Choosing an AI workshop tool your data protection officer will approve

Written by the IdeaHone Editorial Team · · 6 min read

Golden combination padlock on a computer keyboard

The pilot was approved by the team in June and stopped by the data protection officer in September, over a question that could have been asked in May. It happens with most AI tools that listen to workshops: the team evaluates features first and data last. This checklist turns the order around: twelve questions to send to the vendor before anyone records anything, in the sequence your DPO will ask them.

Why the order matters

A DPO works from the data outwards: where it is, what it is, who sees it, how long it stays. A vendor demo works from the features inwards. Ask in the DPO's order and the first four answers tell you whether the rest are worth your time.

The twelve questions

  1. Where is the data hosted and processed, country and provider. Hosting and processing are two answers, and the second is where the surprises live. Good: country and provider named for both.
  2. Is raw audio stored, where and for how long. Audio is the most sensitive thing a workshop produces. Best: processed and discarded. Second best: a named location with a retention period you control.
  3. Which AI models process the content, where do they run, is it used for training. Ask for a written no on training. "We do not train on your data" belongs in the contract, not in a sales email.
  4. Can processing stay in the EU, and what happens to the features then. Many tools have an EU mode that quietly switches off half the product. Good: a list of what still works.
  5. Who can see what. Roles, private spaces, no cross-team or management views of individuals. Ask plainly whether a manager can see what one participant said in another team's workshop. Good: a plain no.
  6. What is biometric, how is consent collected, how is it withdrawn. Voice profiles that recognise speakers are biometric data. Good: consent per person before the first recording, withdrawable in one click, profile deleted on withdrawal.
  7. Retention: configurable, enforced automatically, including exports. "You can delete anything" is not a retention policy. Good: a period per organisation, deleted without a human remembering to, exported files included.
  8. Export and deletion of everything on request, including for one person. The organisation leaves and takes everything; one participant leaves and their contributions go too. Ask how long each takes.
  9. The data processing agreement, sub-processors listed, who signs. Ask for the DPA before the pilot. The sub-processor list is where you find the transcription service nobody mentioned.
  10. AI content labelling and human control. Every AI output is marked as AI and editable by a person; nothing about an idea or a person is decided automatically. The transparency rules of the EU AI Act point the same way; check with your legal team which obligations apply to you.
  11. Nothing that evaluates individuals. Talk time per person, participation scores, emotion detection: each one turns a workshop tool into a monitoring tool. Good: the feature does not exist, rather than it can be switched off.
  12. Security basics. Encryption at rest and in transit, passkeys or a second factor, an audit log, a security review you can read. Without one, your DPO asks IT, and IT says no.

Red flags and green flags

QuestionRed flagGreen flag
Hosting and processing"In the cloud, fully GDPR compliant"Country and provider named for both, in writing
Raw audio"Stored securely"Not stored; processed and discarded
Training"We take privacy seriously"A written no in the DPA
Voice profiles"Speaker recognition is automatic"Opt-in per person, revocable, deleted on withdrawal
DPA"We send it after the pilot"Sent now, sub-processors listed, signed before day one
Evaluating individuals"Engagement analytics included"The feature does not exist

The DPO's job is to say how, not no. Bring answers, not a vendor brochure, and the meeting becomes a review instead of an interrogation.

How to run the evaluation in two weeks

Days 1 to 2: send the twelve questions in writing to every candidate, with a deadline. Days 3 to 7: score each answer as clear, vague or missing. A missing answer on questions 1 to 4 ends the evaluation for that vendor.

Week two: run one real workshop, not a demo, with the two best candidates. Then give the DPO the scored answers, the DPA, and a one-page description in your own words of what the tool does with the data. Decide the same week. Some vendors answer several of these up front; IdeaHone, for example, is hosted in Germany, stores no audio, offers a local transcription model and an EU-only mode, and keeps voice profiles encrypted and consent-based. Ask for the rest in writing anyway.

Ask for the vendor's own privacy policy and read it section by section against what the sales call said. Where the two differ, the policy counts, and your DPO will find the difference anyway.

"We are GDPR compliant" is a claim, not an answer. Every one of the twelve questions has a concrete answer, or it does not.

Five evaluation shortcuts that backfire

  • Features before data. Three sessions comparing summaries, ten minutes on hosting. Reverse the time split.
  • The pilot that starts before the DPA. Workshop data is personal data from the first minute. No signed DPA, no recording.
  • Trusting the badge on the website. A compliance logo proves the vendor bought a logo. Ask the question behind it.
  • Forgetting the works council. A tool that records what employees say is a co-determination topic. Involve the works council in week one.
  • A tool that needs a US account for the facilitator. If the facilitator logs in through an account held outside the EU, the EU mode is decoration.

Capture every idea while people talk

IdeaHone listens in your session and turns what people say into idea cards, questions and decisions, live on a shared board.

Request a demo

What should a DPO check in an AI workshop tool?

Where data is hosted and processed, whether audio is stored, whether content trains models, who can see what, consent for voice profiles, retention and deletion, the DPA with sub-processors, and whether the tool evaluates individuals. The twelve questions above follow that order.

Is an AI transcription tool GDPR compliant?

Compliance is a property of how you use the tool, not a label on it. A tool can support it (EU processing, no training, consent, deletion), but your organisation stays responsible for the listening notice, the DPA and retention. See transcription and GDPR and check with your data protection officer.

What is a data processing agreement and do we need one?

A data processing agreement (DPA) binds a vendor to process personal data only on your instructions, lists its sub-processors and sets security and deletion duties. If a tool processes anything your participants say, you need one signed before the first recording.

All resources