Twenty-Four Minutes, One Timer: The Vetting Ritual Every UK Team Should Run Before Signing Up to an AI Tool
Photo: Wim Vandenbussche, CC BY-SA 3.0, via Wikimedia Commons
Vendor demos have a shape. Fifteen minutes of genuine usefulness, four minutes of pricing, one question at the end that sounds casual: 'shall I get you set up with a trial?' And because everyone in the room is tired and the thing did look quite good, someone says yes.
Eight months later the team is on an annual plan, two seats have become nine, and nobody can quite explain what the tool replaced. This isn't a failure of judgement so much as a missing step. Most teams have a process for buying a laptop and none at all for buying a workflow.
So here's a ritual. Twenty-four minutes, one timer, two people, one scorecard. It won't find you a perfect tool, because there isn't one. It will stop you buying a bad one by accident.
Why twenty-four minutes
Long enough to be real, short enough that nobody postpones it to next quarter. Book it like a meeting, because that's what it is.
Bring two people: whoever actually wants to use the tool, and whoever has to clear up the mess if it goes wrong — usually someone in operations, finance, or whoever owns the client data. Two perspectives, one clock. If the vendor's sales team wants to join, fine, but the timer runs regardless.
Minutes 0–4: Name the job, not the tool
Write one sentence in a shared doc: 'We want to use X to do Y for Z people, starting from this date.' If the sentence won't come, you're shopping rather than procuring, and shopping is what got you here.
Then name the current workaround. If the workaround is a twenty-minute job someone does on Fridays, you're buying convenience, not capability — which is a perfectly good reason, as long as you say so out loud. Convenience purchases should be judged on convenience, not on the vendor's slide about transformation.
Minutes 4–8: Map the data before anyone pastes anything
List the categories of data that would realistically end up in the tool: client names, contract terms, unpublished figures, HR notes, customer personal data. Be concrete. Then agree the red lines — what will never go in, regardless of how good the tool is. Special category data, anything covered by an NDA you haven't read, material you don't have the rights to process.
One more check with the clock running: what does the free tier say about training on your inputs? Free and consumer tiers often reserve rights that business tiers don't. Read the line rather than trusting the demo, and check that the tier you're actually going to buy has the opt-out you think it does.
Minutes 8–13: Test the paperwork against UK GDPR, not against vibes
Seven questions, each scored red, amber or green.
- Is there a data processing agreement you can download without a sales call? If it requires a negotiation, note it.
- Is the sub-processor list public, and are changes notified? 'Trust us' is not a sub-processor list.
- Where is the data stored? 'EU' is not 'UK'. Find the region selector and the actual hosting statement.
- Is there an international transfer mechanism named? Look for the IDTA or the UK Addendum to the EU standard clauses.
- How long is data retained, and can you change it? A fixed retention period is a design choice someone made on your behalf.
- Can you delete data, or only hide it? Test with dummy content, not client content.
- Does your use case need a DPIA? Large-scale profiling or monitoring usually does. If you're unsure, that's a conversation for whoever owns data protection, not a decision for the trial group.
Minutes 13–17: Find the exit while you still have leverage
Export first. During the trial, with dummy data, confirm you can get everything back out in a format you can actually use — CSV, JSON, Markdown, whatever survives being opened in six months. Watch for exports that only cover recent history, or that need a support ticket to unlock.
Then cancellation. Is it in-app, or an email to an address nobody monitors? Note the renewal date. Look for seat creep — the setting that quietly adds a licence whenever someone accepts an invite. Annual plans with a narrow cancellation window are the classic trap, and the trap is described in the terms, not on the pricing page.
Minutes 17–21: Follow the money at three sizes
Price the tool for 6 seats, 12 seats and 40 seats. Somewhere in that range the pricing model changes shape — flat rate becomes per seat, or per seat becomes per usage, and the third number is the one that ends the project. Write it down.
Ask one finance question: which budget line pays for this? If nobody can answer, that's an amber at best. If the pricing is in dollars, note the FX exposure and the VAT treatment — services from overseas suppliers usually land as a reverse charge, and someone in accounts will want to know before the first invoice, not after.
Minutes 21–24: The 90-day question
Write three things that would have to be true in 90 days for you to keep paying. 'Every account manager uses it weekly without being reminded.' 'We've stopped doing the Friday summary by hand.' 'The client-facing output needs no rework.'
Then check that the trial actually tests those things. A trial that exercises a feature nobody will use in week three produces a verdict nobody should trust. If you can't name three conditions, you don't have a trial — you have a free month.
The scorecard
| Check | Green | Amber | Red |
|---|---|---|---|
| Job and owner named | Clear sentence | Vague | Missing |
| Data categories mapped | Listed with red lines | Partial | Not discussed |
| DPA available | Downloadable | On request | Sales call only |
| Sub-processors | Public list | On request | Undisclosed |
| Data residency | UK or named region | Named country only | No statement |
| Export | Tested successfully | Promised | Not offered |
| Cancellation | In-app | Contract only | |
| 90-day conditions | Three written | Two | None |
House rules: any red on data or exit means no, however good the demo was. Three or more ambers means extend the trial by a week rather than sign. Everything green means you have a decision you can defend in a meeting.
After the timer stops
Write the result into a shared document — a paragraph is plenty. Who wanted it, what job it does, what it costs at twelve seats, when the renewal lands, and what would make you cancel. That note is worth more than the scorecard, because it's the thing you'll want when the invoice arrives and nobody remembers why.
Twenty-four minutes, two people, once per tool. Considerably cheaper than an annual plan nobody uses.
