Attribution is broken. Here's what we replaced it with.

Every marketing analytics vendor we've worked with — and we've worked with all of them — eventually concedes the same thing in private: attribution is broken.

Last-click attribution is fiction. Multi-touch attribution is more elaborate fiction. Marketing mix modeling is fiction with regression coefficients. Lift studies are honest, expensive, and don't scale.

After three years of trying to make attribution work for our 200+ clients, we stopped trying to fix the model. We threw it out and built something simpler. It's working better than the model it replaced.

What we threw out

We had been doing what every marketing operations team does:

• A "first-touch" report
• A "last-click" report
• A "linear multi-touch" report
• A "data-driven attribution" report (read: opaque vendor model)
• A separate UTM tracker for each major channel

We had four different numbers for the revenue contribution of paid search, depending on which report you opened. Worse, all four were wrong, just in different directions. Worst of all, our clients knew this. We were spending real time defending models we didn't believe.

What we replaced it with

Three measurements, picked specifically because they're hard to lie about.

1. Direct revenue. A transaction that happens on your domain, from a customer whose first identifiable touch is a channel we can name with confidence. No model. No coefficients. The customer typed in your URL or clicked a link with a known UTM and bought. Count it.

This is a small number. It is also a true number. We started here on purpose.

2. Incremental cohort revenue. For paid spend, we built a holdout cohort. A randomly-assigned subset of your addressable audience does not see your paid impressions. We measure their conversion rate against the treated cohort, week over week. The delta is incremental.

This is annoying to set up. It's annoying to explain to a CMO. It is the only number for paid media that we actually trust.

3. Self-reported channel. Every booking, order, and lead form on every Revenue Engine install asks one question: "How did you hear about us?" Free text. Optional.

The completion rate is around 35%. The data is biased — people forget, people round up, people don't list six channels when they used six channels. But it's a real signal that doesn't come from any of our analytics. It correlates surprisingly well with #2 above.

What this looks like in a board deck

We replaced a 14-tab attribution report with one number plus three explanations:

• The number: total attributed revenue this month, defined as the sum of direct revenue above.
• Explanation 1: "Plus we measured incremental lift of $X from paid this month, holdout cohort attached."
• Explanation 2: "Plus self-reported survey data shows N% of new customers cite Y channel; here's the trend."
• Explanation 3: "Here's what we don't know and aren't claiming."

Operators we've moved to this framework universally tell us the same thing six months in: their decisions got faster, their spend got more efficient, and the political fights between marketing and finance went away.

Why this works

It works because it doesn't try to model reality. It measures the small subset of reality that's measurable, and is honest about the rest.

Most attribution vendors try to give you a single confident number for the contribution of every channel. That number is always false. The honest answer is "here are three things we can measure with confidence; here are several we can't; here's how we'll make the decision anyway."

Operators can run a business on that. They can't run a business on a vendor model they don't believe.

What to do this week

If you're spending more than $50K/mo on paid media, you need an incrementality test. Not a vendor pitch about attribution — an actual holdout. We help our Revenue Engine clients run them; you can also do it yourself with patience and statistical discipline.

If you're spending less than $50K/mo, the survey question on every form is the cheapest, fastest signal you can add. Add it today.

If you're spending $0/mo on paid and just trying to figure out where your customers come from — start with the survey question. You'll be amazed how much you didn't know.