Attribution is the hardest number on the board and the one most worth understanding. This guide is for anyone who has noticed that their platforms collectively claim more revenue than the business earned. It explains how Tatheon reconciles those competing claims, what the different models are for, and how to read match confidence rather than treating it as a score.
The problem being solved
Your platforms do not talk to each other. Each one observes the part of the journey it participated in and reports that part as though it were the whole thing.
The result is that platform reported revenue routinely sums to more than actual revenue, sometimes considerably more. This is not fraud and it is not a bug. It is the predictable outcome of asking several parties to self assess with overlapping windows.
Reconciliation means starting from the orders your store actually recorded and working backwards to the touchpoints that preceded them, rather than starting from what each platform claims and trying to reduce it.
How Tatheon resolves a journey
Tatheon uses a first party session identifier carried by the storefront pixel, plus the click identifiers the platforms append to your URLs.
A visitor arrives with a click identifier attached. That identifier ties the session to the platform and campaign that sent it. The session identifier then persists through browsing, the cart and the checkout, which is where most attribution breaks, and is finally attached to the completed order.
That chain is what turns a claim into evidence. Meta saying it drove an order is a claim. A session that arrived from a Meta ad, browsed, and became order number four thousand and twelve is a record.
Why match confidence matters more than the model
Match confidence describes how much of your revenue Tatheon could tie to an observed journey, rather than how good the attribution is.
Read it before you read anything else on the page. If confidence is low, every model below it is describing a minority of your revenue, and the differences between models are noise. If confidence is high, the models are describing something real and the differences between them are meaningful.
Low confidence is usually a tracking problem rather than a Tatheon problem. The common causes are the pixel not being installed, tracking templates not applied in the ad platforms so no click identifier ever arrives, and a checkout that breaks session continuity. Fixing those raises confidence far more than changing model.
The models, and when each one is right
First touch credits the channel that introduced the customer. Use it when you are asking what creates demand.
Last touch credits the channel that closed the sale. Use it when you are asking what converts existing demand.
Data driven distributes credit across the touchpoints in proportion to how much they moved the outcome. Use it for budget allocation, but only when match confidence is high enough to support it, because it needs enough complete journeys to learn from.
The mistake is treating one as correct. They answer different questions and a channel that looks weak under last touch and strong under first touch is not being measured wrongly. It is doing a job that last touch does not credit.
Reading the gap
Tatheon shows what each platform claims alongside what it reconciles. Expect the platform figure to be higher.
A consistent gap is healthy and tells you how much a platform habitually overstates. Learn your normal. A gap that suddenly widens means fewer journeys are being matched, which points at tracking rather than at performance. A gap that suddenly narrows can mean the same thing in reverse, usually a platform tightening its own window.
What to do with it
Fix tracking until match confidence is high. Then choose the model that matches the question you are asking. Then use the reconciled figure to allocate budget and the platform figure to optimise within a platform.
If you take one thing away, take this: the number that deserves your attention is not which model is showing the highest ROAS, it is how much of your revenue is being matched at all.