What should you ask a board evaluation software provider about using questionnaire data for AI training?

Ask the provider whether your questionnaire responses, comments, and metadata are used to train any AI model, whose model it is, and whether you can opt out in writing. Board evaluation data is among the most sensitive information an organisation holds. Any permission to train on it must be explicit, limited, and revocable. The questions below show what to ask, how to test the answers, and how to weigh the trade-offs.

Why is board questionnaire data riskier than ordinary survey data?

Board questionnaire data is riskier because it contains candid views on named individuals, the Chair, strategy, and boardroom dynamics, often in free-text form. Small respondent groups make anonymity fragile, and the content may touch price-sensitive matters. Training an AI model on it multiplies the exposure beyond a single evaluation.

A board of nine directors yields a small dataset. A remark about the Chair’s handling of a contested acquisition can identify its author even without a name attached. Once that text enters a training set, the organisation loses practical control over where the information travels or whether it can be removed.

Evaluation data also carries regulatory and fiduciary weight. Directors answer frankly only when they trust the process. If they suspect their comments may be reused, candour declines, and the evaluation loses the honesty that makes it valuable.

What questions should you ask a provider about AI training on your data?

Ask whether customer data is used to train, fine-tune, or improve any model, and by whom. Then ask what exactly is used, whether training is opt-in or opt-out, and where the data is processed. Insist on answers in writing, not in a sales conversation.

A disciplined vendor due diligence list covers these points:

  • Is customer questionnaire data used to train or fine-tune any model, including third-party large language models?
  • Is training opt-in by default, or does your data enter the pipeline unless you object?
  • Which data types are covered: scores, free-text comments, respondent metadata, uploaded documents?
  • Which sub-processors and AI model providers receive your data, and in which jurisdictions?
  • Do those third parties retain prompts or outputs, and for how long?
  • Can your data be excluded from any benchmarking or cross-client analytics?
  • What happens if the vendor changes its AI approach after you sign?

Pay close attention to the distinction between using AI to analyse your evaluation and using your evaluation to improve the AI. The first is a service feature. The second is a transfer of value and risk from you to the vendor. Many board evaluation software providers blur the two in their marketing.

How can anonymisation and aggregation claims be tested?

Test anonymisation by asking the provider to explain the technique, the minimum group size below which results are suppressed, and how free-text comments are handled. Request evidence of independent verification. A claim that data is “anonymised” without a method is an assertion, not an assurance.

Several practical tests apply:

  • Minimum threshold. Ask what happens when fewer than five respondents answer a question. Reliable systems suppress or merge such results.
  • Free-text handling. Comments often contain names, events, and distinctive phrasing. Ask whether they are stripped before any secondary use, and who performs that step.
  • Re-identification risk. Ask whether anonymised data can be linked back to a client through industry, board size, or dates.
  • Aggregation scope. Clarify whether aggregation occurs across your own evaluation cycles or across all clients, and whether you consent to the latter.
  • Independent evidence. Request audit reports, penetration test summaries, or certifications, and read what they actually cover.

Ask for a demonstration with a sample dataset. A provider confident in its controls will show how a small board’s comments are processed. Hesitation is informative.

Which contract clauses should govern AI use of evaluation data?

The data processing agreement and master services agreement should prohibit training on customer data without written consent, restrict processing to your documented instructions, and bind every sub-processor to the same terms. Silence in the contract on AI use defaults to the vendor’s terms of service, which usually favour the vendor.

Company Secretaries should look for these provisions:

  1. Purpose limitation. Data may be used solely to deliver the contracted evaluation, not to develop or improve products or models.
  2. Explicit AI training prohibition. A clear statement that customer data, derived data, and outputs will not train, fine-tune, or benchmark any model without separate written consent.
  3. Sub-processor controls. A current list, advance notice of changes, and a right to object.
  4. Confidentiality. Obligations that extend to AI model providers and survive termination.
  5. Audit and transparency rights. The ability to verify compliance and receive notice of any incident promptly.
  6. Change control. Any new AI feature that alters data handling requires your approval before activation.

Involve legal counsel and the data protection officer early. Where the board evaluates directors across jurisdictions, check cross-border transfer terms as well.

What should happen to questionnaire data when the contract ends?

At termination, the provider should return or securely delete all questionnaire data, derived outputs, and backups within a defined period, and confirm deletion in writing. Deletion must extend to sub-processors and any model-related artefacts built from your data.

Define the mechanics before signing. Specify the retention period during the contract, the timeline for deletion afterwards, and the format for any data returned to you. Ask about backup cycles, since data often persists in backups for weeks. Ask whether embeddings, fine-tuned weights, or cached prompts derived from your responses are also removed, because those are the places where training exposure hides.

Request a deletion certificate. Boards also keep records for their own governance reasons, so agree in advance which reports you will retain internally and which the provider must destroy.

How should a Company Secretary weigh AI benefits against confidentiality risk?

Weigh AI benefits by asking whether the value comes from analysing your own data or from training on everyone’s. AI-assisted analysis of your evaluation can sharpen insight without any training on your data. Accept it only where confidentiality is contractually protected and the board has approved the approach.

The benefits are real. AI can surface patterns across committees, compare results across cycles, and draft recommendations faster than manual review. These capabilities serve the forward-looking purpose of an evaluation. They do not require your candid comments to become part of a shared model.

A sound approach follows four steps. Brief the Chair on the issue before selecting a provider. Take the AI data policy to the board or the relevant committee for explicit approval. Run the due diligence questions above and record the answers. Revisit the arrangement annually, since vendors change their models and terms. Treat the evaluation as a governance process first and a technology purchase second. Where the provider cannot give clear, contractual answers, the sensible position is to decline the AI features or choose another provider.

How The Board Practice helps with board evaluation data and AI

The Board Practice combines more than 19 years of board effectiveness methodology with an AI-powered platform launching on 1 August 2026, built by a firm that understands what boards must protect. Dr. Victor Prozesky has led or contributed to board performance programmes in more than 120 instances across industries and continents, so confidentiality is a working discipline, not a policy statement.

  • Tailored evaluations of the Board, Committees, and individual directors, designed in close partnership with the Chair
  • AI-powered analysis that turns questionnaire responses into actionable, forward-looking recommendations
  • Honest, unbiased feedback that identifies both strengths and development areas, with multi-year development plans
  • Guidance for Company Secretaries on vendor due diligence, data handling, and board approval of AI use

To explore how this works in practice, review our board effectiveness evaluation services or contact our team to discuss your board’s requirements.

Related Articles