What if the most important touchpoint is one you can't see?

What if the most important touchpoint is one you can't see?

The question: every attribution model — heuristic or algorithmic — credits only the touches it can observe. But word of mouth, a colleague's recommendation, an offline conversation, a podcast mention with no link: these move purchases and leave no trace in your path data. What does an attribution model do with influence it can't see?

The answer it doesn't admit: it silently reassigns that credit to whatever observable touch happened to be nearby. This is omitted-variable bias — the unobserved cause gets absorbed by correlated observed touches. If people who hear about you from a friend then type your brand into search, your model credits 'organic search' for what a friend actually caused. The model isn't wrong about the correlation; it's structurally incapable of seeing the real driver.

Why this is the deepest problem in attribution: it can't be fixed with a better model. No amount of Shapley or Markov sophistication recovers a variable that isn't in the data. The bias scales with how much of your demand comes from dark, untrackable sources — which for strong brands can be the majority. Algorithmic confidence here is false precision.

The nuance: this is exactly the gap MMM and incrementality experiments exist to cover — both can detect the aggregate effect of unobserved demand even when they can't name it, because they measure total outcomes against total spend.

What to actually do: assume a meaningful share of credit belongs to touches you'll never log. Use survey-based attribution ('how did you hear about us?') and MMM as triangulation against the trackable path.

Bottom line for practitioners: the touchpoint that matters most is often the one with no UTM. Attribution models don't measure influence — they measure trackable influence, and quietly hand the rest to the nearest link.