Who checks the robot? Simple AI rules for finance teams

Finance people are usually the last to say yes to AI, and for good reason. A wrong figure in a marketing draft is embarrassing. A wrong figure in a management report, a VAT working or a supplier payment is a real problem, and it has your name on it.

So the useful question isn't "can AI do this?". It's "who checks it, and against what?". If you can answer that before you start, AI becomes a draft you review. If you can't, it's a risk you've taken on without noticing.

Rule 1: a named person signs off

Every job you hand to AI has one named person who checks the output before it goes anywhere. Not "the team", a person. It's the same rule you'd apply to a trainee's first month: they can prepare the work, but someone with the authority signs it off.

If nobody is willing to put their name to the output, that job isn't ready to hand over yet.

Rule 2: every figure traces back to a source

AI should only use figures from material you've given it: the export, the ledger report, last month's pack. Ask it to say where each number came from.

That turns checking into a short, boring job: tick each figure against its source. It's the same reason the context pack in our earlier article tells the AI to write MISSING rather than guess. A gap you can see is easy to fix. A figure that looks right is the one that gets through.

Rule 3: AI drafts, people act

Write down what AI may never do on its own. For most finance teams the list is short and obvious:

  • send anything to a customer, supplier, bank or HMRC
  • approve or release a payment
  • change a figure in the ledger
  • decide a judgement call (a provision, a write-off, a credit note)

It can prepare all of those. A person does them.

Rule 4: keep what you gave it and what it gave you

When AI drafts something that matters, keep the inputs and the output together, in the same folder as the finished work. If a question comes up later, you can show what the draft was based on and who changed what. It costs nothing and it answers "where did this number come from?" months later.

Rule 5: watch what you paste in

Before anyone pastes in customer names, bank details or payroll, check what your business's AI tool is allowed to see. Some tools your company pays for keep your data inside your account. Free consumer versions may not. If you're not sure which one you're using, ask before you paste, not after.

What this looks like on a real job

Take the Monday sales summary from earlier in this series. With these rules in place:

Prepares: AI, from this week's sales export and last week's summary. Checks: the finance manager, ticking each figure against the export. Never does alone: sends the summary, or changes any figure in the ledger. Kept: the export, the AI draft and the final version, in this week's folder. Data: the export has no customer bank details, so it's fine for the company's AI tool.

Five lines. Write them once, and the job can be handed over without anyone wondering who's responsible for it.

Where are you now?

Trust and checks are where finance teams tend to get stuck, and they're easier to set up before the first job than after the first mistake. The readiness check below takes about three minutes and shows which part of handing work to AI is most likely to trip you up.

XLinkedInGitHubYouTube
Budget MonitoringCloud Cost OptimizationCloud MigrationData & AI IntegrationDevOps AutomationKubernetes & ContainersPerformance MonitoringSecurity & ComplianceServer Admin TiersServer Administration
© 2017-2025 Delta Famiglia Ltd · Registered in England & Wales · Company No. 10017910
Finance Tools - DevOps Answers