← All posts

Weekly reports that write themselves (from numbers you already keep)

Automation · 22 November 2026 · 9 min read

A five-person landscaping business loses half a day every Friday assembling a weekly report by hand, and the numbers still disagree with the spreadsheet. A six-step method: name the five numbers that actually get read, put them in one place, fix the dates, build the smallest thing that assembles it, run it in shadow mode, then make it deliver itself.

The scenario

A landscaping and garden maintenance business of five people on the Sunshine Coast. The owner, a foreman, two gardeners and an office manager who also does the books. Three vans, about forty regular maintenance clients, plus the occasional bigger job where a yard gets cleared and rebuilt.

Every Friday afternoon, the office manager sits down to build the weekly report. Not for a client, and not for a bank. For the owner, so he can see what happened this week and what to chase next week. On paper that sounds sensible. In practice it takes her most of Friday afternoon, and it never quite matches reality.

Here is how it goes. She opens the job management tool and exports the week's jobs to a spreadsheet. She opens the accounting package and exports invoices raised and payments received. She opens the rostering spreadsheet, which the foreman keeps up to date and she does not, and copies the hours across. She opens last week's report and copies the layout. She writes a few lines of commentary from memory. Then she saves it, works out the totals, notices the job tool says nineteen jobs and the accountant's file says eighteen, and spends twenty minutes working out which one is wrong.

The owner reads it on Saturday morning, or sometimes Monday. By then half of it is out of date. And because he suspects the totals are not quite right, he re-checks a couple of numbers against the jobs list anyway.

That last detail is the tell. A report nobody trusts is a report that costs twice: once to make, and again to verify.

This post is about the report assembling itself from numbers the business already keeps, and the unglamorous work that has to happen before that is even possible.

Why the report is slow before any tool is involved

The report feels like a reporting problem. It usually is not. It is three problems stacked.

First, the numbers live in four places run by three people, and each has its own definition. Is a cancelled job a job? Is an invoice raised on Friday for Monday's work part of this week or next? Does a payment that arrived on Saturday belong to last week? Nobody wrote the answers down, so every export answers them differently.

Second, the week boundaries disagree. The rostering spreadsheet runs Sunday to Saturday because that is how the work happens. The invoice export runs Monday to Friday. So Friday afternoon's report compares two different weeks and produces a discrepancy that looks like a data error and is actually a calendar error.

Third, the commentary. Someone writes what happened from memory, which means the report is only as good as a tired person's recall of five days.

None of the three get fixed by buying a dashboard. A dashboard pointed at four disagreeing sources just shows you the disagreement more stylishly.

Step one: work out what actually gets read

Open the last four weekly reports. For each line in them, ask the owner whether he has ever made a decision because of it. Most lines never get acted on. They were put there because reports are supposed to have them.

For this business, the honest list is short. Jobs completed against jobs scheduled. Hours worked against hours estimated. Invoices raised, payments received, and what is still outstanding. Quoted jobs with no answer after a week. And a count of new enquiries.

That is five numbers and one list. Everything else comes off. A smaller report that is trusted and read beats a complete one that is checked, doubted and skimmed.

Write them down, one per line, in plain English, with the sentence that defines each one. "Payment received means money in the bank this week, not invoice marked paid." That sentence is what stops two exports from disagreeing later.

Step two: give every number one home

For each of the five numbers, decide where the value officially lives. Not where it can be found, where it is defined.

Completed jobs live in the job tool. Hours live in the rostering spreadsheet. Invoices and payments live in the accounting package. Quoted jobs waiting on an answer live in the job tool. Enquiries live in the shared inbox.

That is five sources, and that is fine, because problems only start when the same number is defined in two places. The jobs tool and the advertising file both containing a client list is fine. The jobs tool and the advertising file both containing a job count, with different rules, is not.

Where a number is genuinely unclear, fix the source rather than the report. In this business the main one is the account manager's habit of raising a Monday invoice on Friday so the week looks tidy in his invoice list. That is an invoicing rule, not a reporting rule, and it gets written down and stopped regardless of the report.

Step three: agree the week

Pick one week definition and use it everywhere. For a business whose work happens Sunday to Saturday, that is the one. Then make sure every export is filtered to it, using the same date field for the same kind of thing.

This sounds trivial and it is the single step that removes most of the twenty lost minutes. A report that compares Sunday to Saturday jobs against Monday to Friday invoices will never reconcile, no matter how good the automation is. Fix the calendar before you fix the report.

Step four: build the smallest thing that assembles it

The temptation now is to build a dashboard. Do the opposite first.

Write a short script, or build a single automation, that does exactly what the office manager does: pulls the four exports, renames the columns to the agreed definitions, lines the dates up to the agreed week, and writes out one table with a row per number and a column per week. It should also write out a short exceptions list: jobs scheduled but not completed, quoted jobs older than seven days with no answer, and any number that moved by more than a sensible amount from last week.

The exceptions list is what replaces the commentary. The office manager was never adding value by writing "busy week". She was adding value by noticing that a quoted job had gone quiet or a client was two weeks behind on paying. A good exceptions list surfaces those automatically, and then a person writes two or three honest sentences about them.

Keep the output boring. One table, one exceptions list, dated. A big beautiful dashboard will get admired for a fortnight and then ignored, because nobody wants to read a dashboard to find out whether a quote went quiet.

Step five: run it in shadow mode

For the first three weeks, run the automation on Friday afternoon, but still build the report by hand and compare the two. This is boring and you will want to skip it.

Do not skip it. The comparison finds the things that only a person would have caught: a job marked complete that was actually rescheduled, a payment received into the wrong client account, a week where a new job type means the counting rule needs a tweak. Every one of those is a silent wrong number that would have gone into the report and eroded its credibility.

Three weeks is usually enough, because the first week finds the systemic issues and the next two find the rare ones. Write down every difference and what caused it. If a difference keeps recurring, the definition needs changing, not the automation.

Step six: let it deliver itself

Once it reconciles for three weeks, move delivery to a fixed time. Say, Saturday at 7am, in the owner's inbox, as an email with the table inline and the exceptions list below it. It should be possible to read the whole thing in ninety seconds on a phone, which is where it will be read.

Put the exceptions at the top, not the numbers, because the exceptions are the part that needs a decision. And put the revision number and the week it covers in the subject line, so the owner answers "is this current?" without opening it.

Keep one manual thing: the two or three sentences of commentary. Those sentences are the owner's judgement about why the week was what it was, and they are the only part that a machine should not write. Everything else can be assembled.

Failure modes to expect

A number that changes meaning without anyone noticing. Someone starts marking a job complete when they finish the work, and then starts marking it complete when the client pays. The job count doubles. Keep the definitions in a short doc next to the script, and re-read it when a number looks wrong rather than trusting the script.

A silent export failure. The accounting package gets updated, the export format changes, and the script starts writing a table full of zero blanks. Have the automation check that it got a sensible number of rows and send an alert to the owner rather than a report, when it did not.

The report gets too clever. Somebody adds a chart, then a comparison, then a target. Nobody reads it. If it grows past a screen, cut it back.

One person becomes the report. If only the office manager understands the script, the business has replaced a manual job with a fragile one. Keep it short enough that she can explain it to the owner in two minutes.

What good looks like after a month

On a Friday, nobody builds the report. It arrives on Saturday morning in a form that takes a minute and a half to read, with the exceptions at the top and last week's numbers beside this week's. The numbers agree with the jobs list and the bank, because the definitions behind both are written down.

The owner has started using the quoted-jobs-with-no-answer list on a Monday morning, which is a conversation he was not having a month ago. And the office manager has half a day a week back, which for this business is the difference between doing the reconciliation properly and doing it in a hurry at month end.

The setup did not need a dashboard. It needed five numbers with one home each, one definition of a week, and something dull to put them together.

Common questions

Can I automate my weekly business report without buying new software?

Usually yes, if the numbers already live in tools you export. A short script or a simple automation can pull the exports, line the dates up and put them in one table, as long as you first agree what each number means.

Why do my weekly reports never match my accounting software?

Almost always because the week boundaries or the definitions disagree. One system runs Sunday to Saturday and another runs Monday to Friday, or a cancelled job is counted in one place and not the other, so the totals differ by date rather than by error.

How do I make sure an automated report is actually right?

Run it alongside your manual report for a few weeks and compare them line by line. Once the difference list is empty, keep the plain-English definitions next to it so a change in how a number is recorded gets caught.

Want this done for you?

I build one small automation at a time for North Shore businesses, usually starting with whatever is costing you the most hours.

Start a project ↗