Attribution Model.

Hardly anyone buys at the first contact. Between the first search and the purchase there are often several visits through different channels: an organic search, later an ad, then a direct visit. The attribution model determines how that one conversion is distributed across this chain. It is therefore an accounting rule, not a measurement. Change the model and the results shift between channels while the total stays the same. That is exactly why channel comparisons from two different tools are almost never directly comparable.

The common models: last click credits everything to the last contact and systematically favours channels close to the conversion, such as brand search ads. First click, conversely, favours the channels at the start. Linear attribution distributes evenly, time-decay models weight later contacts more heavily, and position-based models give the first and last contact a fixed share each. Data-driven attribution derives the weighting from the observed difference between converting and non-converting paths and is now the default in Google Ads.

An example: under last click, a B2B provider sees that almost all conversions come from people typing in the domain directly and from brand search ads. The logical conclusion would be to cut the budget for generic terms. A position-based model shows a different picture: a large share of the chains begins with a generic term, followed weeks later by a brand search. The same behaviour, two opposite recommendations, purely because of the chosen rule. Anyone who knows only one model makes budget decisions on half an observation.

The honest limit: attribution shows correlations, not causes. No model answers the real question of whether a conversion would have failed to happen without a particular channel. Then there are practical gaps: switching between devices, declined consent, short lookback windows, conversions by phone or in a shop. The question only becomes reliable through an experiment: switch off a campaign for a defined period or region and observe overall demand. That is uncomfortable and costs revenue in the short term, but it is the only way from the report to causality.