Invoices, Excel and AI: sorting out a builder's expenses in two days
A real AI automation case in a small business: what was wrong with the data, a two-layer solution and the next step with invoice QR codes.
Nuno Dias manages the expenses of CivilSul, a construction company, in Excel. Over 400 invoices a year, typed in by hand, with a problem every small business recognises: the same invoice was entered in several places.
What was really wrong
Before automating anything, we looked at the data and found three problems:
- One column doing three jobs. The same code stood for the site, the person or vehicle, and the block within the site.
- The same thing spelled many ways. A single code appeared in 19 different variations. Automating on top of that would only multiply the errors.
- The most important field was missing. The expense type (fuel, materials, labour) lived only in the head of whoever typed it.
The solution: one source, many views
With Claude we built two things in parallel on the same data model:
- A reworked Excel that works as plan B: one sheet instead of 12 monthly ones, closed lists to prevent mistakes, and a summary filtered by site, type, month, year and who paid.
- A mini app with a dashboard, the list of entries, pending items to review, catalogues and an export for the accountant.
Uncategorised spend dropped from about 40% to 1%. The process itself also surfaced invoices from the previous year mixed in with the current one.
The next step
Since 2023 every Portuguese invoice carries a QR code with the tax number, date, amounts and VAT. Reading it is simple: drop the PDF in a folder and the invoice arrives as pending, waiting for confirmation. Most manual typing goes away.
The lesson for other small businesses
AI does not fix messy data; it helps you organise it fast. The value is in getting the data model right first and only then automating.
Have a similar process in your company? Get in touch.