Demand
Automation8 min read
The Attribution Lie: why your marketing dashboard is lying to you
Every marketing dashboard has a story to tell you about what is working. The story is usually wrong, not because the data is fake, but because attribution is answering a different question than the one you think you are asking.
The question dashboards actually answer
Run the exact same customer journey through last click, first click and multi touch attribution models, and you will get three different "winning" channels from identical data. That is not measurement. That is opinion, dressed up as math and delivered with more confidence than it has earned.
Attribution tells you what happened last. It doesn't tell you what caused anything.
A better question: what happens if we turn it off?
Incrementality testing is less elegant and far more honest. Pause a channel entirely for two to four weeks in one market or segment, and watch what actually happens to inbound demand. If nothing changes, the channel was taking credit for demand that existed anyway. If demand drops sharply, you have found something attribution models routinely miss or overstate: real, causal lift.
It is a blunter tool than a beautifully colour coded dashboard. It is also the only one of the two that is actually telling you the truth.
What we track instead of a single blended number
- Leading indicators per stage of the growth system, rather than one blended "ROAS" figure that hides which stage is actually doing the work
- A quarterly holdout test on at least one channel, built around the deliberate question, "what if we turned this off?"
- Marginal cost of the next customer, not the average cost of all past customers combined
- Time from first touch to close, as a proxy for pipeline health that no attribution model can quietly reassign
The businesses that scale well aren't the ones with the most sophisticated dashboard. They're the ones who trust that dashboard the least, and test reality instead of modelling it.
Frequently asked questions
Is multi touch attribution better than last click?+
It is less biased toward the final step, but it is still a modelling choice, not a measurement. Different multi touch models, including linear, time decay and U shaped models, will still assign different credit to the same journey.
How long should an incrementality test run?+
Two to four weeks is usually enough to separate real signal from normal week to week noise, though longer sales cycles may need a longer holdout to capture the full effect.
Isn't turning off a channel risky?+
Running it on one segment or region rather than everywhere limits the downside while still giving you a real answer. The point is a controlled test, not switching everything off blindly.
ScaleKraft Team
Automation & Reporting