Clean the Till Codes First: AI on Your Own Numbers
An operator who has run pubs, a hotel and a visitor attraction has always managed by the numbers. For years that meant pulling data out of the till and the accounting package, adding notes by hand, and merging it all in spreadsheets. One year the figures showed thousands of adult visitors were never being charged admission. That finding changed the whole pricing.
AI makes that kind of analysis faster. It also makes it easier to get wrong. In a later review, the same operator found the till data full of traps. Two different products shared one code. Another product had a misspelt duplicate. "Feed that straight into an LLM and the totals go wrong without anyone noticing."
That is where most small businesses get stuck with AI on their own figures. The model is rarely the problem. The data is.
What goes wrong
Till systems, booking platforms and accounting packages each record the same sale their own way. Product codes drift as staff add items in a hurry. Some exports leave out days with no sales, so the columns stop lining up. One operator could only get PDF reports out of the till.
An AI tool will read all of it and produce a confident answer. If two products share a code, their sales merge. If a duplicate exists, one product's sales split in two. Nothing in the answer tells you.
One owner in the same discussion described the other common block. Three systems that do not talk to each other, and the thought of exporting and cleaning all of it before AI even looks. Most stop there.
Start with the decision
Someone in the replies offered the best way through. Ask what you expect AI to improve before you touch the data. You may not need all three systems.
The operator agreed. In the review, everything came from the till alone: sales by product and sales by hour. That was enough to answer the big winter questions on opening hours, staffing and stock. The accounts would only have mattered for margins.
This is the same advice we give on what to build first: start boring. Pick one decision you make every season, find the one system that holds the figures, and clean that.
Clean it, then check it
Three habits make the difference. Give every product one code and one record, and merge the duplicates. Fill the gaps, so a day with no sales shows as zero instead of missing. Check every AI result against the original totals before you act on it.
The operator does not let AI do the arithmetic. AI writes the calculation in a spreadsheet or a short program, which does not make arithmetic mistakes, and the result gets checked against the till's own totals. AI does the tedious work. The judgement stays with someone who knows how the business runs.
How to tell if you are ready
Data is one of the six pillars of readiness we score: whether your operational knowledge is findable, connected and clean enough for AI to use. Try one test. Export last month's sales by product and add them up. If the total does not match the till's own report, AI will not fix that, and it will hide it.
When the same figures need to feed more than one tool, the cleaning should happen once, in one place you own. That is the job of the Digital Brain, and it is why we argued that unstructured data is the new moat.
Where to start
Our free two-week audit scores your data alongside the other five pillars, ranks the three places AI pays most in your business, and puts a number on each. It takes about three hours of your time, and the report is yours whether or not you hire us.