How I Built an AI Brain From 8 Million Lines of Customer Emails

Not long ago, I had a problem. I manage staff at a real estate company that operates in 46 states, and every day my team would ask me the same kinds of questions — how do you handle an LLC on a listing, what do you do when the seller's name doesn't match the deed, what are the MLS rules in this state versus that one. I was the answer machine. If I wasn't around, things slowed down or someone guessed.

It was a burden on me, because it kept me from doing the tasks that mattered. It also frustrated me that they were the same questions over and over again. In real estate, there is a lot of nuance, so every scenario is different. Employees didn't want to make a mistake — and rightfully so. Real estate is highly regulated, and certain mistakes can become regulatory issues. None the less, it was hard on me.

We also had an agreement with a big box store on the horizon, which would mean an influx of business and new hires to help keep up. And who do you think would have to train them? This guy!

Basically, what I needed was my brain in the cloud — so I actually didn't have to answer employees or clients anymore.

The Brain collapses that timeline. A new hire can ask it the exact same questions they'd otherwise be asking a manager or a senior coworker — and get the same answer, instantly, without taking up anyone else's time. No waiting for someone to be free. No "let me check and get back to you." It's like handing them a coworker who's been there ten years and never gets annoyed at being asked something twice.

I figured: all of this knowledge already exists. Aside from what's sitting in the heads of a couple of people, it's just buried in years of old support messages inside our portal. So I decided to see if AI could organize it into something my team could ask questions to directly — instead of asking me.

Our company runs on something called MySQL — which is just a type of database, basically a giant organized filing cabinet that software uses to store things like messages, customer records, and notes. Every business has some version of this, even if you've never heard the term. It might be MySQL like ours, it might be your Outlook inbox, a CRM system, a folder of old support tickets, or even employee training manuals sitting in a drawer. Wherever your business's knowledge lives right now, the approach below works the same way.

Here's the catch. The full message archive was a single file with 8 million lines in it. I couldn't even open it. My laptop choked just trying to load it. Keep in mind, real estate as a whole is more complex than most businesses — we have 20 years of messages, and a typical HVAC or construction company wouldn't have accumulated this type of complex data.

I'm not a programmer. I had never written a line of code in my life. But I asked a couple of AI tools what my options were, and they both suggested the same thing: use a Python script to filter the file down before trying to do anything else with it.

Python is just a simple programming language that's great at handling big files and repetitive tasks — think of it like a very obedient assistant that does exactly what you tell it, as many times as needed, without complaining. You don't need to know how to write Python yourself. You just need to know how to ask AI to write it for you.

(Quick aside: Python isn't just useful for massive data files. You can ask AI to write you a simple Python script that cleans up your computer desktop — automatically sorting old files into folders by type, deleting duplicates, or archiving anything you haven't touched in over a year. Same idea, much smaller scale. Something like: "Write me a Python script that goes through my Desktop folder and moves any file older than 12 months into a folder called Archive, sorted by file type." AI will write it, explain it, and walk you through running it in about five minutes. Plus many more incredible things you would never think of.)

Back to the 8 million lines. I typed something like this into Claude and ChatGPT:

"I've never used Python in my life. I have eight million lines of messages and I need to remove the ones that aren't useful — things like 'Thank you,' 'You're welcome,' short one-liners that have no real information in them. Can you write a simple Python script that does this, explain what each part does in plain English, and walk me through how to run it?"

That's it. The AI wrote the script, explained every line like I was five, and told me exactly what to do step by step. I didn't understand most of it — and I didn't need to. I just followed the instructions. I did walk away with some understanding, which is exciting.

A few passes later, I'd cut 8 million lines down to about 58,000 — stripping out the noise and leaving only the real stuff. From there, I could actually read what was left and pull out the knowledge that mattered.

I sorted it by the topics my team asked about constantly — ownership verification, trust and LLC paperwork, MLS compliance, status changes, photo and copyright issues — and fed it to the AI in chunks, building it up one category at a time. This isn't just for old emails either. Employee training manuals, SOPs, onboarding docs, even handwritten notes you've typed up — anything that captures "how we do things here" can be fed in the same way. Eventually ChatGPT told me to stop sending the specific categories I was sending because it became redundant.

Now, one more thing worth explaining: where does this actually live?

A lot of companies build what's called a "self-hosted LLM" — basically their own private version of an AI model running on their own servers that nobody else can touch. It's powerful, but expensive and complicated to maintain, and honestly overkill for most businesses including mine.

I don't have one of those. What I built instead lives inside ChatGPT itself, as something called a Custom GPT. Think of it like a special pre-loaded version of ChatGPT that already knows everything I taught it — my company's rules, our procedures, the categories I built. Anyone on my team with the link can open it and start asking questions immediately. No servers, no IT department, no extra software.

That's the part that made this realistic for a business my size. You don't need a tech department or a six-figure budget. You need ChatGPT and a willingness to feed it the right information. And Claude or ChatGPT will show and build you the exact documents to feed it. The hardest part was discarding the useless messages. And this isn't locked to ChatGPT specifically — the same idea works with Claude, Gemini, or whatever AI tool your business already uses. The point isn't the brand of AI. It's that you're handing it your own business's knowledge instead of relying on what it already knows about the world.

What came out the other end is something we now call the CREG Brain. In its own words, here's what it does:

"I was created to help CREG staff make faster, safer, and more consistent decisions on listing issues. My main purpose is to think like an experienced CREG employee: verify ownership, confirm signing authority, identify what documentation is needed, apply MLS rules, reduce liability, and help draft customer-facing responses that protect the company."

Now, instead of staff guessing or waiting on me, they ask the Brain. It doesn't get tired, it doesn't forget, and it's consistent every single time.

I want to be upfront: this wasn't a 20-minute project. It took real time, some trial and error, and yes, a little bit of code I didn't know how to write before I started. But every step of it was something AI walked me through — I just had to be willing to try.

One last thing worth knowing: the Brain isn't frozen in time. Everything it knows lives in something called a markdown file — which sounds technical but is really just a plain text document, like a Word file with no formatting. When I notice a gap or something it got slightly wrong, I just add the missing information in plain English and re-upload it. No coding, no rebuilding from scratch. It's closer to editing a Google Doc than touching software.

If you want the full breakdown — exactly what I asked the AI, the prompt I used, and how I structured the whole thing — subscribe below and I'll send you the detailed version.

[Subscribe to get the full step-by-step PDF]

Ask AI to interview your own business.

Before you can build any kind of knowledge base — even a simple one — you need to get what's in your head out onto paper. Most people skip this step and wonder why the AI doesn't "know" their business well enough.

Here's the fix. Open whatever AI you use and paste this:

"I'm going to describe how my business runs. Ask me one question at a time — don't give me a list, just ask them one by one — until you have enough to write a basic training document for a new employee. Start with the most important question first."

Let it interview you. Answer in plain language. When it's done, ask it to turn your answers into a simple document. That document becomes the foundation of your own business brain — and you can upload it to a Custom GPT, share it with new hires, or just keep it as a reference.

Takes about 20 minutes. Worth more than most employee handbooks that took weeks to write.

Most small business owners say AI is actually getting more useful. A new survey of over 500 small business owners found that nearly 4 in 5 say AI is more useful to them than it was a year ago. The number one use? Research and brainstorming — not writing content like most people assume. Owners are using it to think, not just to produce.

Can you ignore it? No — but in a good way. This is useful signal: if you're still treating AI like a fancy Google search, you're leaving the most valuable part on the table.

Only 18% of small businesses have fully adopted AI. As of late 2025, about 18% of U.S. businesses are using AI in their core operations — up from just 4.6% two years ago. That's fast growth, but it also means 82% of your competition still hasn't figured this out yet.

Can you ignore it? No — the window where being an early adopter gives you an edge is still open, but it won't be forever.

This week: ask AI to write your business's origin story — as a dramatic movie trailer voiceover.

Paste this into any AI:

"Write a 30-second dramatic movie trailer voiceover for the origin story of my business. Make it sound like the fate of the world depends on it. My business is [describe what you do in one sentence]."

Then read it out loud in your best movie-trailer voice.

Share it with your team. Post it somewhere. The more ridiculous the better — and honestly, some of these end up being the best "about us" copy you've ever written by accident.

That's it for this issue. One thing to try before next time: run the "interview your business" prompt above and see what comes out. You might be surprised how much you know that nobody else has ever written down.