Skip to main content

A finance seat falls vacant, and the clock starts…

The assistant hands in their notice, or the management accountant moves on, and within a day the old job spec is back out of the drawer, dusted off, and posted again. Same title. Same duties. Same salary band, give or take. The seat’s empty, so you fill the seat. It’s the most natural move in the world.

It might also be the wrong one.

I sat in a conversation last week with a business that’s grown from around fifteen people to nearly a hundred in a few short years. International now. A deliberately lean finance team, a full-time CFO, a fractional CFO a day or so a week, and one finance seat sitting open. And the question on the table wasn’t ‘who do we hire?’ It was the better one. ‘How much of this seat does a person still need to do?’

That’s a different question, and it changes everything that follows.

 

A role is a bundle, not a title

Here’s the thing a job title hides. A role isn’t one thing. It’s a bundle of tasks, tied together by a label that made sense when the bundle was assembled, often years ago.

Pull the bundle apart and you find the truth of it. The raw processing. The chasing and matching. The reconciliations. The first-draft management accounts. The variance numbers before anyone’s interpreted them. And then, threaded through all of it, the bits that were never really about the processing at all. Spotting the thing that doesn’t look right. Knowing which supplier’s always chaotic. Translating the number into something the rest of the business will act on.

For most of finance history those tasks travelled together, because they had to. The only way to get the judgement was to do the donkey work first. So we tied them into a single seat and gave it a title.

AI unties the bundle, task by task. The processing, the matching, the first drafts, a lot of that can now be handed to an agent. Which means the seat you’re about to refill was never really one job. It was several, wearing one name badge. And some of those jobs don’t need a person any more.

 

The bit nobody wants to skip

There’s a trap on the other side of this, and it’s worth naming before anyone gets carried away.

Untying the bundle does not mean handing the lot to a machine and walking off. The business this week wasn’t trying to delete the seat. They were trying to work out, carefully, which tasks an agent could take, with what controls, before a human was hired into what was left.

That word, controls, is the whole game. An agent can draft the journal, but someone builds it and someone signs it off. Maker and checker, the oldest control in finance, pointed at the machine. The output gets validated before anyone trusts it, the access is staged from read-only to anything riskier, and there’s a trail an auditor can follow. Get that wrong and you haven’t saved a salary. You’ve automated a mistake and given it a clean, audit-looking finish.

So the real work isn’t ‘can AI do this seat’. It’s slower and more useful. Untie the bundle. Decide task by task what a machine does well under proper control, and what still needs a human in the loop. Then, and only then, look at what’s left and ask what kind of person it adds up to.

 

What stays human

What’s left is rarely the same job you started with.

Strip out the processing and the first drafts, and the residue is the part that was always the actual point. The exceptions the agent flags but can’t resolve. The judgement when the data’s ambiguous and the model’s confident anyway. The translation between finance and a sales team that doesn’t believe the numbers. The trust that holds when the forecast wobbles in front of the board. And now, a new one. Owning the agents themselves, checking their work, knowing when they’ve quietly drifted.

That’s not an assistant’s job description from three years ago. It’s a more senior shape of work sitting inside a more junior seat, and if you refill the old spec you’ll hire for the wrong one. You’ll get someone brilliant at the tasks that are leaving, and untested on the ones that are staying.

This is why the hire changes, not just the headcount. The profile you want now leans less on ‘can run a clean ledger’ and more on ‘can challenge what the system produced and back their own read of it’. Real AI fluency, not the occasional query. Comfort with ambiguity, because the agent will be wrong sometimes and someone has to catch it. You can test for that in an interview, if you design for it. Less ‘talk me through your month-end’, more ‘here’s an output an agent produced, tell me what you’d challenge before you believed it’.

 

What this means in practice

Four things worth sitting with before you repost that spec…

One. Untie the bundle before you advertise. List the actual tasks in the seat, not the title. Mark which ones an agent could take under proper control, and which a human keeps. You’re hiring against what’s left, not against last year’s job description.

Two. Don’t automate without the checker. Every task you hand to an agent needs a maker who built it and a human who signs it off, with a trail behind it. The saving isn’t real until the control is.

Three. Hire for the human skill, not the old role. The work that stays is judgement, exceptions, translation and trust, plus owning the agents. Test for that directly. A confident output and the question ‘what would you challenge here?’ tells you more than any month-end walkthrough.

Four. Sometimes the answer is still a person, just a different one. Untying the bundle doesn’t always mean fewer hires. Often it means a better-aimed one. The seat you cut and the seat you need are rarely the same seat.

 

Where this leaves you

The next time a finance seat falls vacant, the reflex will be to refill it. Same spec, same title, out the door by Friday.

Try the slower question first. Not ‘who replaces them?’ but ‘what was this seat actually made of, and how much of it still needs a person?’

It takes an afternoon you won’t feel you have. It saves you from hiring, at some expense, for a job that’s disappearing underneath the title. The bundle’s already coming apart. Its probably better you untie it on purpose than discover, a year and a salary later, that the machine untied it for you.

• • •

This is the work itself, and it’s dominating our conversations with CFOs in scaling businesses right now. So, i’ve developed a service to tackle this problem head on. It’s called the Finance Team Blueprint: designing the finance team of tomorrow. A 1 hour session with me to map the future of your finance team.

You leave with a clear read on where your team’s time actually goes, what AI can take off you today, what has to stay human, and the first step worth taking.

If that question’s on your desk, drop me a note. For the CFOs and FDs tackling this right now, I’m happy to walk you through what we’re seeing. [email protected].