Data, Analytics & Engineering · Enterprise reporting
Three person-hours a week, returned
The weekly report had the same shape every week: export, clean, analyse, write, format, send. Two people, an hour and a half each. The pipeline now does the mechanical part in ten minutes and the humans do what humans are for — checking that the numbers deserve to be trusted.
At a glance
- Practice
- AI & Automation
- Cadence
- Weekly
- Volume
- ~50 MB per file
- Before
- 2 people × 1.5 h
- After
- 10 min tooling + 20 min review
Python · Pandas · LLM analysis · Word/PDF generation · Scheduled delivery
The situation
The report was not hard. It was just relentless
Two analysts each spent about an hour and a half a week producing the same weekly report: pulling the export, cleaning it, running the analysis, writing the observations and formatting the output. Nothing about it was intellectually difficult. It simply arrived every week, forever.
The obvious automation would have been the wrong one: an AI that writes a confident summary nobody can verify. At around fifty megabytes per file, a wrong figure is easy to produce and nearly impossible to spot downstream — so data integrity had to be part of the design, not a hope.
What we did
Four steps, in this order.
Measured the manual baseline
Two people, an hour and a half each, every week. That figure became both the budget for the work and the criterion for calling it successful.
Automated the mechanical half
Ingestion, cleaning, calculation and draft narrative run in about ten minutes across roughly fifty megabytes per file.
Kept a human on integrity
Twenty minutes of curation and refinement remain deliberately in the loop. Figures are computed in code and traceable to source; the reviewer confirms the reading, not the arithmetic.
Left the schedule alone
The report still lands weekly, in the same format, for the same audience. Nobody downstream had to change how they work.
Results
The numbers, as measured.
The takeaway
What this case actually shows
The twenty minutes that stayed are the interesting part. It would have been easy to claim the whole ninety, and the demo would have looked better — but a reporting pipeline nobody checks is a liability dressed as a saving. The human step is what makes the other eighty-nine percent safe to bank.
Start here
Have a version of this problem?
Describe the problem in a few lines. You get a written reply with a first read on it, whether or not there is an engagement in it. The message is sent from this page — no email client, no third-party form service, no trackers.