Questions to Ask Before Buying an AI Tool

Most AI purchasing mistakes are not caused by choosing the “wrong” product — they are caused by not asking the right questions before signing. The questions below work for any AI purchase: a standalone assistant, an industry-specific AI product, or an AI feature bolted onto software you already pay for. Put them to the vendor in writing and keep the answers; they matter if something goes wrong later.

First: what problem is this solving?

Before any vendor conversation, be able to answer this yourself. “We should be using AI” is not a requirement. Write down the specific task the tool should improve, who will use it, and what a successful result looks like in numbers — hours saved, faster responses, fewer errors. If you cannot describe success, you cannot judge a trial, and you will end up renewing a subscription nobody can defend.

Data handling: where does our information go?

This is the question set that separates business-grade AI tools from consumer products with a business price tag. Ask the vendor:

• Is our data used to train your AI models, or anyone else's? The answer for a business tier should be a clear no, in writing — not a settings toggle staff have to remember.
• Where is our data stored and processed — in the UK, the EU, or elsewhere? If outside the UK, what safeguards cover the transfer?
• Who at your company can see our prompts, uploads and outputs, and under what circumstances?
• How long is our data retained, and can we delete it — including from backups?
• Which other companies (sub-processors) handle our data behind the scenes, and will you tell us when that list changes?
• What happens to our data if you are acquired or shut down?

Pricing: what will this really cost?

• Is pricing per user, per usage (credits, tokens, messages), or both? Usage-based pricing can make a cheap-looking tool expensive once the whole team uses it.
• What happens when we hit a usage limit — does the tool stop working, slow down, or silently bill us more?
• Is the advertised price an introductory or promotional rate? What is the standard price at renewal?
• Is the price in pounds or dollars? Dollar pricing means your cost moves with the exchange rate.
• Are the features in the demo included in the tier we are buying, or in a higher tier?
• Is there a minimum number of seats or a minimum contract term?
• How much notice will we get of price increases?

Then calculate the yearly cost for your actual headcount and realistic usage — not the per-user monthly figure. A £20-per-user tool across ten staff is £2,400 a year, before VAT and before any usage overages.

Lock-in: how hard is it to leave?

Every subscription is easy to enter and harder to exit. Before you build the tool into daily work, ask:

• Can we export our data — conversations, documents, prompts, custom configurations — in a usable format, and can we do it ourselves without asking support?
• If we build custom assistants, templates or workflows inside the tool, do they belong to us? Can they move anywhere else?
• What is the notice period for cancellation, and does the contract auto-renew?
• Does the tool only work inside one vendor's ecosystem, and are we comfortable deepening that dependency?
• If we stop paying, how long do we have to retrieve our data before it is deleted?

Support and reliability: who helps when it breaks?

• What support is included — email only, chat, phone — and what are the response times for the tier we are buying?
• Is there a published service-level commitment (uptime), and a status page we can check during an outage?
• How are changes communicated? AI products change models and features frequently — will we be told before behaviour changes?
• Is there a UK or European support presence, or do time zones mean a day's delay on every reply?
• How long has the company been operating, and is it likely to exist for the life of our contract? The AI market moves fast and products are discontinued regularly.

UK GDPR: your obligations do not transfer to the vendor

If the tool will touch personal data — customer details, staff information, email content, meeting recordings — you remain responsible for it under UK GDPR. Buying a tool never outsources that responsibility. Before real data goes in:

• Get a data processing agreement (DPA) from the vendor. Reputable business AI tools offer one as standard; if a vendor does not know what a DPA is, walk away.
• Check for international transfers: if data leaves the UK, the vendor should explain the legal safeguard that covers it (such as an adequacy decision or standard contractual clauses).
• Consider whether you need a data protection impact assessment (DPIA) — sensible whenever an AI tool will process personal data in a new way, and the ICO provides free guidance for small businesses.
• Be able to honour individual rights: if a customer asks what data you hold on them or asks for deletion, can you find and remove their data inside this tool?
• Tell people where you should: if AI is used in decisions that affect customers or staff, your privacy notice may need updating.

Before you commit: trial and verify

Once a vendor has answered these questions well, run a small paid trial — two or three licences for a month on real work — and judge it against the success measure you wrote down at the start. And whatever the tool, treat its output as a draft, not a fact. Before AI-generated content influences a business decision or reaches a customer, run it through a verification process:

AI Answer Verification Checklist →