The report that writes itself: automated reporting for management
The most expensive person in the company is often the one spending Friday compiling numbers for Monday’s report. This shouldn’t be happening anymore in 2026.
In a growing SMB, the weekly management report is almost always done by hand by a senior person: pull sales from the CRM, revenue from billing, tickets from support, spend from the bank — and paste everything into a document with commentary. Costs four to six hours a week. Multiplied by 50 weeks, that’s a whole month a year of an expensive person.
In 2026, this shouldn’t be happening.
What a good reporting system does
Three layers, all indispensable:
- Automatic collection from source systems (CRM, billing, support, banking, ads).
- Calculation and comparison against the previous week, the month, and the same week last year.
- Natural-language drafting of the highlights — what went up, what went down, what deserves attention.
The last step is where AI changes the game. A Looker or Power BI has delivered the first two layers since 2018. What was missing was the narrative — and the narrative was what the senior person was writing by hand.
What AI writes well — and what it still doesn’t
Writes well:
- "Sales grew 12% vs. last week, driven by client X who closed €25k."
- "Average support ticket time rose 40 minutes. Strong correlation with Tuesday’s release."
- "We’re 8% above the quarterly target with two weeks to go."
Still writes poorly:
- Causation — saying why something happened. Correlations, yes; causes, no.
- Market nuance — impact of a strike, local holiday, industry news. Depends on context the model doesn’t have.
- Action recommendations — usually comes out generic or dangerous. Leave that to the reader.
The pattern that works
A hybrid report: facts written by AI, interpretation written by the person who leads. The person saves four hours and gets a document where the only remaining task is the one they should be doing — thinking.
How to start without a big project
- Pick one report — the most annoying to build, the one that hurts most on Friday.
- List the five to seven data fields that always show up. More than that, leave for later.
- Automate only collection and structure in week one. Only when that’s stable, layer in the AI-written narrative.
Three weeks in, the senior person has Friday afternoon back. Three months in, they can’t remember how they used to do it before.