I thought I was building a CRM

I thought I was building a CRM. I was wrong.

There is a pretty common saying in tech circles right now: nobody should vibe-code their own CRM.

And honestly, I mostly agree.

There are plenty of CRMs already. Contact records, pipeline stages, deal fields, reminders, notes, tasks. None of that is new. None of that is especially interesting. Most businesses should not spend their time rebuilding it from scratch.

But after building CRMs for a few businesses I’m involved with, I’ve changed how I think about the category.

The CRM itself is not the point.

The point is the dataset underneath it.

A standard CRM stores what a human remembered to type in. An AI-native system should gather, reason, and act across everything the business is learning: calls, emails, web signals, hiring patterns, customer objections, competitor movement, test results, conference notes, scraped data, purchased data, and the tiny behavioral signals that only matter inside one specific market.

That is where the advantage is.

And it only works if the niche is narrow enough.

Try to become the leading expert on accounting firms and you’ll lose. There are too many firms, too many specialties, too many geographies, too many existing players with a head start.

But accounting firms under twenty people that serve dental practices?

Or mid-sized firms facing a succession cliff?

Or firms that just switched practice management software?

That starts to get interesting.

The narrower the market, the more completely you can know it. You can track who is hiring, who is merging, who is complaining, who changed vendors, who spoke at the conference, what they said, what messaging they respond to, what offers they ignore, and what finally makes them move.

That is not just CRM data.

That is market intelligence.

AI makes this strategy practical for a much smaller operator. What used to require a research team, an analyst, a data engineer, and a sales ops person can now be stitched together by one focused business with the right workflows.

You scan the web and use AI’s deep research functions.

You buy the data that exists.

You capture every customer conversation with AI transcriptions services.

You run small tests.

You feed the results back into the system.

You ask what changed this week that you would have missed.

Over time, the CRM becomes less like a database and more like a learning loop.

That is also why “build vs buy” is the wrong argument. Build it, buy it, customize it, wire tools together. The implementation matters, but it is secondary.

The real question is whether your system is helping you become the number one data owner in your niche.

Because a competitor can copy your product positioning. They can buy the same CRM. They can use the same AI models. They can even copy your website.

They cannot instantly copy years of knowing exactly how your narrow segment behaves, what they have tried, what they complain about, what they ignore, and what finally gets them to act.

That is the moat.

So if I were starting from scratch, I would not ask, “What CRM should we use?”

I would ask:

What is the smallest market we would be proud to dominate?

What signals can we observe before everyone else?

What data can we buy, scrape, capture, or generate through tests?

And how do we turn that into a system that makes us smarter every week?

Do that for long enough and you do not just have a CRM.

You have a position nobody else can easily take.