A connection is authorized exclusively by the employee personally in the struct-i-vio user interface (regular, current login — no more than 15 minutes old —, selection of the tenant, and express confirmation); the confirmation page cannot be embedded in third-party pages. Access tokens, refresh tokens, and authorization codes are stored exclusively as hash values, are short-lived or are rotated upon each renewal; reuse of a code already redeemed or a refresh token already renewed terminates the connection, except in the case of simultaneous renewal by the same client within 30 seconds. A connection applies to exactly one tenant and only to the designated functions of the interface, not to the administration of the user account, API keys, or connections, and not to the administration of the tenant. Each request is checked against the current authorization status; use is subject to rate limits. The name specified by the client and its redirect address are not verified and are identified as such upon confirmation; redirect addresses are technically restricted. This restriction does not constitute verification or a recommendation of a client.
Notice for employees regarding the administrator function and visibility
Tenant administrators can disconnect all connections of a specifically named employee. In the overview provided for this purpose, they see, for each employee, only the number of connections — neither the name of the client nor usage or content. In addition, the overview displays a list of the client types connected in the tenant with their number, without allocation to individual persons. The list appears only once at least three persons have active connections. This reduces allocation to individual persons, but does not rule it out: in small companies, the number and client list may make it possible to infer which clients individual persons have connected. The action log containing the name of the client can be viewed only by the employee taking the action; changes to the data stock appear in the tenant's change log as described above. struct-i-vio does not evaluate this data for monitoring employees' performance or behavior.
13. Special categories of data (Art. 9 GDPR)
The application enables employees to upload certificates of incapacity for work as part of absence management. These documents may constitute health data within the meaning of Art. 4 No. 15 in conjunction with Art. 9 GDPR.
The legal basis for this processing is Art. 9(2)(b) GDPR (processing in the context of employment law) and § 26(3) BDSG. The employer (customer of struct-i-vio) is responsible for the lawful collection of these data.
Employees are advised to upload only the certificate of incapacity for work (without diagnosis). Uploading documents containing diagnostic information or other health data beyond the certificate is not necessary and should be avoided.
14. Use of the mobile app (Android / iOS)
struct-i-vio is currently available as a mobile app for Android. An iOS version may be made available in the future. The app is a web-based application (Capacitor/WebView) and processes the same data as the web version.
Device access
The mobile Android app technically declares the following system permissions: internet access as well as supporting permissions for network status, vibration, wake lock, and the receipt of push notifications (including technical permissions for the push receipt service). Notifications (push) are used only following the user's explicit consent. Camera, microphone, location, or contacts are not currently actively requested as runtime permissions. Should more extensive device access become necessary in the future, this will occur only following prior privacy notice and – where required – with separate consent or via the respective operating system dialog.
Account deletion
Users may request account deletion. If in-app account deletion is unavailable, the request may be sent by email to tigran.varazhian@struct-i-vio.com; it is not submitted automatically when an attempt fails. We handle erasure requests without undue delay and will inform you of the measures taken no later than one month after receiving the request. Where necessary due to the complexity or number of requests, the response period may be extended by up to two further months; we will inform you of the extension and its reasons within the first month. After review, personal data is deleted or anonymized without undue delay unless statutory retention duties prevent it, and access is terminated. Technically necessary backups follow their separately described deletion cycles; those cycles do not delay deletion of data no longer needed in the active production environment. Files assigned exclusively to the user account may require separate removal. Ordinary company work documents remain subject to the applicable trash, retention, and deletion rules. Personal language preferences tied to a tenant are removed from the active data set when a person permanently leaves a tenant, when an account is deleted, or when a tenant is deleted. If membership ends solely because a recorded departure date takes effect, cleanup runs in limited hourly batches, no earlier than 24 hours after the membership record was last changed. Temporary suspension may preserve the inaccessible preference. Backlogs or failed runs may delay actual removal.
You can find a public description of the deletion process at /legal/account-loeschung.
15. Public demo access
At /demo, we offer interested parties publicly accessible, read-only test access to an environment containing exclusively fictitious sample data. In addition to the preceding sections, this section describes which data are processed in this context.
Data processed
When you submit a request via the form, we process your email address (mandatory information) as well as, optionally, your name and company, the times of your consents, source and campaign information (e. g. UTM parameters), where available, as well as technical status information and timestamps of the request and a salted hash of your IP address. A plaintext IP address is not stored at any time; the hash is used exclusively for rate limiting and abuse detection on this publicly accessible form route and cannot be traced back to an IP address without a server-side secret that is not stored in the database. If you subsequently use the AI assistant available in the demo, the chat messages described in Section 12 are added.
Legal bases
For the setup and provision of demo access itself (creating access, sending the login link, enabling the read-only environment), we rely on Art. 6(1)(b) GDPR: you request this access yourself, so processing takes place at your instigation in the context of initiating potential future use of struct-i-vio. For the processing of your inputs by the AI assistant, we additionally obtain your explicit consent pursuant to Art. 6(1)(a) GDPR, as this processing is not strictly necessary merely to view the demo. You confirm both by means of the same checkbox in the form; without this checkbox, we do not set up demo access. Any contact for sales purposes is subject to separate, voluntary consent pursuant to Art. 6(1)(a) GDPR via a second, independent checkbox, which has no effect on access itself. For the IP hash as well as source and campaign parameters, we rely on our legitimate interest in operating this public form route in a functional manner free from abuse (Art. 6(1)(f) GDPR).
Pseudonymization within the demo environment
Within the demo environment itself, you do not appear under your real identity. We create a pseudonym for you ("Demo Guest <Nummer>" with a non-deliverable sample email address) and use this pseudonym exclusively for your user account within the demo tenant. Your actual email address, your name, and your company are not included at any time in the area visible to other demo visitors; they remain exclusively in our registration system and in a separate repository for access requests with strictly restricted access. Other visitors to the demo therefore do not see your real details at any time.
AI use in the demo
The same provisions as described in Section 12 generally apply to the AI assistant (Luna) available in the demo: your inputs are transmitted via the Azure OpenAI Service to Microsoft Deutschland GmbH as a processor and processed within the Microsoft EU Data Boundary; no AI models are trained on these data, and no automated decisions within the meaning of Art. 22 GDPR are made. Unlike the regular process, we obtain your consent to AI processing not only when you first use the assistant, but already through the checkbox in the form at /demo. In the demo, only a single model intended for this purpose with limited functionality is also active (including without internet research, dictation, document creation, or attachments), and your usage volume is limited.
Recipients
To send the sign-in link, we use the email delivery service enabled for that channel: Resend until its separate migration, then Amazon SES (see Section 8). It receives only your email address and sign-in link, not the other form details. When using the AI assistant, the recipient group described in Section 12 also applies (Microsoft Deutschland GmbH, Azure OpenAI Service).
Storage period
We store your access request (email address, name/company where provided, consent times, source information, IP hash, status) for 180 days from your request; it is then automatically deleted. Conversations with the AI assistant in the demo are removed considerably earlier as part of regular automated maintenance of the demo environment, irrespective of the 180-day period for the access request.
Withdrawal and deletion
You may request the early deletion of your access request and withdraw your consents at any time with effect for the future. To do so, use the contact address specified in Section 4. Your access to the demo environment will be deactivated in this case. Since only a pseudonym is stored within the demo environment in any event, such deletion concerns your registration record and your access request – not the fictitious sample data of the demo environment itself.
No contractual relationship
For the data collected in connection with demo access, we are the controller within the meaning of Art. 4(7) GDPR for data protection purposes, not a processor. Requesting demo access does not establish a contractual relationship for use of the SaaS platform; the General Terms and Conditions and the data processing agreement (see Section 7) therefore do not apply to demo access.
16. Changes to this privacy policy
This privacy policy may be updated from time to time. You can always find the current version on this page.
Last updated: 12 November 2026 (Privacy Policy, version 2026-11-12.1)
