Where does our data go, and who can see it?
You're handing over tender pricing, client lists, supplier rates, drawings, maybe staff records. Before any of that leaves your building, you need to know where it lands. Ask whether your data is used to train the vendor's models, whether it sits on UK or EU servers, and who at the vendor's end can read it.
A straight answer comes back in plain terms: stored here, encrypted, not used for training, deleted on request. A vague answer, or one that points you to a forty-page policy, tells you they haven't thought it through for a business like yours.
Who owns the outputs?
If the tool drafts your quotes, your reports, or your client emails, you want it in writing that the output belongs to you and not to the vendor. Same goes for anything you build inside their system: templates, prompts, workflows you've spent weeks refining.
The question to put to them is simple. If we leave in two years, do we walk away with our work, or does it stay locked in your platform? The answer shapes how much you should depend on the tool.
What does this replace or speed up, in our workflow?
Not in the demo. In yours. Push the vendor to name the exact task this tool takes off someone's plate. Is it the eight days your estimator spends pulling a tender together? The hours your office manager loses chasing supplier quotes? The report your PM writes every Friday night?
If they can't point to a specific job that gets faster or cheaper in your business, you're buying a capability, not a result. Capabilities sit unused. Results show up on the P and L.
Who on our team has to run it, and how long to learn?
Software doesn't run itself. Someone owns it, feeds it, fixes it when it stalls. Ask who that person is in your business, how many hours a week it takes, and how long before they're competent rather than just trained.
The honest vendors give you a real number: a fortnight to get going, a month or two before it's second nature. The ones who say it's "intuitive" and "needs no training" are the ones whose tools end up abandoned by month three, because nobody was ever responsible for them.
What happens when it gets something wrong?
It will get things wrong. AI tools produce confident answers that are sometimes plain incorrect. The question is what your business does when that happens. Does a human check the output before it reaches a client? Can you trace why it made a given decision? Who carries the liability if a wrong number goes out on a quote?
A vendor who treats this seriously will show you the checks and the human sign-off points built into the workflow. A vendor who waves it away is leaving the cost of their mistakes on your desk.
How do we measure whether it paid back?
Agree the measure before you buy, not after. Pick one or two numbers you'll watch: days off the bid cycle, hours saved a week, error rate on invoices, jobs quoted per estimator. Write down where they sit today, and where the tool is meant to move them.
If you can't define what success looks like in advance, you'll never know whether the spend was worth it. You'll just have a subscription nobody wants to be the one to cancel.
What does it cost, fully loaded?
The licence fee is the easy part. The real cost includes setup, the data clean-up to get your information in a usable state, the staff hours to run it, and the time it pulls from the people learning it. A tool at £5–15k a year in licences can carry several times that in hidden time before it settles.
Ask the vendor to walk you through the first-year total, not the monthly headline. If they only ever talk about the per-seat price, do the fuller sum yourself. The tools worth buying still pay back once you've counted everything. The ones that don't only look cheap because half the cost was never named.