Why I paid $1,000 for one word

I paid more than $1,000 for a domain: unmodeled.com.

It wasn't planned. I was writing a batch of posts to schedule, and when I read them back, they all had the same thread. I kept writing unsampled, unbiased, unmodeled. And the one that came back again and again was unmodeled data.

When a word shows up in everything you write, it's probably what you actually believe. So I bought it.

What "unmodeled" means

Most marketing data today isn't measured. It's modeled.

A visitor rejects cookies, so the platform estimates what they probably did. A report is too big, so it gets sampled. A conversion can't be attributed, so a model decides which channel gets the credit. Each step looks reasonable. Together they produce numbers that are partly observation and partly someone else's guess.

Unmodeled data is the opposite: what actually happened, counted, with nothing filled in. Clean, complete, and agnostic about which channel, vendor or tool should win.

Why it matters more with AI

Your AI is only as good as the data you feed it. That's it.

Feed an algorithm modeled data and you get a model of a model: the guesses of the first system become the facts of the second. The recommendations sound confident, and they're built on assumptions nobody can check.

Feed it unmodeled, clean data and it has a chance of telling you something true.

That's also why I built Sealmetrics the way I did, and it's the reason it's becoming a data layer for AI. Full disclosure: I'm its founder. Unmodeled is separate. It's where I test what AI actually does to marketing, with data, whoever wins.

What Unmodeled is

Everyone has an opinion about AI. I want the data.

Unmodeled tests the claims being made about AI in marketing and publishes what happened: experiments with a hypothesis, a method and a verdict; benchmarks with the sample behind them; and field notes like this one, labelled as opinion.

It's not another newsletter of tools, prompts and launches. I've just finished the home for it, and I hope the look already says that. Have a look.