Before You Buy the AI Tool, Answer These Five Questions

A CFO kept getting the same board packet three quarters running: an article about AI copilots, forwarded with the note "should we be doing this?" She wanted a straight answer. The straight answer is five questions, and none of them are about the technology.

That's deliberate. The models mostly work now, and the demos are all impressive. Whether a tool earns its keep in your shop is a business question, and you can answer it before you sit through a single pitch.

The first two questions are about the work

Question one: what decision does this speed up? Not what task, what decision. A tool that digests fifty pages of claims correspondence so an adjuster can rule in an hour instead of a day is speeding up a decision someone is paid to make. A tool that drafts documents nobody was waiting on is a hobby with a subscription fee.

Question two: who checks the output? Somebody has to, and their time is the hidden line item in every AI business case. If the answer is "nobody, it goes straight out the door," you've designed a system where the model's worst day becomes your organization's worst day. If the answer is "a senior person reviews everything," fine, but then admit you've bought a drafting assistant and price the benefit like one.

The next two are about the bad day

Question three: what happens when it's wrong? When, not if. A tool that's right most of the time is still wrong on a schedule, and the cost of those wrong days varies enormously. A clumsy first draft of a memo costs you an edit. A wrong number in a regulatory filing costs you a hearing. Map the blast radius of a bad output before you buy, because the vendor's demo will not do it for you.

A tool that saves an hour and adds an hour of checking has bought you a story, not a saving.

Question four: where does the data live? Every prompt your team types is data leaving the building. Customer records, contract terms, the strategy deck that isn't final yet. Read the vendor's terms on retention and training use, then ask them directly whether your inputs improve their product. If they can't answer that in one plain sentence, that is your answer.

The last one is about money

Question five: what does success cost? Everyone budgets the licenses. Almost nobody budgets the review time from question two, the process redesign, the training, and the awkward quarter where productivity dips before it climbs. A public agency we worked with piloted a document tool that saved each analyst about an hour a day, then watched most of that hour go to verifying citations. The tool wasn't bad. The business case just hadn't met question five yet.

There's also the cost of stopping. Once a team reshapes its workflow around a tool, the switching cost climbs every month you use it. Negotiate exit terms and data export while you're still easy to walk away from, because you won't be for long.

Run all five questions against one real workflow, with the people who actually do the work in the room. Where the answers hold up, buy the tool and move fast. We've watched these things pay for themselves inside a quarter when the fit is right.

Where the answers wobble, wait. The tools will be better and cheaper next year, and your board will forgive you for being deliberate. It will not forgive you for being first and wrong.