The Billion-Dollar Manufacturer Whose Executive Dashboard Was Built by Hand

A manufacturer with revenue in the billions was producing its executive reporting by hand.

Someone wrote production numbers into a paper register. Someone else typed those numbers into Excel. A third person rebuilt the charts in PowerPoint. Then a human got on a call and explained the slides, because the slides on their own didn't carry the reasoning.

Four handoffs. Every one of them a place where a row goes missing and nobody finds out until it matters.

This was an operation with the money to buy any system it wanted, running its production and procurement decisions off a deck someone assembled by hand.

What actually breaks in a chain like that

The obvious failure is transcription error, and that's real. But it isn't the expensive one.

The expensive one is that nobody can ask a follow-up question. A stakeholder sees a number that looks wrong and asks why. Answering means finding the register, finding the Excel file, working out which version is current, and rebuilding the analysis. So the question gets asked once, and after that people stop asking. Decisions get made on the number in the deck because interrogating it costs a day.

The second is latency. By the time a deck exists, it describes a period that already closed. You are steering by a rear-view mirror that somebody polished by hand.

The third is the one that costs real money, and it's why revenue leakage lives here. A leak is a pattern, not an event. A pricing exception that recurs on one customer, a contract term nobody reconciled against what shipped, warranty claims drifting above what was priced in. None of those show up in a monthly snapshot assembled by a person. They only surface when you can ask questions across the whole history, and this operation could not ask questions at all.

What we built

A production and procurement engine, one system, with every step digitised end to end.

Entry that validates at the source. Data goes in through forms with validation rules, not into a register and then into a spreadsheet. An error gets caught by the person who has the context to fix it, at the moment they have it, instead of surfacing three handoffs later when nobody can reconstruct what was meant.

Storage with a shape. Structured, with required fields enforced. That matters more than it sounds: a blank field quietly drops out of every downstream total, and nobody notices the number is short.

Visualisation that builds itself. The PowerPoint step disappears, and a dashboard nobody assembles by hand can stay current.

An AI layer over the top, answering questions in natural language and surfacing revenue-leakage patterns across the history. This is the part that changes what the operation can do rather than how fast it does it. The question that used to cost a day now costs a sentence.

What we can say now

The system is live. It has not been running long enough to publish outcome numbers, and this piece is not going to invent them.

What can be said is what the architecture removed. There is no longer a step where a person copies a figure from paper into a spreadsheet. There is no longer a version of the truth that exists only in one analyst's working file. There is no longer a question that costs a day to answer, because the question can now be typed instead of scheduled.

Those are structural claims about the build, and they are the honest thing to say this soon after launch.

The results piece comes later, and it will have numbers in it.

What surprised us

The gap between how much money moves through this industry and how little of it is digitised.

This was not a company that couldn't afford better. It could have bought anything. The reporting ran on paper and PowerPoint because nobody had ever been made to own the pipeline end to end. Each handoff belonged to a different team, and every team's part worked fine in isolation. The failure only existed in the joins, which meant it was nobody's job.

That's the thing worth saying out loud to anyone in this sector: the constraint is almost never budget.

Frequently asked questions

Why digitise reporting before doing anything with AI?

Because an AI layer answers questions using whatever data reaches it. If production numbers arrive by spreadsheet four days after close, with fields missing, the model will analyse that faithfully and confidently. Structured entry is unglamorous and it determines whether everything built on top is worth having.

Did this require replacing their ERP?

No. In most operations we look at, the individual systems work and the joins between them don't. Connecting what's already running is cheaper, faster, and doesn't ask a team to relearn their jobs mid-quarter.

How long does a build like this take?

Smaller automations go live in one to two weeks. A production and procurement system spanning entry, storage, visualisation, and an AI layer runs in phases, so working pieces arrive early instead of everything landing at once at the end.

What size of company does this apply to?

Size matters less than whether anyone owns the pipeline end to end. A thirty-person business can have the same broken chain as an operation with revenue in the billions.

If you recognise your own reporting in the chain above, a fixed-scope AI audit maps where automation actually pays before you commit to building anything, including the processes we think you should leave alone.