The Monthly Report That Takes Two Days to Build

In most businesses between five and fifty million there is one person who disappears for two days near the start of each month.

They are exporting from the accounting system, pulling figures out of the CRM, downloading spend from the ad platforms, and reconciling all of it in a spreadsheet that somebody built four years ago and nobody fully understands.

At the end they produce a report describing a period that finished a fortnight ago. Everybody reads it, nobody decides anything differently because of it, and next month they do it again.

The cost is not the two days

Two days a month is roughly twenty four days a year of a reasonably senior person. That is the visible cost and it is the smallest part of it.

The real cost is latency. A number describing January that arrives on the fifteenth of February is a historical record rather than a control. Whatever went wrong in the first week of January has now been going wrong for six weeks.

The second cost is that manual assembly makes frequency impossible. Nobody is going to do this weekly, so the business runs on a monthly cadence regardless of how fast things actually move.

The third is trust. A spreadsheet with manual steps has errors in it. Not many, but enough that somebody eventually finds one, and after that the whole report carries an asterisk in everybody’s head.

Why it is still manual

Rarely because nobody thought of automating it. Usually because of three specific obstacles.

The sources do not agree. The CRM says eleven deals, finance says eight. Somebody has to reconcile that, and reconciliation feels like judgement, so it stays with a person.

Some of it is not structured. A supplier statement arrives as a document. A subcontractor sends an invoice by email. A project update lives in a message thread. Rules based tools cannot read those, so a human transcribes.

The report was never specified. It grew. Somebody asked for a figure once, it was added, and now nobody can say which parts are load bearing. Automating something nobody can define is genuinely difficult.

Fix the definitions before touching the tooling

The reconciliation problem is not a data problem. It is that two systems are counting different things and calling them the same word.

Agree what a closed deal is, when revenue counts, whether a returning client is a new sale, and whether a cancellation reverses a previous month. Write it down where people can find it. That afternoon removes most of the reconciliation, and it has to happen first, because automating a disagreement produces the disagreement faster and with more confidence.

Then cut the report in half

Before automating, ask which figures anybody has ever acted on.

Most monthly reports contain twenty numbers, of which perhaps five change a decision. The rest are there because somebody wanted them once, or because the template came from somewhere else.

Start from the decisions. Where does next month’s marketing budget go. Do we need to hire. Which clients are quietly unprofitable. Is the pipeline enough for the quarter. Which service line is carrying the business. Five questions, three numbers behind each, and everything else is interesting rather than useful.

Where AI does the part that blocked automation

Connecting structured systems has been possible for years. What kept this manual was the unstructured material, and that is precisely what changed.

AI reads the supplier statement, the emailed invoice, the free text note in the CRM and the message thread where an exception was agreed. It extracts the figures and writes them into the same place as everything else. That single capability removes the reason a person had to sit in the middle.

It also does the second job, which is explanation. A chart tells you cost per client rose eighteen percent. It does not tell you that it rose because one channel scaled while converting worse and the other three were flat. Producing that sentence every week, reliably, is work a system can do and a person does only when they have time.

Stanford’s AI Index for 2024 found that assistance of this kind both speeds up task completion and improves output quality, with the largest gains among people less experienced at the task. For reporting that matters, because it means the analysis no longer has to wait for the one person who knows how the spreadsheet works.

Send a summary, not a dashboard

The instinct is to replace the report with a dashboard. Dashboards get built, admired for six weeks, then quietly abandoned, because somebody has to remember to open them and busy people do not.

A short written summary arriving before the week starts gets read, because reading it is easier than not reading it. Five numbers, what changed, what caused the change, and what needs attention.

It also changes the meeting. When everybody has seen the same five numbers beforehand, the hour goes on what to do rather than on establishing what happened, which is where most management meetings currently spend their first half.

Every figure has to be traceable

One rule protects the whole thing. Any number on the page must be followable back to the record it came from.

The first time a figure looks wrong and nobody can explain where it came from, credibility is gone permanently. People return to their own spreadsheets and you are worse off than before, because now there is another version of the truth in circulation.

If a number cannot be traced, leave it off the page.

What to do this month

Time the exercise honestly. Ask whoever builds the report how long it actually takes, including the chasing, and multiply by twelve.

Then list the figures and ask, for each one, what decision it has changed in the last year. Delete the ones with no answer.

Then agree the definitions for what survives. At that point the automation is a small technical job rather than a project, because the difficult parts were never technical.

Sources

• Stanford Institute for Human-Centered Artificial Intelligence, Artificial Intelligence Index Report 2024, 15 April 2024.

AI Optimize connects the systems, reads the material that was never structured, and sends the summary before the week starts. That work sits under Reporting & Data.