Attribution is the rule that decides which marketing touch gets credit for a sale. A visitor clicks a paid ad in March, reads two articles in April, comes back through a newsletter in May, and buys. Attribution is how you answer “what earned that revenue” — and there is no measured answer, only a rule you pick.
That is the one thing to hold onto: attribution is a choice, not a measurement. The same deal, the same data, a different rule, and the revenue lands on a different channel.
The common models
Each model is a different answer to “which touch counts”, and each is right for a different question.
| Model | Credits | Answers |
|---|---|---|
| First touch | The first interaction | Which channel finds new people |
| Last touch | The interaction before the sale | Which channel closes |
| Any touch | Every interaction, split | Which channels are involved at all |
| Linear / time-decay | All touches, weighted | The whole path, not one point |
None is “correct”. First touch overstates awareness channels and hides the ones that close; last touch does the reverse. A channel that never gets the first or last click — a retargeting campaign, a comparison page — looks worthless under both and busy under any-touch.
Why it moves your numbers
Every downstream number that divides by attributed revenue inherits the choice of model. Return on marketing investment is the clearest case: its numerator is attributed revenue, so switching the report from first-touch to last-touch moves revenue between channels and every channel’s ROMI changes — while nothing about the marketing did.
This is why a ROMI, a cost-per-acquisition, or a channel ranking is meaningless without its attribution model stated next to it. Two dashboards can both be “correct” and disagree completely because they answer different questions.
The window is part of the model
Attribution also has a lookback window — how far back before the sale a touch still counts. A 30-day window and a 90-day window credit different touches, and if your sales cycle is longer than the window, the touch that started the deal falls outside it and gets no credit at all. A model without a stated window is only half specified.
How KO handles it
KO computes reports under first-touch, last-touch, or any-touch, and the model is part of the report rather than a global setting — so the same funnel can be read two ways side by side. The point is not to find the one true model; it is to stop reading a single model as if it were the truth.
Where attribution is missing entirely — a lead with no recorded source — KO does not guess a channel. It counts the lead as unattributed and says so, because a fabricated source is worse than a visible gap.