Cost of Marketing

It Launched at 137% COM. Fourteen Demos Is Not a Verdict.

We’ve published our kill bands: investigate any channel over 25% cost of marketing, cut anything over 50%, no exceptions. So here’s an awkward line from our own ledger: in 2024, a new lead aggregator’s first year ran 137.1% — nearly triple the kill line — and we didn’t cut it. The next year it ran 14.5%, inside the target band.

Was that a violation of the rule, or the rule working? It depends on a number most contractors never look at next to COM%: the sample size underneath it.

Why the kill rule didn’t fire

Look at what was actually underneath that 137%: fourteen demos. Five sales at the source’s 35.7% close rate that year. At those volumes, two jobs landing or dying swings the ratio by fifty points — the same source could have printed 90% or 180% on the bounce of a coin flip’s worth of outcomes. A COM% computed on fourteen demos isn’t a performance measurement. It’s noise wearing a percent sign.

The kill bands exist to stop sunk-cost rationalization on operated channels — the ones with enough history that the ratio means something. Our Detroit TV test earned its execution: six months, six figures, a consistent ratio with real volume behind it. A $13.5K pilot in its first year hasn’t yet bought enough data to be guilty. The honest distinction isn’t a loophole; it’s the difference between a verdict and a coin flip, and writing it down in advance is what keeps it from becoming a loophole.

A pilot gets a data budget. An operated channel gets a verdict. Confusing the two kills good sources young and lets bad campaigns die old.

What year two actually proved

With real budget and real volume, the source ran $22.1K against $152.4K — 14.5%, inside the band, at $818 a demo. Same platform that printed 137% twelve months earlier. The launch-year number wasn’t measuring the source; it was measuring the sample. And the discipline cuts both ways, which is why the third column is on the page: 2026 is drifting to 17.5%, and at the volumes the source now runs, that number is real. It closes well — 60% year to date — so it stays funded. It also stays on the leash. Surviving the pilot phase buys a channel a seat at the review, not a pardon from it.

The test before you cut — or keep

Before any kill decision, put the demo count next to the ratio. Under a few dozen demos, you don’t have a verdict in either direction — set a defined data budget (a spend cap and a time box, written down before launch) and let the source finish buying its sample. Past real volume, the bands apply with no appeals: that’s what they’re for. The two most expensive moves in channel management are executing a pilot for its launch-year noise and pardoning a veteran for its glory-year memory. Both are cured by reading the sample size column first. The full framework: COM%, the number that governs a contractor marketing budget.

137% said kill it. Fourteen demos said wait. The next $152K said the demos were right.