What incrementality actually measures

Attribution and incrementality answer different questions. Attribution asks: given a deal closed, how do we split credit? Incrementality asks: given a channel ran, how much extra revenue did it cause?

Both are useful. Neither replaces the other. The mistake teams make is using attribution to justify spend — which is like grading a salesperson on what they were next to when the deal closed.

The counterfactual is the whole game

Incrementality lives or dies on the quality of the counterfactual. The question isn't "what happened" — it's "what would have happened if we'd turned this off." Three ways to estimate it:

• Geo-holdout. Run the channel in some regions, dark in others, for a fixed window. Compare conversion rates. Cleanest signal, hardest to operate.
• Matched-pair synthetic control.For each treated account, find firmographically similar accounts that didn't see the channel. Compare. Works when geo-holdouts are politically impossible.
• Ghost ads. Auction platforms can log impressions that wouldhave been served if you'd bid. Comparing "shown" to "ghost" gives you observational lift without holding back budget.

Why teams skip this

Because it's harder than reading a dashboard, and because the answer is sometimes unflattering. Channels that look good on attribution often look modest on incrementality. Channels that look modest on attribution sometimes turn out to be the ones doing real work.

That gap — attributed minus incremental — is the headline number. We put it next to every channel row by default, because that's the conversation worth having.