Reef TRHThe first question a managing partner asked me about AI was which product to buy. A colleague had demoed a tool at a conference, a competitor down the road was "already using AI", and her inbox held six vendor pitches. She wanted a recommendation.
I didn't have one — not because the tools are bad, some are genuinely good, but because the question skips the step that decides whether any of them pays off.
Firms that get real value from AI don't start by choosing a tool. They start by finding the one place in the month where the work is repetitive, high-volume and stubbornly manual, then ask whether AI can take it off a person's desk. That's a mapping problem, not a shopping problem — and it's the step almost everyone skips.
Why "which tool" is the wrong first question
A demo is built to impress. It runs on tidy data, on a workflow the vendor picked, in five clean minutes. Your firm runs on a receipt photographed at a bad angle and forwarded over WhatsApp, on the client who sends twelve emails and one spreadsheet, on a chart of accounts that has grown its own private logic over fifteen years. The distance between those two worlds is where most AI purchases quietly die.
Buy first and you learn too late that the tool solves a problem you don't have — or can't touch the messy version of the one you do. The budget's gone, the staff are trained, and the subscription sits unopened.
Start where the work already hurts
The better starting point is boring, and that's exactly the point. Look at your month and find the task that is done over and over, arrives in the same shape every time, and burns hours of a qualified person's day on something that isn't judgement.
For many firms that task is intake and month-end reporting. One firm we work with — a practice with a few dozen monthly-reporting clients — used to build each client's report by hand from whatever turned up: PDFs, receipt photos, spreadsheets, documents forwarded by email and WhatsApp. Now a pipeline reads it as it lands, pulls out what matters, and drafts a structured report per client in the firm's own format. A professional still signs off on every figure. The work didn't vanish; it changed from assembly into review.
What made it work wasn't cleverness. It was that the task was the most repetitive one on the desk, and the kind where a wrong answer is caught by a human before it ever reaches a client. That discipline — a person on the numbers that matter — is also what separates a pilot that ships from one that stalls after the applause.
A map is cheaper than a mistake
Mapping where AI is worth it sounds like a delay. It's the opposite. A week or two spent deciding which single workflow to automate — and which three to leave alone — costs a fraction of a tool bought, rolled out and abandoned.
Three things, before anyone builds: which task is worth automating first, what accuracy and sign-off it needs to be safe in a regulated profession, and what it would take to actually run it — not just demo it.
Half the value is the "leave alone" list. Learning that a workflow isn't ready — data too scattered, volume too thin, risk too high — saves you from finding it out the expensive way. That map is exactly what an AI roadmap for a firm is meant to produce.
So when the next vendor email lands, the sharp move isn't to book the demo. It's to open your own calendar, find the task your team dreads every month, and ask the smaller question: is this the one worth mapping first? Answer that, and the tool — if you even need one — picks itself.
Common questions
Where should an accounting firm start with AI? Start with the single most repetitive, high-volume manual task in your month — usually intake or month-end reporting — not with choosing a tool. Map where AI is genuinely worth it before you buy or build.
Do we need to clean up our data first? No. Scattered, messy intake is the normal starting point, and structuring it is part of the work, not a prerequisite you have to finish alone.
Will AI get our clients' numbers wrong? It can, which is why the workflows worth automating first are the ones where a professional reviews and signs off before anything reaches a client.
There will always be a newer tool, a louder competitor, another demo that dazzles for five minutes. None of that tells you where to start. Your own calendar does — the task nobody looks forward to, the one that repeats whether you like it or not. Begin there, and AI stops being a purchase you hope pays off and becomes a decision you can defend.