CodelessOps · AI with receipts
Automation for finance teams.
AI that shows its working.
I set up the AI you already pay for, safely. I build custom automation for the processes that eat your month end. And when the whole drive needs to be answerable and proven, I build your team itsknowledge base. CIMA-qualified, fifteen years in FP&A, and I build the systems myself.
You ask
"Why did Q2 opex jump against budget?"
Grounded answers
Q2 opex was £4.82M against a £4.20M budget, £0.62M over, driven by a one-off licensing true-up of £410k in May.
And when the documents can't support it"I can't confirm that from the documents provided."
01 · What I do
Three ways in. Most teams start with the first.
The Audited AI Close
Your existing Claude or ChatGPT licence, wired into the finance function properly: scoped connectors, close workflows, a guardrail policy your CFO signs, sign-off gates, and an audit pack every close. No new software unless it's absolutely necessary.
Custom automation, measured
One process at a time: invoice flows, reconciliations, reporting pipelines. One agreed metric, measured over one month end, so the improvement is provable, not promised.
Your finance knowledge base, built
A knowledge base of your own documents: packs, accounts, policies, contracts. Every answer cited to the exact cell and checked, and you sign off when it passes 25 questions your own team wrote.
02 · Who you're hiring
One person, both languages.
I spent fifteen years inside finance teams. Month end, board packs, planning models. CIMA-qualified, contract FP&A builds at places like Goldman Sachs, Vodafone and the NHS, then the full planning stack at an AI drug discovery biotech. Now I build the systems instead.
Everything is built to enterprise practice: documented, versioned, auditable, and yours when I leave. I publish what I learn, including a public benchmark on AI accuracy in finance.
03 · Common questions
Asked by every finance leader I talk to.
Doesn't AI get numbers wrong?
It does, and that's the premise of everything I build. Nothing I set up posts a number without showing its working, and a named reviewer signs off before anything reaches the ledger. I take this seriously enough that I published a public benchmark on it: raw frontier models scored 66 and 71 of the 81 machine-checkable points on a real board pack; a verified pipeline scored 81 of 81, the ceiling. My first system failed this way. That's why mine now don't.
What will our auditors say?
They'll ask about AI in your close this year. This setup is the answer: an audit pack every close, each AI-assisted output with its sources and the reviewer's initials.
We already pay for ChatGPT / Copilot. Why would we need you?
Good, you're already paying for the licence. Having it isn't the same as getting value from it. My job is making it safe and actually useful for the close. Almost nobody's done the wiring.
Is our data safe?
No-training workspace tiers, least-access connectors, and a one page leak policy your CFO signs before anything gets connected.
We're too small, and our processes are a mess.
Messy is the normal case, and small means we fix it faster. That's what the discovery call is for.
Will this scale, or is it a toy?
Everything's built to enterprise practice from day one: scoped access, documentation, versioned components, audit trails. I built at enterprise level for fifteen years before AI. The first automation is deliberately small. The standards behind it aren't. It scales when you do.
What happens when you leave?
Your team takes over, and I mean it. The routine changes are yours to make: the run-book marks exactly which changes are safe to self-serve, and your named admin makes a few of them, live, before I sign off the handover. For the structural work, new processes, new integrations, anything touching the controls, I'm one email away. It's documented well enough that any competent engineer could pick it up. Most clients call me because I built it and it's faster.
04 · Thirty minutes
Tell me where your month goes.
Walk me through your close and I'll tell you exactly what I'd automate first, and what I'd leave alone. You end the call knowing where you stand, whether or not we work together.
- No pitch unless you ask for one
- Free discovery call, written estimate within 48 hours
And if the pain is answers buried in documents, start withthe knowledge base build: bring your last close pack and watch your own questions answered with the source, or refused, the same way.
Featured
RSS FeedIntroducing Grounded: AI that returns the version-correct number, with receipts
Grounded is an audit-grade retrieval layer for financial documents: version-correct, sourced to the exact figure, and honest when it can't answer. Here's why we built it the way we did.
Recent Posts
The anatomy of a financial knowledge base
Engineering teams have shown how to build company knowledge bases. Finance needs a different animal. Seven organs, what each one is for, and how I built them.
The questions your finance team has stopped asking
The entire pitch for what I built, asterisk included: why the board pack questions that cost an afternoon stop getting asked, what changed this year, and how I'd test any AI that says it's fixed that, including mine.
How to use AI at work without leaking data
Is it safe to put company data in ChatGPT? A working protocol for finance teams: what to paste where, how to anonymise a question properly, and how to get IT to yes.
The Board Pack Test: I gave four AIs the same board pack, then graded them with a script
I published an exam for finance AI: 34 realistic documents, 25 questions, a deterministic grader, answer key included. The frontier models got most answers right. Proving them right was a different story.