Reef TRHIt surfaces in the first meeting, every time, usually before anyone's touched the coffee: if we let AI into the firm, does it replace our people? The honest answer is no — not as reassurance, but as a plain description of what happens. What AI takes off the desk is the collation and the keying: receipts retyped by hand, figures copied between systems, the first draft of a report assembled line by line. None of that was ever where an accountant's value sat.
But "will it replace us" is the question firms open with, not the one that decides whether AI is worth having. The ones that decide it are quieter, and they tend to arrive once the first fear is out of the way. Here they are, with straight answers.
So what does it change, if not the headcount?
It moves your people off assembly and onto judgement. A firm we work with runs its monthly client reporting this way today, across dozens of clients: the documents still arrive as a mess of email, WhatsApp and photographed receipts, but an AI pipeline reads them as they land and drafts each client's report in the firm's own format. The team reviews instead of assembling. The hours that used to disappear into collation go into the work a client actually pays a professional for. That's the shape of it — AI doing the high-volume grind, people doing the thinking.
Is our clients' data kept separate?
In a multi-client practice this is the non-negotiable, and it's a design decision rather than a hope. Each client's data is scoped and access-controlled so one client's information can never surface in another's work, and every output is attributable back to the source it came from. A system that can't guarantee that separation has no business touching a book of clients — so it's built in from the first day, not bolted on after a scare.
What happens when it gets a number wrong?
It will, sometimes — so you build for that instead of pretending otherwise. Two things make it safe. Every figure is source-linked, traceable back to the document or transaction it came from, so nothing is a black box you have to trust blind. And a professional signs off where it counts: the AI drafts, a human decides.
It's the same rule that governed an AI support agent we built for an airline — anything touching money waited for a person before it went out. A confidently wrong number that reaches a client is the one outcome you cannot have, so the workflows you automate first are precisely the ones a human reviews before anything leaves the building.
Do we need to clean up our data before we start?
No — and this is the objection that stalls the most firms for no reason. Messy, scattered intake is the normal starting point; structuring it is part of the work, not a prerequisite you're expected to finish on your own first. If you wait until the data is tidy, you'll wait forever, because it never quite is.
Won't clients feel like they're dealing with a bot?
A fair worry, and the answer isn't to make the AI sound human for its own sake. It's that the client gets a faster, accurate, on-brand answer with a professional standing behind it. When that airline's support agent reached its premium cabins, the bar was exactly this — a passenger must never feel they're talking to a machine — and it cleared it because a human still owned everything that mattered. The client feels speed and accuracy; the firm keeps the relationship.
So where does a firm actually start?
Not by buying a tool. By mapping where AI is worth it in your specific month — and, just as important, where it isn't. That's a short, fixed-scope exercise, and it's the honest first step for a firm that has heard AI matters but doesn't want to gamble the budget on the loudest idea in the room.
Strip it back and the opening question was never really about headcount. It was about value — whether the thing that makes a firm worth hiring can be handed to software. It can't. What can be handed over is the part your people already resented: the typing, the chasing, the assembling. Give that away, and what's left is exactly the judgement the client was paying for all along.