Time-Decay Attribution

Time-decay attribution gives more credit to touchpoints closer in time to the conversion and less to earlier ones. A click on the day of purchase counts for more than a click two weeks before. The credit falls on a half-life curve: after a set number of days the weight is cut in half, then in half again. It is built on the idea that recent interactions carry more of the final nudge, which suits businesses where people research for days or weeks, such as furniture, education, or home services, and the most recent contact usually signals real intent. Think of a ripening curve: a touch loses weight the longer ago it happened, so last week's email counts for more than last month's ad.

The mechanics: each touch gets a weight based on how many days it came before the conversion, and the weights are normalized so they add up to one conversion. A touch on the conversion day gets the largest weight, one a half-life earlier gets half as much, and so on. The half-life is a fixed setting chosen by the tool, not something fitted to your customers. That makes the model sensitive to the number picked: a short half-life behaves almost like last click, while a very long one starts to look like linear. The model can also be unfair to channels that work slowly. Podcasts, video, and content often act weeks before a purchase, exactly where decay gives the least credit.

Availability matters. Google announced that, starting in June 2023, it would stop letting advertisers pick time decay for new conversion actions and, starting in September 2023, would switch conversion actions still using it to data-driven attribution, with last click as the alternative. GA4 stopped offering time decay, first click, linear, and position-based in November 2023. So the model now lives mainly in other analytics products, in spreadsheet exports, or in a business intelligence tool. Meta reports use click and view windows instead. If you move from a time-decay view to GA4's data-driven model, expect credit to move, because the weights now come from observed behavior instead of a decay curve. If you rebuild it yourself in a spreadsheet, test two or three half-lives and see whether the ranking of channels changes. If it does, the conclusion depends on an assumption you cannot verify.

The common mistake is assuming the decay rate fits your business. A buyer who researches for two months and then decides in one day may have been won by the earliest content, which time decay discounts heavily. The model also penalizes top-of-funnel channels by design, so a cut-by-credit decision under time decay will gradually shift budget to retargeting and branded search. Another mistake is applying it to short buying cycles, where all touches fall inside a day or two and the decay barely changes anything. It tends to flatter retargeting and branded search, which cluster near the end, so beware using it to defend spend that sits there. If your sales cycle is short, a day or two, the decay hardly changes anything and you can skip the model. It earns attention only when journeys run long enough for earlier touches to lose meaningful weight.

Use it when you want to test the view that recency matters. Compare a channel's credit under time decay and under last click: if it is similar, your paths are short or the channel closes. If time decay gives much more than last click, the channel often appears a few days before purchase, which is a mid-funnel role. This week, check how long your typical path takes, using days to convert in your analytics tool, and ask whether a decay curve with a half-life that long or short would match what you see. Compare it against data-driven credit when you can. Where both models agree, you can act with more confidence; where they disagree, test.