The marketing spend with no attribution

Six hundred thousand dollars a quarter, four channels, one dashboard. Paid search, paid social, content syndication, outbound. Blended CAC at that company read $480, below benchmark and improving every month. Every channel attributed cleanly. The growth team carried quarterly OKRs against that dashboard and hit them.

Then the Series B lead asked, during diligence, what CAC would look like if paid search went dark for thirty days. Nobody knew. The team had never run the test. The investor made it a condition of the term sheet and funded it. Paid search went off for a month. CAC did not move. Paid search had been capturing customers who arrived anyway through brand search, and the dashboard credited paid search for all of them.

The same investor asked for the same test on content syndication. New pipeline dropped forty percent. That channel had been generating demand the dashboard under-credited, because the journey was multi-touch and the dashboard counted only the last click.

The budget had been miscalibrated for over a year. The team doubled down on the channel that wasn't producing and starved the one that was. Nobody lied. The dashboard was wrong, the dashboard looked clean, and the clean look produced the confidence that stopped anyone from questioning it. It's the same failure as four green metrics hiding a red quarter: the number that reassures is the number nobody audits.

Attribution is the most overconfident data in marketing. Last-click credits whatever touched the customer last, usually a brand-search query from someone who'd already decided to buy. Multi-touch improves on that and stays wrong, because the models assign weights from heuristics, not causality. Dashboards print these numbers to two decimals, and the precision sells an accuracy the method never had. It's the paid-media version of a vanity metric on the homepage, a figure that looks like performance and measures comfort.

The only attribution method that works is testing. Holdout tests, turning a channel off in a defined region or cohort, give direct causal evidence of what the channel does. They cost money up front, because you give up known revenue to learn about future allocation. They're cheaper than the alternative: years spent optimizing against a dashboard with no relationship to causality.

Holdout tests stay rare for political reasons. The manager running a channel doesn't volunteer it for a test that might show the channel isn't doing what the dashboard claims. The CFO would benefit from the test but lacks the marketing context to demand it. So the tests get proposed, deferred, and forgotten.

Some channels can be measured cleanly. Some can't. Measure the first group on a standing basis. Evaluate the second through holdout tests on an annual schedule. Everything in between is a story the marketing team tells itself. Run the test. The number will hurt, and the hurt is the data.