AI in the finance team: how, what, and when?

You’re two months into a search for a finance hire. Maybe it’s a Financial Controller, maybe it’s your first FD. The candidates come through, the CVs look sound, and then somewhere around interview two, you ask the question that’s on every CEO’s lips right now: “How are you using AI, and have you ever created an AI agent?”

It’s become the reflex response to every finance hire. Why pay £70k, £90k, £120k for a role that isn’t paying for itself through automation, and more accurate, timely and relevant financial information? You’re not wrong to ask. AI can do a great deal in a finance function, more than most finance teams are currently using it for. But “can AI do this?” is the wrong first question. The right one is: is the finance team ready to explore, adopt and execute it?

The naive first move

The obvious next step feels simple. Pick a tool, say you’re adopting AI, and ask a select few if they’re using it with guardrails around the data. Your team say they’re using it, it’s amazing — but all they’re doing is research and spell-checking emails. The starting point is to look at your data strategy and every process map, to understand what you’re doing versus what you could be doing differently.

Data should be central to your strategy

Before a single spreadsheet or ledger entry goes anywhere near an AI tool, you need to know where your data lives, who can see it, and what happens to it once it leaves your system. Finance data is some of the most sensitive information in your business: customer pricing, payroll, banking details, supplier terms. Feed it into the wrong tool, on the wrong terms, and you’ve created a data protection problem that dwarfs whatever time you were trying to save.

Get a competent CIO or CTO involved first. Internal or external, it doesn’t matter which, but someone whose job is to interrogate where your data goes needs to sign off before your finance team starts experimenting. Once your data is properly ring-fenced — access controls in place, storage and processing terms understood, encryption in play if there’s any doubt — you can start testing. Not before.

Nobody’s actually mapped what finance does

Most finance teams can’t tell you, in writing, exactly what happens between an invoice landing and it being paid, or between month-end starting and the board pack going out. It’s tribal knowledge, held by whoever’s been there longest, done slightly differently depending on who’s covering for whom. You can’t apply AI sensibly to a process nobody’s documented. Skip this step and you’ll automate the wrong thing, or automate a broken process and just make the mess happen faster.

Not every process is worth automating

Once you can see the whole map, the next question is where AI actually earns its keep. Some processes are ripe for it: high volume, repetitive, rules-based, where speed and accuracy genuinely move the needle. Others aren’t: judgement-heavy work, exceptions-driven reconciliations, anything where the commercial context matters more than the mechanics. Diagnosing this properly, process by process, is where most businesses skip a step and end up disappointed with a tool that was never going to solve the problem they had.

Implementation stresses your team more than you’d expect

Say you’ve mapped it, diagnosed it, and picked your first two or three candidates for automation. Start small — a sandbox, a single process, tightly scoped — before scaling anything. What catches CEOs out is the transition period itself. For weeks, sometimes months, your team is running the old process and the new one in parallel, checking one against the other, while still doing their day job. That overlap is exhausting, and it’s exactly when things get missed. Bringing in a temporary project accountant to manage that changeover is often the difference between a smooth rollout and a team that quietly resents the whole initiative.