Customer data and AI: the supplier questions a processing-mode label cannot answer
2026-08-06 · Reviewed 2026-11-04

Customer data and AI: the supplier questions a processing-mode label cannot answer

A processing-mode label is not a compliance conclusion. Before an AI service receives customer information, the buyer needs a documented purpose, a data path and accountable decisions. GDPR and the AI Act are distinct frameworks; choosing an EU region or excluding a model does not settle every obligation. This guide is a technical procurement worksheet, not legal advice or permission to process a particular dataset.

Start with the smallest necessary dataset

Describe the output before collecting inputs. If the task is to categorise a support request, ask whether names, full attachments and years of customer history are necessary. Use synthetic or appropriately redacted examples for discovery. Removing an email address alone does not establish anonymity: free text and combinations of details may still identify someone. A filtering demonstration should state what it misses, not promise perfect anonymisation.

The procurement team should obtain legal review of the actual roles and processing conditions. Record why each data category is needed and which organisation decides its purpose. A generic statement that the customer has consent is not a substitute for checking the appropriate basis and obligations. Sensitive data and consequential decisions require more scrutiny than an ordinary internal drafting example.

Ask for evidence along the data path

Supplier due diligence — evidence to request
AreaQuestionUseful artefact
DecisionCan the actual flow be explained and controlled?Written ownership and open issues
InputWhich fields and files leave our systems?Field-level data map
AccessWho can read inputs, outputs and logs?Roles and permission matrix
StorageWhere do copies persist and for how long?Retention and deletion procedure
SuppliersWhich services process or access the data?Current contractual/subprocessor information
ChangeWhat happens after a model or region changes?Review and release record

CNIL's linked recommendations concern AI development and training. They are useful for understanding development-specific questions, but must not be presented as a blanket authorisation for operational use of a third-party assistant. Keep development, testing and production flows separate in the inventory. Your production supplier terms and national advice must address the configuration you actually intend to buy.

Test the technical claims you can test

  • Try cross-organisation reads, writes and exports using controlled synthetic records; a company filter on the screen is not enough.
  • Check whether logs and failed-job payloads retain information that the main database is supposed to delete.
  • Verify that a revoked user or connection stops working, including background jobs and old links.
  • Treat retrieved documents and emails as untrusted content; they cannot grant permission to access unrelated data.
  • Define incident ownership, access removal and an exit procedure before the workflow becomes business-critical.

Be precise about the seller and the promise

LEMAIT serves European businesses through TRIQUETRA COMPANY W.L.L. in Bahrain, as disclosed on our company pages. Serving Europe is not a claim of EU incorporation or EU-only processing. The proposed hosting, support access, third-party services and transfer arrangements must be assessed for the individual engagement. We do not describe a provider list as an official EU approval list.

Our interface may expose a restricted processing configuration. Its useful meaning is narrower: selected routing and control settings that require verification. It does not establish that every personal identifier is removed, that data never leaves a region, or that the customer's purpose is lawful. Contractual terms, configuration and observed behaviour should agree; a badge is not evidence when they conflict.

A maintainable review assigns owners for access, retention, supplier changes and incident follow-up. Recheck after adding a connector or enabling a write action. A technical diagnostic can identify missing evidence and implement bounded controls; a legal adviser decides the legal questions. The free supplier brief is designed to make that conversation specific without collecting customer records in the initial enquiry.

Prepared with AI assistance and checked against the linked public sources by LEMAIT. Scenarios are illustrations, not customer results. No ranking, ROI or legal-compliance guarantee.

TRIQUETRA COMPANY W.L.L., Bahrain · LEMAIT · Privacy