Monday at the tire shop on University opens the same way it opened last Monday. Run the sales export from the point-of-sale at the counter, pull the deposits from the bank feed, count what the bays finished against what the board promised, and paste it all into a spreadsheet with a tab for every week of the year. Three hours later, the owner has the same five numbers he had last week, and the week is already moving without him.
The tour, step by step
It is worth naming the steps, because the owner performing them has usually stopped seeing them. The point-of-sale export comes out as a CSV whose date column Excel misreads as text, so it gets fixed by hand — every week, the same fix. The bank feed lags a day, so Friday's deposits show up Monday and have to be matched backward. The job board counts a rescheduled alignment as complete because someone closed the ticket to clear the screen, so completed jobs need a manual correction the spreadsheet applies but the job board never learns. The receivables aging for the fleet accounts comes out of QuickBooks, filtered to past thirty days. Then the paste, the dragged formulas, and — one February — the formula that was not dragged far enough, which made the month look down nine percent until someone found the error in March.
Call it three hours, generously. And it is the owner's three hours: the most expensive labor in the building, spent on assembly rather than on anything the assembly is supposed to inform.
The five numbers are the spec
Ask what actually gets read out of the finished report and the answer is short. For this shop: cash collected last week. Car count through the bays. Average repair order. Fleet receivables past thirty days. Payroll as a share of sales.
A different trade reads a different five — a retail store watches sell-through and shrink, a marketing agency watches billable hours against retainer burn — but the number is almost always about five, and the owner already knows theirs cold. Writing those five down is the entire design document. Everything else in the report is decoration that got into the template years ago and never left.
What scheduling the assembly means, mechanically
Every one of those systems has some export path: an API, a report it can email itself on a schedule, or a file it can drop somewhere on Sunday night. A small workflow collects the pieces, then applies the same corrections the owner applies by hand — the date-column repair, the rescheduled-job rule, the account mapping — and computes the five numbers, appending the week's row to the same spreadsheet the shop already uses.
Notice what is not happening: no new methodology, no new metrics, no consultant's framework. The rules are the owner's own rules, learned over years and currently re-performed fifty-two times a year. Encoding them once is the whole project. That is also why the first conversation is about the corrections, not the software — the value is in the rules, and the owner is the only person who knows them.
A report should notice things, not just record them
Two kinds of sentence are worth adding once the assembly is automatic. The first flags movement: fleet receivables past thirty days have risen three straight weeks. The second flags mismatch: Thursday's deposit total does not match Thursday's counter total.
The mismatch line matters more than it looks. Hand-assembled reporting is where reconciliation errors hide, because the person pasting is racing, not checking — the February formula error survived a month of Mondays. A machine that compares two totals every single week, without ever getting bored, is doing something the manual tour never actually did, even though everyone assumed it was.
Delivered into the inbox that already gets read
The finished report is a short email that arrives at 6:30 Monday morning: five numbers in the body, the spreadsheet attached, one flag line if something moved. No dashboard, no login, no password to forget.
That choice is deliberate. Dashboards are where reporting projects go to die quietly — built with care, admired at the demo, unvisited by week six. The owner already reads email in the truck at 6:45. A report should go where the reader already is, not where the software would prefer an audience.
Honest limits
Automation reproduces the data it is given. If techs close jobs three days late, the completed count is wrong on Monday and right on Thursday whether a human or a workflow assembles it — the report can flag the pattern, but it cannot fix the habit. Some older systems genuinely export nothing, and the workaround — a scheduled report emailed to an inbox the workflow reads — is ugly, though it works.
Expect the first two or three weeks to be reconciliation weeks: the automated report runs next to the manual one until they agree, and at first they will not. Most of the disagreement usually turns out to be the manual process's own quiet inconsistencies surfacing — a definition applied one way in January and another way in June — which is uncomfortable and useful in equal measure. And the whole exercise assumes an owner who knows their five numbers. If the honest answer is "just send me everything," the result will be a report as unread as the dashboard it replaced.
Where to start
Next Monday, run the tour exactly as usual — but keep a running note: every login, every export, every paste, every hand correction, and the clock time. That single page, written by the person who performs the tour, is the entire technical specification. It costs nothing to produce, and it is worth more than any discovery call.