AI vendor due diligence: questions to ask before you sign

AI vendors for advice firms tend to demonstrate the output first: a polished suitability letter, a tidy meeting summary, a scored file. The output matters, but it is the last thing to judge. What sits behind it, where client data goes, who can use it, how mistakes are caught and how you leave, decides whether the tool is safe to adopt. That is the work of AI vendor due diligence, and it belongs before the contract, not after.

This checklist is vendor-neutral. Use it with any supplier, including us. At the end we give our own answers, so you can see what a direct response looks like.

Why AI vendor due diligence sits with the firm

The FCA has not created AI-specific rules. Its AI Update relies on the Consumer Duty, SM&CR and SYSC, and its June 2026 blog says it will "rely on existing frameworks, including the Consumer Duty, the Senior Managers and Certification Regime (SM&CR), and our expectations on governance and controls." In practice, a senior manager should be able to explain why the firm chose a tool and how it is controlled.

The ICO (since 30 September 2026, the Information Commission) is just as clear on the data side. Its guidance on AI and data protection says that "In the vast majority of cases, the use of AI will involve a type of processing likely to result in a high risk… and will therefore trigger the legal requirement for you to undertake a DPIA." Good vendor answers make that DPIA straightforward to write.

The AI vendor due diligence checklist

Put these questions to the vendor in writing and keep the answers with your DPIA.

1. Data flows and hosting

• Which organisations receive, process or store our client data? Ask for a full list, including AI model providers and transcription services.
• In which countries is data stored and processed? Is that a default or something we must configure?
• Is the vendor our processor, or a controller for any purpose? The ICO expects the firm to decide and document this for each party.
• Which steps, if any, run on our own computers rather than the vendor's service?
• Does the tool connect to other systems, such as our CRM, email or meeting recorder, and what data passes across each connection?

A good answer is a written data flow diagram or list that names every party. Be cautious of any answer that says data "never leaves" somewhere unless it is precise about which data and which step. We explain the two main designs in where client data goes with AI tools.

2. Model training and retention

• Is any of our data, including prompts, documents, transcripts and outputs, used to train any model, the vendor's or a third party's?
• Is that position contractual, or only a statement on a web page?
• How long is each type of data kept, and can we delete it ourselves?
• What happens to our data in backups after deletion?

Several vendors publish clear statements. Aveni's Trust & Security page says customer data is "never used to train our AI systems or those of any third-party AI provider" and that this is "enforced contractually", and Saturn's security page says "We never train our models on your client data". Ask for the contractual wording behind any such statement.

3. Security evidence

• Which certifications does the vendor hold, such as ISO 27001 or a SOC 2 report? Can we see the current certificate or report and its scope?
• Is data encrypted in transit and at rest?
• How are access credentials and API keys stored?
• How will we be told about a security incident, and how quickly?

Certifications are a good sign, but check that the scope covers the product you are buying.

4. Accuracy, evidence and supervision

• How does the tool show where each statement or figure came from?
• Does it tell us when it could not read a document or part of one?
• Can we adjust it to our firm's own standard, house style or scoring?
• What review step does the vendor expect before anything reaches a client?
• How are changes to the tool's rules or models dated and communicated?

This section matters for Consumer Duty. Under PRIN 2A.2.8R, "A firm must avoid causing foreseeable harm to retail customers." A tool that cannot show its sources makes that harder to evidence. For outputs used in competence or quality assessment, a dated, versioned standard is especially useful, so you can say which version applied to each assessment.

5. Contract, exit and continuity

• What is the minimum term and notice period?
• If we leave, how do we export our outputs, and in what formats?
• What happens to our data at the end of the contract, and is deletion confirmed in writing?
• If the vendor changes its AI provider, will we be told in advance?
• What support is included, and how is it reached?

6. Pricing

• Is the price published, or quoted per firm?
• Is it per user, per adviser, per firm or per item processed?
• Are AI usage costs included, or paid separately?
• Is there a trial, and what happens to data created during it?
• How much notice is given of price changes?

Published prices vary across the market. In our research in October 2026, AdvisoryAI publishes its prices on its pricing page, while we did not find published prices for Aveni, Saturn, PlannerPal, intelliflo IQ or Timeline on the pages we reviewed. An unpublished price is not a criticism; simply ask for a written quote that includes every cost.

A quick scoring sheet

Our own answers

For fairness, here is how IFA Tools answers the main questions:

• Data flows: our apps run on the firm's own Mac or Windows computers. The AI work goes through the firm's own Claude or ChatGPT account, so content the AI needs is sent to that provider under the firm's account. The meeting apps connect to Fireflies, and Carbon and Freeboard can connect Google. IFA Tools does not hold the firm's client data.
• On the machine: Scrutineer's text recognition for scans and its transcription of recordings in a client file run on the computer, with nothing uploaded for that step.
• Keys: stored in macOS Keychain or Windows Credential Manager.
• Training: governed by the firm's own agreement with its AI provider, which the firm should check and record. We explain this in why we use your own AI account.
• Evidence: for example, Freeboard scores against a versioned, dated standard and quotes the client's own words for every conclusion.
• Terms and price: no contract, cancel monthly. Published prices from £49 to £99 per adviser per month ex VAT, or £299 for all six. IFA Tools users also pay separately for their own Claude or ChatGPT account.

Frequently asked questions

Do we need a DPIA for every AI tool?

The ICO says AI use will, in the vast majority of cases, trigger the legal requirement for one. Plan for a DPIA for each tool that processes client personal data.

Is ISO 27001 or SOC 2 enough on its own?

It is a helpful signal, not a full answer. Check the scope, and still ask about data flows, training and exit.

Who in the firm should own AI vendor due diligence?

Usually the senior manager responsible for operations or compliance, with input from whoever handles data protection.

To put these questions to us directly, book a demo: twenty minutes on Teams with Adam Booth, on a demo firm, with nothing to prepare. Or read about Freeboard.