AI
Monthly Reports From Messy Excel With AI, Without Making Up Numbers
The report is due tomorrow morning. The data is scattered across messy Excel files. And every time the numbers do not line up, you guess.
If you are an office worker stuck in this routine every month, this article is for you. I want to break down the workflow I actually use, plus the dark side nobody talks about. AI is at its most dangerous precisely when you are writing a report.
I sit in two chairs. As a Director of Product, every month I read reports from four teams. And every month I also write a report for my own boss. So I know exactly how both sides feel: the writer panicking, and the reader who can tell within seconds whether the numbers are real.
Why monthly reports freeze people
The problem is almost always the same. The data exists, but it is spread across several Excel files with different formats. One sheet per date, another per product, some with columns nobody can explain. Total rows mixed in with raw rows.
At this point, most people make one of two mistakes.
First, doing everything manually. Copying cell by cell into PowerPoint, calculating with a calculator, and losing half a day. Second, and more dangerous: pasting everything into an AI, asking it to write the report, and sending it without checking. Fast, but the numbers can be fabricated, and you are the one accountable in front of your boss.
There is a middle path, and it is short.
The workflow I use: paste, summarize, verify
Step one, clean the data just enough. It does not need to be perfect. It just needs to be readable by AI: remove empty rows, label what each column contains, and merge files where you can. Five to ten minutes.
Step two, paste it into the AI with clear instructions. The prompt looks roughly like this:
"This is this month's sales data. Build a monthly summary for my boss with sections: key achievements, trends versus last month, the three biggest problems, and recommendations. Quote numbers exactly as they appear in the data. Do not round, do not recalculate. If any number looks odd or inconsistent, flag it. Do not guess."
That last part is the most important. You are telling the AI it is allowed to say it does not know. An AI that is forced to answer will answer, including made-up answers.
Step three, and this cannot be skipped: verify the numbers manually. Every number that appears in the report, go back to the Excel file and find its source. Use Ctrl+F, filters, or a pivot table. This is the fastest part of the whole process, because you are checking, not computing from scratch.
Total time? Work that used to take half a day now takes about an hour. Even less if the data is already tidy.
From the chair of the person reading the report
Now the other side. Every month I read reports from four teams, and I usually need only a few minutes to know whether a report can be trusted.
How? I ask about one number. "Where does this come from?" If the writer can point to the source within a minute, my trust in every future report from them goes up dramatically. If they mutter "I think the AI generated it," it is over. I will read every report from that person under a microscope for months.
That is what is at stake. It is not about how fast you produce the report. It is about the reputation of your numbers in the eyes of the person who makes decisions based on them.
I have seen savings figures on people's resumes and known instantly which parts were embellished. People who actually did the work can always describe the process, not just the result. Monthly reports are the same. A number whose source you understand looks different from a copy-paste number.
What to remember about AI here
AI is great at structuring and lazy at arithmetic. It is excellent at fixing structure, finding patterns, writing narrative. But it can miscalculate with total confidence, especially when the data is messy and you have not explained the format.
So divide the work like this: AI builds the frame and the narrative. You own the numbers. As long as that division is clear, you are safe. The moment you swap them, you are the one in danger.
Saving companies money is an old habit of mine, long before AI was trendy: $4M+ per year from a mix of initiatives I pushed as a product leader, including process fixes, cost optimization, and smarter product decisions, with automation as just one contributor among many. Back then, that kind of leverage required seniority, an engineering team, and expensive systems. Now AI is the sharpest tool for the same habit, and it can be trained into everyone on your team.
If you want to learn workflows like this in more depth, alongside people practicing them every week, check out AI Circle at brianarfi.com/ai-circle/. And if you lead a team and want everyone producing reports like this the right way, the corporate page is at brianarfi.com/corporate/.
This month, run the workflow once. Paste, summarize, verify. Feel the difference when you send your report at 10 a.m. instead of 7 p.m.