A vague list
The brief arrived as a list: count the assets, count the stock, work out what the goods cost. Sensible, and vague.
I've learned that vague briefs hide the real problem. So I didn't start counting. I turned the list into a clear plan: a physical count, then costing, then reporting set-up, then handover, run as six tracks in parallel, so nothing waited on anything it didn't have to.
Four answers to one question
Then I asked the question that mattered: what's your margin? I got four different stories, depending on which version of the model I looked at, and one of the 'actual' columns was empty.
I didn't pick the most believable one. I told her I would not use any number until there was evidence behind it.
I would not use a number until there was evidence behind it.
Rules before numbers
So I set the rules before touching a spreadsheet. Build forward, not backward. Value things by the best evidence available, in a strict order: an invoice, then a bank record, then a quote, then a clearly labelled estimate. And check everything the owner tells me against something else, including the things she is most sure about.
It sounds strict. It's the opposite. It's what lets everyone trust the final numbers, her most of all.
Counting, and cost cards
We counted the stock, wrote down every recipe for every product, and built a cost card for each one. The recipe sheet went through eleven versions as the real quantities came to light. The cost cards settled the argument that the spreadsheets couldn't.
In the first batch alone, verified unit costs were set for 49 products across 6,250 units.
The contract I didn't sign as it was
While the work was under way, the contract for it arrived from the client's lawyers. I could have signed. Instead I sent it back with seven amendments: clear acceptance tests, a cap on liability, and a plain statement that estimates are estimates, not an audit.
A good agreement protects both sides. It's one of the least glamorous things an adviser does, and one of the most important.
The cost that looked wrong
One product's cost looked far too high, and I wouldn't let it pass. I traced it to a stale stored cost, then recalculated the whole history, with checks before and after, so the correction could be proved. That meant correcting 7,957 cost entries across 1,632 documents and posting ten journals, while changing nothing the client had entered herself.
Traced to the kobo, and fixed across the whole history.
A business that reports for itself
With the numbers settled, I built the structure around them: an annual report, an audit pack, and a monthly management pack. I prepared funding materials for two programmes, writing one pitch deck two ways, one fact-led and one emotional, so she could choose the voice that fitted the funder.
Then came the part I care about most. From 1 July 2026 she recorded on her own, and the first monthly position report was checked line by line against her bank statements. The business no longer needed me to know its numbers.