The Questions to Ask Before You Trust an AI Tool With Client Data

A law firm we spoke to had already connected a popular AI transcription tool to their client intake calls before anyone asked where the recordings were stored. The tool was useful. Nobody had checked what happened to the data after the meeting ended.

This is not a story about a reckless business. It is the normal pattern. A team finds a tool that solves a real problem, signs up with a work email, and starts feeding it client information within a week. The excitement about what the tool can do usually arrives well before anyone asks what it does with what you give it.

You do not need a legal background to ask sensible questions before that happens. You need a short list of things to check, and the discipline to check them before the data starts flowing, not after.

Start with where the data actually goes

Every AI tool sits on top of infrastructure somewhere. That might be the vendor's own servers, a cloud provider like AWS or Google Cloud, or a subprocessor you have never heard of. Ask the vendor directly: where is client data stored, physically and legally, and does it ever leave that location.

This matters for two reasons. First, some industries and some clients have expectations about where their information sits, particularly in legal, healthcare, and financial services. Second, if data crosses borders, different data protection rules may apply, and you want to know that before a client asks you and you have no answer.

A vendor that cannot answer this question clearly, or answers with marketing language instead of specifics, has told you something useful. Move on, or ask again in writing.

Ask whether your data trains their model

This is the question most business owners forget to ask, and it is the one that matters most. Some AI tools use customer input to train or improve their underlying model. That means a client's information, or patterns drawn from it, could theoretically shape how the tool responds to other customers entirely unrelated to your business.

Reputable vendors will state clearly whether this happens and give you a way to opt out. Many enterprise-tier AI products already exclude customer data from model training by default. Free or consumer-tier versions of the same tool often do not carry that protection, which is one reason the free version and the paid version of the same AI product can carry very different risk profiles for a business handling client information.

Read the terms. If the answer is buried in a privacy policy you cannot find in five minutes, ask the vendor to point you to the exact clause.

Ask who has access, and how that access is controlled

Data protection is not only about the vendor's servers. It is about people. Which of the vendor's staff can see client data, under what circumstances, and is that access logged. Can your own team members see more than they should inside the tool itself. Does the tool support role-based permissions, so a junior team member cannot pull up every client's full record by default.

Internally, this is worth applying to your own team too. An AI tool that gives every staff member full access to every client file is a bigger risk than the tool itself, regardless of what the vendor promises on their end.

Ask what happens if the vendor is acquired or shuts down

Software companies get bought, merge, or fold. When that happens, what contractually happens to the data sitting inside their systems. Does client data transfer automatically to a new owner, does it get exported to you first, or is there a defined process for deletion.

Smaller or newer AI vendors, which describes a large share of the tools flooding the market right now, are more likely to be acquired or to close within a few years. That is not a reason to avoid new tools. It is a reason to ask the question before you sign up, while asking still costs you nothing.

Ask for a data processing agreement

A data processing agreement, often shortened to DPA, is a contract that sets out how a vendor is allowed to handle personal data on your behalf; what they can do with it, how they must protect it, and what they must do if something goes wrong. Most established AI vendors offer one, usually available on request or already published on their site for business-tier accounts.

If a vendor cannot produce a DPA, or treats the request as unusual, that tells you something about how seriously they take this part of the relationship. Ask for it in writing, before you connect the tool to anything containing client information.

Ask about retention and deletion

How long does the vendor keep client data after you stop using the tool, or after a specific record is no longer needed. Can you request deletion, and how long does that deletion actually take once requested. Does deletion mean removed from active systems only, or removed from backups too.

This question matters because a client can ask you, the business they trust with their information, to have their data removed. If you cannot answer what happens on the vendor's side, you cannot honestly answer the client either.

A note on legal and compliance advice

None of the above is legal advice, and this article should not be treated as a compliance checklist. In South Africa, the Protection of Personal Information Act, generally known as POPIA, sets out rules for how businesses must handle personal information, and equivalent frameworks exist in other jurisdictions such as the GDPR in the European Union. Requirements differ by country, by industry, and by the type of data involved.

If your business handles sensitive client information, whether medical records, financial details, or legal matters, get advice from a qualified legal or compliance professional in your own jurisdiction before adopting a new AI tool at scale. This article gives you the practical questions to ask a vendor. It does not replace proper legal counsel for your specific situation.

Where this fits into how we work

Data questions are part of the Assess stage of the Claro Build Framework, before a single tool gets connected to anything. We do not sell AI tools. We help service businesses fix the process first, choose the right tool for what the process actually needs, and make sure the team understands what that tool does with the information passing through it. Every build we complete also meets the Adoption Standard, our minimum handover requirement, so the people running the system day to day know exactly how it works, not just that it works.

If you want to see how this plays out in practice, our case studies cover real builds across different industries, and our AI explained simply pillar has more articles written for business owners rather than technical teams. If you are asking these questions because a specific tool is already on your desk, a short discovery call is the fastest way to work through them properly. We will tell you honestly if the tool is fit for purpose, and if it is not, we will tell you that too.