Attribution is the part of analytics where a tool can quietly make a decision for you. Pick the wrong model and a channel that is doing real work looks like a waste of money, even though the data is the same.
So we made it a setting you choose. Settings → Revenue has two models and a lookback window, set once per project. Together they decide which channel gets the credit for each sale. The totals never change, only how they are split between channels.
Last-touch
The final interaction before a conversion takes all the credit.
This is the default because it is the easiest to understand. If somebody clicked a paid ad and bought ten minutes later, the ad gets the sale.
What it hides is everything that came earlier. Under last-touch, the search visit three weeks ago and the newsletter click last week get no credit, even if they are what put you on the shortlist. Run a business on last-touch alone and the work that brings people in early looks like a waste, because it is never the last click.
Linear
Credit is split evenly across every touch inside the lookback window.
If someone arrived from organic search, came back from a newsletter, and converted on a paid click, each of the three gets a third of the revenue. Channels that helped along the way now get some credit.
The cost is that "evenly" is a claim nobody can defend. A three-second bounce and a fifteen-minute comparison session are not equal contributions, and linear scores them the same.
A typed address never takes the credit
In both models, a direct visit gets no credit. That is someone typing your address or using a bookmark. A person who found you on Google and came back later by typing your address is still Google's sale.
A payment counts under Direct only when no other visit falls inside the lookback window.
The lookback window is the other half of the setting
Both models only see touches inside the window: 30 days by default, anywhere from 1 to 90.
A 7-day window on a product with a two-month sales cycle throws away most of the path and quietly converges on last-touch, because the early touches fall outside it. A 90-day window on an impulse purchase drags in visits that had nothing to do with the sale. Set it near your actual time-to-purchase and check it again when that changes.
Why not show both at once
We considered it, and it makes the page worse rather than better. With two revenue figures per channel, every number would need a note saying which model it uses. One model, chosen deliberately and written down, is easier to run a team on.
Switching is one setting, and reports use the new model the next time they load. You can still compare the two when you need to. A chart read before the switch and one read after are not comparable, so note the date when you change it.
What we did not build
We left out weighted and time-decay models with coefficients we picked. Their weights are always somebody's guess, and you would just have to trust them. We shipped linear as our multi-touch model because its rule fits in one sentence.
See Revenue. For the tools that offer more models, and who each one suits, see marketing attribution software compared.