Last-touch Attribution

The reporting engine walks backwards from the conversion event and stops at the first touchpoint it finds, handing it the full unit of credit. Most analytics tools apply a last non-direct rule, which skips a session with no recorded source and passes credit to the last identifiable channel before it. Ad platforms use a different variant: they credit their own click whenever one occurred inside their conversion window, regardless of what the visitor did on other channels afterwards, which is why two dashboards can both claim the same sale.

What moves last-touch numbers is rarely campaign quality. Retargeting and branded search sit closest to the conversion by design, so raising remarketing budget raises their credited conversions without necessarily raising sales. The direct-skipping rule shifts credit toward email and organic search, and a shorter conversion window pushes it further toward whatever fired most recently. The advantage is that the figure is reproducible and nobody argues about weights; the cost is that the model rewards proximity to the checkout rather than persuasion.

Last touch stays the right tool for in-platform bid optimisation and for fast, low-consideration purchases where the path genuinely is one or two clicks. Keep it as the operational metric for campaigns whose job is to close, and set separate targets for campaigns whose job is to introduce. In a quiz funnel the completion or the result page usually collects the credit, which is useful when comparing two quiz variants against each other but says nothing about the ads that filled the quiz.

The model degrades as the number of touches per path grows. In considered B2B purchases most conversions arrive through branded search or a direct visit, so last touch mainly records that the buyer already knew the name and stays silent about who taught them. It also cannot detect cannibalisation: pausing a branded search campaign often leaves conversions unchanged because organic listings absorb the clicks. Any channel that does its work early will be undervalued whenever last touch is the only view.

Track credited conversions and cost per credited conversion for each channel, then add assisted conversions: the number of converting paths where the channel appears but is not last. The ratio of assists to last touches tells you whether a channel closes or contributes. A channel with a high ratio is being systematically underpaid by the view you are currently reading.

Watch the share of conversions credited to direct and branded search over time. When that share grows while total conversions stay flat, existing demand is being harvested rather than created. Then run a spend pause on one branded campaign or one region and compare total conversions rather than credited ones; that comparison is the only dependable check on a last-touch conclusion.

Teams cut a channel because its last-touch conversions sit near zero, then watch total conversions fall a month later with no explanation visible in the dashboard. Before pausing upper-funnel spend, check how often that channel appears anywhere in converting paths rather than only at the end. A channel with many assisted appearances and few last touches is doing introduction work that this model was never built to see.

The second error is comparing platform-reported last-click conversions with analytics-reported ones and treating the gap as fraud or breakage. The two use different windows, different identity rules and different reporting dates. Pick one source as the system of record for decisions, keep the other for in-platform optimisation, and stop trying to reconcile the totals. Record the window and rule beside every figure so later readers do not restart the argument.