How I Use AI Agents In Business Operations: Raw Data to Report in Minutes
Jason BrownShare
I gave an agent a raw data file and one prompt. It came back with:
- The top 5 things I need to know right in the chat.
- A multi-tab Excel report ready to send to the team or present to leadership including a summary, a supplier risk scorecard, and an action list, all done in a matter of minutes.
Here's how I got this result.
The Background
Earlier this year I built a Supplier Audit Excel Tool and walked through it in a YouTube series. That series runs 6 videos and 60 minutes in total. It's a great resource for those looking to get started in this world where everyone is a developer building tools.
Recently I've been working with AI agent teams, discovering the real benefits and risks for a business. This week I launched a new walkthrough. The difference? A single video under 20 minutes, and it gets better results faster. That's what an AI agent can do for real business operations.
A fully customized macro system like my original tool could take many people hours to set up. The agent approach takes minutes. It's faster for sure, but speed is the bonus, not the headline. The real gain is the quality of the analysis and the summary reporting.
The new video covers one example of a real business operation, a supplier audit process. However, the same process works for any business process that starts with a raw data export and ends with a report you need to present. And, most importantly, it's done in a way that protects your proprietary data.
The Start: Proprietary Data
The question is how to use an agent to take over some tasks, like the analysis itself, when the file is full of data you can't share. A supplier file holds exactly what a company protects: who it buys from, what it pays, how each supplier performs, and much more.
My answer to this problem is to swap out the proprietary data first with generic data using simple Excel macros. You keep the key that matches the proprietary to the generic so it never leaves your system. Then the agent can go through the masked data at large and do advanced analysis, without the real company names in the upload.
The Workflow: Getting Real Results
When I analyze where companies are with agents right now, I see two camps.
On one hand, you've got companies that are still trying to figure all of AI out. They're trying to be safe, they're trying to use AI, but everything's a little bit disjointed. They have several options for agents, whether it's Copilot or something else, and none of it connects to their data, or very little of it.
On the other hand, you've got teams that have been running agent workflows for quite some time. For them, the major problems are around being very strategic about how you deploy your agents and creating a really streamlined and effective workflow and process.
These two camps drove me to the method I show in the video: one hybrid method that speaks to both. It works on any data set, whether it's been in Excel or exported from another system or a database. And it relies on something very generic, like macros in Excel, to clean out the proprietary data and later put it back in.
Most importantly, this method gives flexibility and security while enabling an agent to do the heavy lifting of analysis, structuring, and preparing the data in the exact way you need it. Then, after the agent is done, we use another macro to quickly swap back in the protected data we pulled out so nothing is lost in the process.
The Hybrid Method: 3 Easy Steps
All data in the video is synthetic. The demo file is built to look like an export from a plant ERP or MRP system, and every supplier, name, and number in it is made up.

1. Mask the Company Names in Excel
I wrote my own simple macros (with the help of AI), each with one discrete job.
The first, MaskCompanyNames, replaces every company name with a generic name like Supplier 001 and writes the key (generic name to real name) on a new sheet called Mask Key. A second one, RestoreCompanyNames, puts the real names back in the file when the agent is done and you're ready to finalize the report.
The critical step: Right after masking, I copy the Mask Key and save it in its own Excel file and then delete it from the data file before upload, so the uploaded file has no key in it. The Mask Key stays with you. It never goes in the upload.
In this example, only the Company Name column is masked. The same steps work on any other column you need to protect.
2. Give the Agent the Masked File and One Prompt
The agent starts with basic instructions. I upload the masked file and give it one prompt: its job, its rules, and the weekly report I want. We're just looking at the masked file, and in this case, it's a supplier audit.
The agent comes back with:
- A "top 5 things I need to know" summary in the chat.
- A multi-tab Excel report: a summary, a supplier risk scorecard, an action list, and more.
- Answers to my follow-up questions on the same data.
Then I ask the agent to set up a weekly routine, so the report is ready to go out to the team first thing every Monday morning. As long as it has the new masked data file every week, it runs the same routine and I start my week ready to go instead of spending hours updating reports.
3. Restore the Names in the Report
To finalize the report, I download it and add the Mask Key back in as a separate sheet. Then, I run the RestoreReportNames macro. The macro swaps the real supplier names back on every sheet, in plain cells and inside sentences.
The key only joins the report after the download, on your machine. As long as you keep the key safe, the real company names never leave your machine.
The Payoff: The Weekly Report
A weekly report on everything that we want to know about is only worth the effort if the team uses it.
The summary tab gives leadership the big picture. The supplier risk scorecard shows where each supplier stands. The action list tells the team what to do next. And with the Monday routine, the report lands at the start of every week instead of whenever someone has time to build it.
The agent does the analysis. You still review the report before it goes out, because you own anything that goes out under your name.
Watch the Walkthrough
The video does the heavy lifting this week. In just under 20 minutes, it shows the macros, the masking, the agent setup, the upload, the report, the follow-ups and the weekly routine, and the restore step that puts the real names into the final report.
Watch it on YouTube: How I Use AI Agents In Business Operations: Raw Data to Report in Minutes
Closing Thought: Start With One Report You Already Build
This method is not a whole new system. It's a scoped, repeatable workflow built on generic Excel macros and an agent that only sees the data you choose to upload. It's also The 7 Agent Rules applied to one real job.
Your action this week: pick one report your team builds by hand from a data export, decide which column you can't share, and watch the walkthrough to see the full loop from raw data to a report ready to present.
If you want more walkthroughs like this one, subscribe to my YouTube channel. I'm building this method in public, one walkthrough at a time.