Your team already wrote the documentation. It's called a spreadsheet.
Everything AI needs to be useful in your business is already written down, in the workbook only one person really understands. Here's how that becomes a governed AI team of your own.
Why Excel hell is worth taking seriously, and why killing it is the wrong instinct
Every finance team has one: the workbook with forty tabs, six years of history, and one person who genuinely understands column BF.
94% of audited operational spreadsheets contain at least one error, across independent field audits by firms including KPMG and Coopers & Lybrand, compiled by Raymond Panko and published through the European Spreadsheet Risks Interest Group (EuSpRIG). The risk is real and measured, not a reason to delete the workbook, a reason to stop trusting it unsupervised. Source →
The workbook, read as documentation rather than as debt
"If she leaves, we're in real trouble, and everyone knows it."
"There are four versions. Three of them are called final."
That workbook is the most honest documentation of how your business really works.
So no, this isn't another Excel killer. The formulas encode decisions nobody ever wrote in a policy document: which accruals reverse, why that cost centre is treated differently. Throw the file away and you throw away the knowledge with it. Excel stays exactly where it's genuinely good, as the place people think and model. What changes is that the logic inside it becomes the foundation: agents do the reliable, repeatable part of the work on top of it, and a person still signs off the part that needs judgement.
02
Why the chatbot didn't fix it
The difference between personal productivity and a process you can trust
Most organisations have now run the experiment. An AI assistant went out to the business, and something real did happen: emails drafted faster, documents summarised, meetings minuted.
And then it stopped. Month-end takes exactly as long as it did last year. The reconciliation is still manual. The forecast still rebuilds from the same fragile file.
95% of generative AI pilots at companies are failing to reach production, according to an MIT report covered by Fortune. Not because the models are weak, because the foundation underneath them isn't there yet. Source →
The reason is mechanical, not mysterious. Ask an off-the-shelf AI for "net revenue" and there are probably six fields across your systems that could plausibly mean it.
Six plausible fields, one silent choice
The point in one line: the model doesn't tell you it guessed. That's fine for a first draft of an email. It's not fine for a number that goes to the board.
A business process has one right answer. A chat reply doesn't.
What actually closes the gap. Not a smarter model, a governed structure around the same one everyone already uses. The calculation and the rule still get built the traditional way: real Power BI, real Fabric, logic that runs identically every time. What's new is that AI can now build that layer with you, and work on top of it once it exists.
What just happened has a name
Self-Service AI
Business teams getting trustworthy answers and finished work out of AI on their own, because the rules of the business are written down somewhere the AI can actually reach.
Self-service BI made the same promise: drag, drop, done. It delivered charts.
03
The organisational brain
Five layers, built one process at a time, not a transformation programme
Getting the logic out of that workbook and somewhere AI can reach it is the whole job. What it lands in, we call the organisational brain: five layers your business already half has.
One process enters here ↓
01DataSource systems and schemas, structured once, referenced not copied.IT & Data
02LogicThe definitions your business runs on: net revenue, period close, this month's exception.Finance
03ToolsPermissioned connections into the systems you already run: SAP, NetSuite, D365.IT & Data
04SkillsEach process you capture becomes a certified, reusable skill. This is the layer that grows.Business
05GovernanceWho sees what, and where a human signs off, from the first skill, not bolted on later.IT & Data
↳ and leaves as one certified skill
You don't stand all five up before anything works. The first process you capture touches all five at small scale: one workbook, one set of definitions, one permission model, one approval point. The brain grows by one certified skill each time you do it again. IT and data own the platform and governance; the business team owns the definitions and the skills, because that's the knowledge only they have.
04
Your digital team
One accountable human, six agent roles - not one assistant with a prompt box
A single agent pointed at a business process behaves like a brilliant new joiner with no job description: capable, confident, and quietly ungovernable. So the work is split the way you'd split it across people, and each part of the brain gets someone accountable for it.
Each agent is a part of the brain. You're who it serves.
Your weak spots become your team's strengths.
One human lead and six agent roles, each introduced in a short video. Ellie stays accountable for the process; the agents divide the specialist work around her. Meet the team:
Human lead
EllieFinance Process Owner · Human
Defines the outcome, supplies the business context and signs off the work that still needs judgement.
The worker
MayaData & BI Analyst
Gets your data ready and turns your process into a reusable skill.
The skull
CamAI Concierge
The harness that holds the team together - and the agent you talk to.
Memory
RichardResearch Analyst
Brings your data, business logic, numbers, definitions and rules into the shared context.
Senses
DanielPlatform Architect
Connects the team safely to your ERP, CRM, SharePoint and other systems.
Learned behaviour
TanyaTools & Skills Mgr
Maintains the reusable patterns the organisation has approved for new problems.
Inhibitions
GabrielleGovernance Mgr
Defines what the team may do, what it must not do and who may ask.
Ellie remains accountable for the outcome. All six agents read from the same organisational brain, so the number your finance director sees matches the one the report was built from.
05
Where you'd start
One process, chosen because it hurts, not a programme
Most organisations sit at the first rung. The gap to the third isn't a budget cycle away; it's one captured process away.
MOST TEAMS ARE HERE
01Personal productivityLeaves with the person.
02Process transformationThe process owns the result.
03Organisational capabilityThe team improves it safely.
01
Excel-to-Fabric
When reporting or reconciliation depends on a workbook nobody wants to touch. The logic comes out intact; Excel keeps doing what it's good at.
02
Finance Process Automation
When planning, budgeting or forecasting is slow, fragmented and manual, and the cycle time is the problem.
03
BI Modernisation
When the reporting estate has more duplicates than definitions, and the technical debt is compounding.
Not sure which applies? The Process & AI Readiness Assessment takes about five minutes and comes back with one specific next step, rather than a maturity score you'll never look at again. Take the assessment →
Hear it from the room
The Finance × AI Lab interview
Rishi sat down with Finance × AI Lab to walk through the spreadsheet-to-agent journey this page just summarised, in his own words.
