IMScollect
Back

Data processing agreement

As of 17 November 2026

This is a translation provided for convenience. Only the German version is legally binding: Vertrag über die Auftragsverarbeitung.

between

the customer, with company name and address as stated in the accepted offer or in the other Main Contract (§ 1(1))
(hereinafter the “Controller”)

and

IMScollect GmbH, Scharnhorststraße 77a, 48151 Münster
(Amtsgericht Münster, HRB 23870)
(hereinafter the “Processor”)

§ 1 Subject matter, term and order of precedence

(1) The Processor makes the IMScollect platform available to the Controller for running its management system (hereinafter the “Main Contract”). Free-of-charge access is also a Main Contract, in particular a trial access under § 3(2) of the General Terms and Conditions and partner access under the Partner Access Addendum (Zusatz Partnerzugang) to the General Terms and Conditions. If the parties conclude a paid contract for the same organisation following a trial access, this agreement continues to apply to it without needing to be concluded again. In doing so, the Processor processes personal data on behalf of the Controller. This agreement governs the obligations of the parties under Art. 28 GDPR.

(2) The Controller remains responsible for the lawfulness of the processing (Art. 4(7) GDPR). It decides on the purposes and means.

(3) This agreement runs for as long as the Main Contract runs. It ends with the Main Contract, but not before the obligations under § 12 have been performed in full.

(4) In the event of conflicts, the provisions of this agreement take precedence over those of the Main Contract and of the General Terms and Conditions insofar as they concern data protection.

§ 2 Nature, scope and purpose of the processing

(1) Purpose. Provision and operation of the IMScollect platform for setting up, running and auditing a management system in accordance with ISO standards.

(2) Nature of the processing.Collection, recording, organisation, structuring, storage, adaptation, retrieval, consultation, use, disclosure by making available, restriction, erasure and destruction of the data entered by the Controller and its users. Sending of system messages to the persons designated by the Controller. When triggered by a user: processing of content, including uploaded files, by the AI-supported functions and the AI advisor of the platform, including storage of the AI advisor's conversation histories, in each case in accordance with § 7(5).

(3) Categories of data subjects.

  • Employees of the Controller, including executives and appointed officers
  • Contact persons at suppliers, customers and other interested parties of the Controller
  • External auditors and consultants to whom the Controller grants access
  • Persons whom the Controller surveys via the survey module

(4) Categories of personal data.

  • Master data: name, business contact details, personnel number, date of joining and date of leaving
  • Organisational data: company, site, department, function, role, responsibilities
  • Qualification data: competences, training attendance and training records, certificates, project references
  • Activity data: tasks, deadlines, acknowledgements, contributions to audits, nonconformities, processes and documents
  • Access data: email address, time of the last login, status of two-factor login
  • Conversation histories of the AI advisor: users' requests and the AI advisor's answers and suggestions, including the Controller's content reproduced in them
  • Responses from surveys and customer feedback
  • Files uploaded by the Controller and its users (documents, evidence, training content, certificates, attachments), including their file names
  • Usage data of the AI-supported functions: organisation, account, function, time and extent of the call, without content

Special categories of personal data under Art. 9 GDPR are not the subject matter of this agreement. The Controller ensures that such data is not entered into the platform. If it intends to do so, for instance in connection with occupational health care or fitness examinations, it informs the Processor in advance; the parties then conclude a separate agreement and examine whether a data protection impact assessment is required.

(5) Place of processing. Exclusively within the European Union, with the exception of the cases expressly designated in Annex 1.

§ 3 Instructions of the Controller

(1) The Processor processes the data only on documented instructions from the Controller. This also applies to transfers to a third country, unless the Processor is required to do so by Union or Member State law. In that case, it informs the Controller of that legal requirement before processing, unless that law prohibits such information on important grounds of public interest.

(2) The provisions of this agreement and the use of the functions of the platform by the Controller and by the users authorised by it are deemed to be instructions. This also includes a request by an authorised user to the AI advisor; such a request is an instruction only within the scope of that user's permissions and in accordance with § 7(5). Individual instructions beyond this are given in text form to kontakt@imscollect.de. The Processor is not obliged to carry out an individual instruction that requires a change to the platform. The Controller remunerates the effort for individual instructions in accordance with § 11(4).

(3) If the Processor considers that an instruction infringes data protection provisions, it informs the Controller without undue delay. It is entitled to suspend execution until the instruction is confirmed or changed.

§ 4 Obligations of the Processor

(1) The Processor processes the data only within the scope of this agreement and the instructions. The data is not used for its own purposes.

(2) The Controller's data is not used to train AI models. The Processor also ensures this contractually with the sub-processors named in Annex 1.

(3) The Processor maintains a record under Art. 30(2) GDPR.

(4) The Processor informs the Controller without undue delay if a supervisory authority takes measures against the Processor that concern the Controller's data.

§ 5 Confidentiality

(1) Before they take up their activity, the Processor commits all persons who have access to the data to confidentiality in writing, unless they are already subject to an appropriate statutory obligation of confidentiality. The obligation continues after the activity has ended.

(2) The group of these persons is limited to the extent necessary for performing the contract.

§ 6 Technical and organisational measures

(1) The Processor takes the measures described in Annex 2 under Art. 32 GDPR and maintains them throughout the entire term of the agreement.

(2) It may develop the measures further. The level of protection achieved in doing so may not fall below that described in Annex 2. Significant changes are documented and communicated to the Controller on request.

§ 7 Sub-processors

(1) The Controller grants its general authorisation to engage the sub-processors named in Annex 1 (Art. 28(2), second sentence, GDPR).

(2) The Processor notifies any intended change, in particular the addition of a further sub-processor or the replacement of an existing one, at least 30 days in advance in text form.

(3) The Controller may object to the change within 30 days of receipt for an important reason under data protection law. If no agreement is reached, the Controller has a special right to terminate the Main Contract.

(4) The Processor selects sub-processors carefully and contractually binds them to obligations that correspond to those agreed here. It is liable for their conduct as for its own.

(5) AI-supported functions and AI advisor.

a) The Controller can switch off the AI-supported functions, including the AI advisor, completely for its organisation. Once this has been done, no content is transmitted to an AI service provider. The AI advisor is available only if the Processor has enabled it for the organisation.

b) Content is transmitted to the AI service provider only when triggered by a user: for the individual AI functions, by expressly triggering them; for the AI advisor, by a request from the user or by a file that the user gives it for import. No automatic analysis takes place, for instance in the background, on a schedule or without a request.

c) For the individual AI functions, only the content selected in each case is transmitted; for a PDF file, this is the file itself. For the AI advisor, the following are transmitted: the request, the previous course of the conversation, the page of the platform on which the request is made, together with the title of an entry opened there, the name of the organisation, the user's first name and role, content of the Controller that the AI advisor retrieves to answer the request, and files that the user gives it for import. It retrieves only content that the requesting user is allowed to read in the platform themselves, and only to the extent that answering the request requires; each retrieval is technically limited in scope (Annex 2, section 5). The data set as a whole is not transmitted.

d) The retrieved content may include information about employees and other persons named in entries of the Controller, including competences, certificates, project involvements and training assignments of individual persons, insofar as the user is allowed to see them in the platform. If the user has read permission for personnel management, this also includes the following information on persons in the organisation: name, department, company, position and function, line manager, whether active or having left, the designation as external or as auditor, and the email address and telephone number stored in personnel management. Absences, personnel numbers, notes in personnel management, notes on certificates of individual persons and answers from surveys are not retrieved.

e) The AI advisor only suggests changes to the data set. They are carried out only when the user confirms them, and only within the scope of the user's write permissions; this also applies to approvals and publications that the AI advisor prepares. The AI advisor does not reach accesses and invitations, roles and permissions, settings of the organisation, the AI settings, billing or the deletion of the organisation, not even as a suggestion. The platform records every accepted change without content in a log (user, type of entry, action, time), which is deleted together with the organisation; in the change history of an entry, the change appears as “über Conni” (via Conni).

f) The conversation histories of the AI advisor are stored together with its suggestions so that the user can continue a conversation. They are visible only to the user conducting the conversation; the platform provides no view for the Controller or the Processor. The user can delete any conversation. Otherwise, conversations are deleted 90 days after the last message, and immediately when the user's membership in the organisation ends and when the organisation is deleted. Files that a user gives the AI advisor for import are deleted when the import is completed, at the latest 30 days after it began together with the associated conversation, unless the user takes them over as a document or evidence. Conversation histories are not subject to return under § 12.

g) If a user sends feedback on an answer and expressly chooses to send the conversation with it, the Processor receives an extract with the texts of the requests and answers, not the retrieved content itself. Feedback is deleted after 12 months.

h) For each call to a language model, the Processor records, without content, which organisation and which user used which function to what extent. This serves billing, limiting the costs per organisation and protection against misuse. The Processor removes the link to the user after 24 months, and immediately when the user account is deleted.

i) The AI service provider is Vercel Inc. (Annex 1). It processes the requests via its AI Gateway with a fixed processing location in the European Union at its sub-processors Amazon Web Services (Amazon Bedrock) or Google Cloud (Vertex AI). Neither Vercel nor these sub-processors store the transmitted content beyond answering the request; the exception is a cache for recurring parts of a request, which lasts at most a few minutes. A request may pass through a Vercel region outside the European Union before being forwarded, and the providers of the language models may check requests for misuse outside the European Union; the basis named in Annex 1 applies to this.

(6) Services that the Processor uses as purely ancillary services, in particular telecommunications, postal services, maintenance and cleaning, are not deemed to be sub-processing. Here too, the Processor takes appropriate precautions to protect the data.

§ 8 Rights of data subjects

(1) Data subjects address the exercise of their rights to the Controller. The Controller is obliged to fulfil them.

(2) The Processor assists the Controller in this by appropriate technical and organisational measures, insofar as this is possible. For this purpose, the platform provides functions for exporting, rectifying and erasing the data.

(3) If a data subject contacts the Processor directly, the Processor forwards the request to the Controller without undue delay and does not answer it itself.

(4) If the assistance goes beyond the range of functions of the platform, the Controller remunerates the effort in accordance with § 11(4). The Processor notifies the expected effort in advance.

§ 9 Further obligations to assist

Taking into account the nature of the processing and the information available to it, the Processor assists the Controller in ensuring compliance with the obligations under

  • Art. 32 GDPR (security of processing),
  • Art. 33 and 34 GDPR (notification and communication of personal data breaches),
  • Art. 35 GDPR (data protection impact assessment) and
  • Art. 36 GDPR (prior consultation of the supervisory authority).

Assistance is provided primarily by making the existing documentation available, in particular Annex 2. The Controller remunerates any effort beyond this in accordance with § 11(4); this does not apply to the notification under § 10.

§ 10 Personal data breaches

(1) The Processor notifies the Controller of every personal data breach that has come to its knowledge without undue delay, where possible within 48 hours of becoming aware of it, in text form to the address designated by the Controller under § 13(2). If it has not designated one, the notification is sent to the email addresses of the persons who hold the Administrator role in its organisation and to the person who confirmed this agreement in the platform.

(2) Insofar as known, the notification contains: the nature of the breach, the categories and approximate number of data subjects and records concerned, the likely consequences, and the measures taken and proposed. Missing information is provided subsequently without undue delay.

(3) The notification to the supervisory authority under Art. 33 GDPR and the communication to the data subjects under Art. 34 GDPR are the responsibility of the Controller.

§ 11 Evidence and inspections

(1) On request, the Processor demonstrates compliance with the obligations under this agreement in text form, in particular by providing the current version of Annex 2, a self-assessment and existing certificates and audit reports of its sub-processors.

(2) The Controller may demand an inspection beyond this only if the evidence under paragraph 1 is demonstrably insufficient in the individual case and the Controller sets out the reasons for this in text form. The inspection is carried out by the Controller itself or by an auditor engaged by it who is bound to confidentiality and is not in competition with the Processor.

(3) The inspection takes place after notice in text form of at least 28 days, at most once per calendar year, primarily as a remote audit, otherwise during normal business hours and without disrupting operations, with a duration of at most one working day. It does not extend to data of other customers or to details of protective measures whose disclosure would jeopardise their effectiveness. After a personal data breach that affects data of the Controller, the notice period is five working days and the limitation on frequency does not apply. The powers of the supervisory authorities remain unaffected.

(4) The Controller bears the costs it incurs and remunerates the Processor's effort for inspections under paragraphs 2 and 3, and for services for which this agreement refers to this paragraph, at EUR 150 net per hour or part thereof. If an inspection reveals a material breach of this agreement by the Processor, the Processor bears its own effort for this inspection.

§ 12 Deletion and return after termination

(1) After termination of the Main Contract, the Processor deletes all data of the Controller or returns it, at the Controller's choice. The Controller communicates its choice within 30 days of the end of the contract. For a trial access not followed by a paid contract, the end of the contract is the expiry of the trial access.

(2) For a period of 30 days after the end of the contract, the Processor provides the data to the Controller in a common, machine-readable format upon the Controller's request in text form. If the Controller does not make a choice within this period, the data is deleted when it expires.

(3) Backup copies. Deletion in the production system takes place without undue delay. Data contained in routine backup copies is deleted when the respective backup cycle expires, at the latest 90 days after deletion in the production system; this also applies to any recycle bin or version history at the storage location of the backup. Until then, it remains subject to this agreement and is not restored, except to remedy a system failure.

(4) A statutory retention obligation remains unaffected. In this case, the data concerned is blocked and processed exclusively to fulfil that obligation.

(5) Deletion is confirmed in text form on request.

§ 13 Obligations of the Controller

(1) The Controller is responsible for the lawfulness of the processing and for safeguarding the rights of data subjects.

(2) It designates a point of contact for data protection matters in text form to kontakt@imscollect.de or in the platform, where the platform provides a field for this, and keeps its details up to date. Until then, § 10(1) sentence 2 applies.

(3) It manages the accesses of its users and their permissions on its own responsibility. It withdraws accesses without undue delay when the authorisation ceases.

(4) It ensures that no data under Art. 9 GDPR is entered as long as there is no separate agreement under § 2(4).

§ 14 Liability

(1) In relation to data subjects, Art. 82 GDPR applies. Under it, the parties are jointly and severally liable insofar as they are responsible for the same damage. This liability can be neither excluded nor limited by agreement between the parties; the liability provisions of the Main Contract do not apply in this respect.

(2) In the relationship between the parties, each party bears the share of the damage that corresponds to its part of the responsibility (Art. 82(5) GDPR). The decisive factor is whose breach of duty caused the damage.

(3) If one party has paid full compensation for the damage, it may claim back from the other party the share that corresponds to the other party's part of the responsibility. The party against which a claim is made informs the other party without undue delay and gives it the opportunity to comment before it acknowledges or settles a claim.

(4) The Processor is liable to the Controller, subject to paragraph 6, for damage that arises because the Processor

a) has not complied with its obligations under this agreement or under the Regulation, insofar as these are specifically directed to processors, or b) has acted outside or contrary to lawful instructions of the Controller.

It is not liable for damage resulting from instructions of the Controller unless it failed to point out their unlawfulness under § 3(3) although the unlawfulness was obvious.

(5) Fines are borne by the party on which they are imposed. Compensation between the parties takes place only insofar as the fine is demonstrably based on a breach of duty by the other party.

(6) Claims of the parties against each other are governed by the liability provisions of the Main Contract, in the case of partner access including the liability provision of the Partner Access Addendum. Liability for intent and gross negligence, for injury to life, body and health, and under the Produkthaftungsgesetz (German Product Liability Act) remains unaffected.

§ 15 Final provisions

(1) Amendments and additions require text form. This also applies to any waiver of this form requirement.

(2) If a provision is invalid, the remainder of the agreement remains valid. The parties replace the invalid provision with a valid one that comes closest to its economic purpose.

(3) The law of the Federal Republic of Germany applies. The exclusive place of jurisdiction is Münster, provided the Controller is a merchant (Kaufmann), a legal entity under public law or a special fund under public law.

Annex 1: Sub-processors

As of 17 November 2026 (backup of the database outside the database provider: Amazon Web Services EMEA SARL for the nightly backup run, Microsoft also for the backup copy of the database). Changes are notified in accordance with § 7(2).

CompanyServicePlace of processingBasis where a third country is involved
Supabase Pte. Ltd., 65 Chulia Street, OCBC Centre, SingaporeDatabase, authentication, file storageIreland (AWS eu-west-1)Registered office outside the EU. Standard contractual clauses. Data stored exclusively in Ireland
Vercel Inc., 440 N Barranca Ave, Covina, CA, USADelivery of the website and the application. Processing of requests to the AI-supported functions and the AI advisor via the AI Gateway, only when the function is activated and when triggered by a user, without storage beyond answering the request (§ 7(5)(i)); the sub-processors of Vercel Inc. for this are Amazon Web Services (Amazon Bedrock) and Google Cloud (Vertex AI)Frankfurt am Main (region fra1); AI processing in data centres in the European UnionEU-US Data Privacy Framework, supplemented by standard contractual clauses; also for the forwarding of AI requests via Vercel regions outside the EU and for the misuse checks by the providers of the language models
Brevo GmbH, BerlinSending of system messagesFrance and GermanyNo third-country involvement
Microsoft Ireland Operations Limited, Dublin, IrelandNightly backup copy of the files stored in the application and of the database (SharePoint, the database encrypted); copies of deleted files are permanently removed after 30 days, invoices after the statutory retention period has expired, daily database exports after 30 days, weekly ones after twelve weeksGermanyStorage in Germany. Parent company in the USA: EU-US Data Privacy Framework and standard contractual clauses
Amazon Web Services EMEA SARL, 38 Avenue John F. Kennedy, LuxembourgNightly backup run of the database: export, encryption, hand-over to the backup copy at Microsoft; the exports are not retainedSweden (AWS eu-north-1, Stockholm)Processing in the EU. Parent company in the USA: EU-US Data Privacy Framework and standard contractual clauses

Annex 2: Technical and organisational measures

Annex 2 “Technical and organisational measures” in the version of 17 November 2026, which is attached to this agreement, applies.

Technical and organisational measures

Annex 2 to the data processing agreement.

As of17 November 2026
Earlier editions14 October 2026; 12 November 2026 with AI via the Vercel AI Gateway in the EU and the AI advisor. New on 17 November 2026: backup of the database outside the database provider in sections 1.1, 3 and 8, permanent removal from the backups in section 3
Applies toThe IMScollect platform including all sub-processors

1. Confidentiality

1.1 Physical access control (physical access to the facilities)

Processing takes place exclusively in data centres of the sub-processors. No servers of our own are operated.

  • Supabase (database, authentication, files): AWS region eu-west-1, Ireland. Physical access control according to the standards of Amazon Web Services (including ISO 27001, SOC 2 Type II).
  • Vercel (delivery of the application): server functions in Frankfurt am Main. The AI-supported functions and the AI advisor run via the Vercel AI Gateway, only when the function is activated and when triggered by a user; computation takes place in data centres in the EU at the sub-processors of Vercel, Amazon Web Services (Amazon Bedrock) or Google Cloud (Vertex AI).
  • Microsoft (nightly backup copy of the files and of the database): SharePoint, storage location Germany.
  • Amazon Web Services (nightly backup run of the database, without retaining the exports): AWS region eu-north-1, Stockholm.
  • Brevo (sending of system messages): data centres in France and Germany.

For the operator's workstations:

  • Full disk encryption
  • Automatic screen lock
  • No third-party access to the device
  • Access credentials exclusively in a password manager, not in files or in the source code

1.2 System access control (who may log in)

  • Login via Supabase Auth with email and password. Passwords are stored exclusively as a hash value, never in plain text.
  • No open self-registration. An account is created only through an invitation by an administrator of the client, through an invitation link generated by such an administrator, or when the organisation is set up by the Processor. As a result, there is no anonymous route into the application.
  • Two-factor authentication via TOTP is available to every person and can be made mandatory by the client for all members of its organisation. When the requirement is activated, a block in the middleware enforces it: accounts without a confirmed factor do not get into the application, and sessions without a second factor are directed to confirmation.
  • Time-limited accesses are possible. The time limit takes effect not only in the interface but in the database itself: the function that checks membership takes it into account, and an expired access no longer reads a single row.
  • If a person is recorded as having left in personnel management, their access is blocked automatically.
  • Passwords have at least eight characters and, when they are set, are checked against known data leaks; a password known from such leaks is rejected.
  • A login ends after seven days without use.
  • Limitation of login attempts by the authentication service, blocking after repeated failed attempts at the second factor, and logging of repelled access attempts.

1.3 Data access control (who may see and change which data)

  • Roles with individually configurable permissions. Separately per module and per direction (read / write), via 33 permission keys. Five system roles are predefined (administrator, management system officer, editor, reader, external auditor); the client can freely create and tailor its own roles.
  • Checks in two places, not in one. Every write operation checks the permission on the server before it reaches the database. Independently of this, the database itself checks membership and module permission again via row-level policies on every read and write operation. An error in the application therefore does not open up any path to the data.
  • The database does not accept write operations from outside. The public roles anon and authenticated have no write permission in the public schema, not even on tables created in the future. Writing is possible exclusively via the checked route on the server.
  • The AI advisor inherits the permissions of the person asking.It reads via that person's database connection, so that row-level policies, distribution lists and module permissions apply without a second check that could diverge. What it suggests is written only by the action of the module after the person's click, with that person's permissions. Details in section 5.

1.4 Separation control (separation of the clients from one another)

  • Each client is a separate organisation with its own identifier. Every record bears this identifier.
  • Separation is enforced in the database by row-level policies, not in the application. Even if a foreign identifier is known, no foreign row is delivered or changed.
  • Every modifying and deleting access is additionally restricted to the organisation. Knowing the identifier of a record alone is never sufficient.
  • An automated test checks on every run that no new table is created without being assigned to a module permission.

2. Integrity

2.1 Transfer control

  • All transmission is encrypted via TLS. Unencrypted connections are not accepted.
  • Data at rest is stored in encrypted form according to the providers (Supabase on AWS, Vercel).
  • Files are kept in the file storage of Supabase. They get there only via a one-time link that the server issues after checking permissions; the server then checks the file. Reading in the file storage requires the same module permission as the associated entry.
  • Security-related response headers (including HSTS, Content Security Policy, framing protection) are set.
  • The data export of a tenant at the end of the contract (in the operator area) contains signed links to the tenant's files and invoices, which are valid for seven days; anyone who has the export file can retrieve them during this time without logging in. The export is therefore handed over only to the customer itself.

2.2 Input control

  • Administrative actions are logged immutably. Every action with operator rights is recorded with the time, the acting account and the organisation concerned. The log table does not permit changes or deletions, not even via the technical service key. This block is anchored in the database, not in the application.
  • Substantive changes to content subject to approval (documents, processes, wiki, competence records) are versioned. The earlier version remains traceable.
  • Repelled access attempts and suspicious login attempts are recorded separately.

3. Availability and resilience

  • Operation on infrastructure with redundancy and failover at the providers.
  • Protection against overload and automated attacks via the upstream firewall of Vercel.
  • Separation of development and production environments.

Data backup

  • Daily backup of the database by the database provider, going back seven days.
  • Daily backup of the files stored in the application (documents, evidence, invoices) outside the database provider: at Microsoft 365 in a separate SharePoint site, storage location Germany, with the application having access only to this site. Copies of deleted files are permanently removed after 30 days, invoices after the ten-year retention period has expired.
  • Daily backup of the database outside the infrastructure of the database provider.A nightly run (AWS, Stockholm) takes a complete export using a role that may only read, encrypts it with a separate key before it leaves the machine (AES-256; the private key is not held at Microsoft but only in the key store of the backup run and in the operator's password manager) and stores it in the same SharePoint site as the files. Daily exports are permanently removed after 30 days, the Sunday export after twelve weeks. A backup held exclusively with the same provider protects against data loss, but not against failure or blocking of that provider.
  • Permanently removed means: without a recycle bin. Whatever drops out of the backup does not go to the recycle bin of the SharePoint site but is deleted immediately and irrevocably. Data deleted in the platform is therefore also gone from all backups no later than 90 days afterwards; invoices with their statutory retention are excepted.
  • Monitoring.If an export is missing for more than 36 hours or reports an error, the operator's operations overview shows this the same morning.
  • Restoration is tested, not assumed. Every month an automatic run restores the latest export into a separate database that is discarded afterwards and compares the row count of every table with the state at the time of the export. At least once a year and after significant changes to the database structure, the application is additionally started on a restored backup and its functionality is verified, using the key from the password manager. The result is recorded with the date.
MetricCommitment
Maximum tolerated data lossup to 24 hours (daily backup of database and files)
Target recovery timeone working day
Retention of backupsdatabase at the database provider seven days, outside daily exports 30 days and weekly ones twelve weeks; files continuously, copies of deleted files 30 days, invoices ten years
Last tested restorationFiles: 3 October 2026, all files read from the backup and compared with the original, one restored. Database: 17 November 2026, export of the production database restored into a separate database, all 328 tables with the same row count; monthly and automatic thereafter

4. Regular testing and evaluation

  • Automated guards in the build process. Before every deployment, tests run that check, among other things: that no table is created without an assignment of permissions, that the public roles have no write permission, that every list of values matches in the database and in the code, and that every write operation with a state condition evaluates its result.
  • Separate test suite against the real database for permissions and row-level policies.
  • Checking of dependencies for known vulnerabilities.
  • Recurring review of the security advisories of the database provider.
  • Review of this document at least annually and whenever there is a change to service providers, processing locations or protective measures.

5. Data minimisation and pseudonymisation

  • The record of cookie consent stores the IP address exclusively as a hash value with a salt that changes daily and a secret key. This makes it impossible to infer the address.
  • The AI-supported functions and the AI advisor can be switched off completely by the client for its organisation. When the function is switched off, no content leaves the platform towards an AI service.
  • AI analyses never run automatically in the background, but only when triggered by a user: for the individual AI functions via their button, for the AI advisor via a request or a file that the user gives it for import. For the individual functions, only the selected content is transmitted; for a PDF file, the file itself. A file uploaded for a process draft is not stored and is deleted after processing.
  • Processing location EU, technically fixed.Every request to a language model goes via the Vercel AI Gateway with the setting “region EU, only Amazon Bedrock and Google Vertex AI, without storage”; there is no fallback region outside the EU. Every response must report the processing location EU, otherwise it is treated as an error. Known limits: a request may pass through a Vercel region outside the EU before being forwarded, and the providers of the language models may check requests for misuse outside the EU.
  • Technical block against processing outside the EU. If the configured AI service does not process in the EU, the AI advisor does not accept any request for organisations of customers; this is checked on enabling and with every message.
  • Retrievals by the AI advisor are limited: at most 50 entries per query; texts shortened to 300 characters in lists, 2,000 characters in an individual entry and 6,000 characters for the text of a document; at most 30,000 characters per retrieval and at most 12 retrieval steps per request. Texts are never shortened silently. The platform itself calculates metrics across entire areas; only the figures and a limited number of anomalies go to the AI service.
  • Not transmitted to the AI service are absences, personnel numbers, notes in personnel management, notes on certificates of individual persons and answers from surveys. The AI advisor reads competences, certificates, project involvements and training assignments of individual persons only as the competence matrix shows them to the person asking. It reads personal details from personnel management (name, department, company, position and function, line manager, status, designation as external or as auditor, the stored email address and telephone number) only with read permission for personnel management. It is instructed to state only facts about persons and not to give any assessment of performance, suitability or reliability; a fixed test set measures the rules of the AI advisor before every new expansion stage and before every change of model or provider.
  • Exception to reading via the person's connection: two visibility checks for process drafts that have not been approved and for the risks attached to them, which the database does not map as a row-level policy. They return only identifiers and serve exclusively for hiding.
  • Conversation histories are visible only to the person conducting the conversation.There is no view for the administration of the organisation or for the operator. The tables are blocked for the login roles of the database; reading takes place only via the server, filtered by organisation and person. A nightly run deletes conversations 90 days after the last message; if the person's membership ends, they are deleted immediately.
  • Files of an import are kept in a separate, private storage that is included neither in the data export nor in the backup copy. They are deleted when the import is completed, at the latest 30 days after it began by a nightly run, together with the conversation and the suggestions.
  • Measurement without content. For each call to a language model, a measurement line is created with organisation, person, function, model, extent and costs. The link to the person is removed after 24 months, and immediately when the account is deleted. Measurement lines never contain text from a conversation.
  • The AI service does not store content beyond answering the request; a cache for recurring parts of a request holds it for at most a few minutes. The content is not used to train models.

6. Job control

  • Sub-processors are named in the data processing agreement (Annex 1) and published in the privacy policy.
  • A contract under Art. 28 GDPR is in place with each sub-processor.
  • Any change of sub-processor or new engagement is notified to the client in advance; the client has a right to object.
  • Processing locations are verified on the system itself, not taken from the documentation of the providers.

7. Handling of personal data breaches

  • Detection via the security logging and the error overview in the operator area.
  • Notification to the client concerned without undue delay, where possible within 48 hours of becoming aware, with the information under Art. 33(3) GDPR, so that the client can meet its own notification deadline of 72 hours.
  • The notification to the supervisory authority is the responsibility of the client as controller. IMScollect assists the client in this.

8. Expressly not implemented

These points are deliberately left open or not provided for.

ItemStatusAssessment
Content Security Policy with a nonce for scriptsNot active. A first attempt blocked all scriptsThe other security headers are set. Resumption planned
Encryption of individual data fields beyond the database encryptionNot implementedNot currently necessary in view of the types of data processed. To be reassessed if health data is included
Separate permission for the AI advisor per roleNot implemented. The AI advisor is available to every member of an enabled organisation with that member's own permissions, including external consultants and auditorsSwitching off is only possible for the entire organisation (AI switch). A separate permission key is planned
Certification to ISO 27001 or SOC 2 attestationNot availableNot planned for the current scope. The data centres used are certified

The general terms and conditions also apply.

Earlier versions: As of 14 November 2026 (in German), As of 14 October 2026 (in German)