There are over 4,000 AI tools on the market. Most are wrappers around the same handful of underlying models. Some are excellent. Almost all of them have a great demo video, an enthusiastic founder on LinkedIn, and a pricing page that gets more expensive the closer you look at it.
If you're a UK SMB, you don't have time to evaluate them all. Here's the buying playbook we recommend - the one that quietly separates the genuinely useful tools from the ones that win demos and lose deployments.
Ask for a real workload, not a demo
Bring your own data (suitably anonymised) and ask the vendor to run their tool on it live, with you in the room. Demos are choreographed. Real workloads aren't. You'll learn more from watching a vendor handle one of your messy real-world examples than from any number of polished case study slides.
If they can't or won't, that tells you something. Either the tool doesn't generalise as well as the demo suggests, or the sales team isn't trusted to operate it without a script. Both are problems.
Insist on a 14-day trial with named users
If the vendor won't agree to a trial, walk away. If they will, pick two or three named users from your team and time them on real tasks - before and after. Not 'how do they feel about the tool' - how long does the actual job take, and is the output good enough to send.
Two weeks is enough to get past the novelty effect. The first three days, everyone loves it because it's new. Day 14 tells you whether they still love it once it's just another tab.
Check the data exit story
Ask: if we leave in 12 months, what happens to our data, our prompts, and our custom workflows? A vague answer is your answer. The good vendors have a clear, documented offboarding process. The risky ones say 'oh, we can probably sort something out'.
Related: what does the vendor do with your data while you are a customer? Is it used to train their models? Can you turn that off? Where is it stored? These should be one-line answers, not white papers.
Watch the per-seat maths
AI tools love per-seat pricing. A tool at £25/user/month for 30 staff is £9,000 a year - more than most SMBs spend on accounting software. Make sure the value is genuinely there before rolling it out org-wide. A useful exercise: what would have to be true for this tool to pay for itself? If the answer involves heroic adoption assumptions, the answer is probably no.
Ask about volume discounts, but don't lead with them. The best discounts come after you've demonstrated commitment, not before.
Buy for one team first
Org-wide rollouts of unproven tools are how budgets disappear. Pick the team most likely to use it well, prove it, then expand. The team most likely to use it well is usually the smallest one with the most curious manager - not the biggest department with the loudest demand.
Set a calendar reminder for 90 days after purchase to check actual usage. Most AI tools have a usage dashboard. If your seat utilisation is below 40%, you've either rolled out too widely or chosen the wrong tool. Either way, fix it before the renewal.
Pricing red flags
- 'Get in touch for pricing' on a tool aimed at SMBs. Either it's enterprise-priced or they want to size you up first. Both make budgeting harder.
- Annual-only contracts on a brand new product. The vendor is hedging against churn. You should be hedging back.
- Sky-high overage fees for going over usage limits. Read the small print on tokens, requests, or 'AI credits'.
- Aggressive auto-renewal with short cancellation windows. Diary the cancellation date the day you sign.
Questions worth asking on the first call
- Who builds the underlying model, and what happens to my workflows if that model is deprecated?
- What's the most common reason a customer churns within the first year?
- Which other UK SMBs of our size are using this in production, and can we speak to one?
- What does your roadmap look like for the next two quarters, and how will I know if it slips?
- If we found a bug today, how would it get fixed, and how would we hear about it?
Vendors who answer those questions clearly are usually good to work with. Vendors who deflect are usually not.
After you've bought
The buying decision isn't the end of the work - it's the start. Pick an internal owner, define the success metric, agree the review date, and treat the first 90 days as a paid pilot. If it isn't working by then, it probably won't be working at the renewal either, and you've still got time to course-correct.
Buying AI tools well isn't about being suspicious of every vendor. It's about doing enough of the boring questions early that you don't have to do them later, when the budget is gone and the team has moved on.