Questions to ask an AI supplier before you sign anything
By Keith Hamilton · · 4 min read
The questions to ask an AI supplier before you sign matter more than the demo. AI demonstrations are designed to impress: the supplier picks the examples, the data is clean and everything works first time, and your business does not look like that. This article sets out the questions to put to any AI supplier, what a good answer sounds like, and a short checklist to take into the meeting.
It focuses on the commercial and practical side of choosing an AI supplier. For the data protection questions, see our UK GDPR and AI checklist.
What problem will this solve for us?
The supplier should describe what the tool does in terms of your process, not their product. Ask who does the task today, what will change for them, and what they will still do themselves.
A good answer is specific and modest. Be wary if the answer is mostly about features, or if it promises to transform everything at once.
If you have not yet worked out which task matters most, our guide on where to start with AI gives you a clear brief to take into the conversation.
Can we see it working on our own examples?
Yes, and you should insist on it, because a demo built on the supplier's examples tells you little. Ask to see the tool handle a sample of your real work: a handful of genuine customer emails, typical calls, or the invoices that cause your team trouble.
Remove or anonymise personal details first. Ask, too, for a short trial before any long commitment.
What happens when it gets something wrong?
Every AI tool makes mistakes, so the useful question is what happens next.
- How does the tool behave when it is unsure? Does it say so, ask for help or guess?
- When does it hand over to a person, and what information does that person receive?
- Can you see a record of what it did and why, so mistakes are found and fixed?
A good supplier talks about errors openly. Be cautious of one who says their tool does not make mistakes.
How much does AI software cost at our volumes?
AI pricing is typically a mix of a set-up fee, a monthly subscription, per-user charges and usage charges. Ask for all of them in writing, with an estimate at your normal volumes and at your busiest.
Usage charges are where bills move. At 20p per conversation, 1,500 conversations a month costs £300. If a busy season doubles the volume to 3,000 conversations, the same tool costs £600 that month.
Also ask:
- Is there a minimum contract term, and what is the notice period?
- How much notice will you give of price changes?
- What counts as usage, and what happens if we go over an allowance?
With the full picture, compare it with what the task costs today. Our article on the real cost of manual work shows the arithmetic.
What will it connect to, and who builds that?
The tool needs to connect to the systems you already use, because that is where most of the value from AI comes from: your accounts package, CRM, phone system or order system. Ask which of your systems it works with already, who builds any connections that do not exist, and who maintains them when either system changes.
Ask what access it needs, and limit it to what the task requires.
Where does our data go?
Your data goes wherever the supplier stores and processes it, so ask exactly where that is. Keep this short in the commercial meeting, but do ask it:
- Where is your data stored and processed?
- Will anything you put in be used to train the supplier's models?
- Which other companies does the supplier pass data to?
- Will they sign a data processing agreement?
Then follow up properly using the checklist linked above.
Who owns what we build, and can we leave?
You should own your work and be able to take it with you. Over time you put instructions, prompts, workflows and records into a tool, so ask who owns them and whether you can export them in a usable form.
Ask what happens to your data when the contract ends, how long it takes to get it back, and confirm that it will be deleted afterwards.
How will we know it is working?
You will know it is working by measuring against a baseline agreed before you start. Record how the task performs today, such as hours spent, response times or calls answered, and agree what the supplier will report afterwards.
Be precise about what each measure means. In our own work with an SME telecoms business, inbound calls answered rose from 23% to virtually 100%. That figure measures whether a call was picked up, not whether the caller's query was resolved by the AI, and any supplier should be just as clear about what their numbers do and do not show.
What support will we get?
Support should be named, timed and priced in writing. Ask who you contact when something goes wrong, how quickly they respond, and whether support is included in the price.
Ask how they handle changes to the underlying AI model, since a change made by the model provider alters how the tool behaves.
What is the short checklist of questions to ask an AI supplier?
These eight AI vendor questions cover what matters when buying AI software for a small business:
- What will change for the people who do this task today?
- Can we test it on our own examples first?
- What happens when it is unsure or wrong?
- What is the full cost at normal and busy volumes?
- Which systems will it connect to, and with what access?
- Where does our data go, and is it used for training?
- Can we export our work and leave cleanly?
- How will results be measured against today's baseline?
Where Forwardcycle fits
Our AI Governance Pack includes a vendor checklist you can reuse with every supplier, and our fractional Chief AI & Commercial Officer service covers vendor management on your behalf. If you are about to choose a supplier and want a second opinion, arrange a free call.