Attribution models explained: last click, linear, time decay and data-driven

What is the difference between last click, first click, linear, time decay and data-driven attribution, and which one should I use?

A buyer rarely goes from one ad to a purchase. They see an ad, search for you days later, open an email, and then buy. An attribution model is the rule that decides how the credit for that one sale is split among those touches. It never changes how many sales there were, only who gets the credit, and that is exactly why the choice of model can move budget between channels without a single sale changing.

One journey, credited six ways

Take one order worth $1,000 with three touches before it: a click on a Meta ad on day 1, a click on a Google brand search ad on day 5, and a click in a promotional email on day 6, followed by the purchase the same day. This is an illustration, not a real order, and each model splits it like this:

The time decay row assumes credit halves every three days before the purchase; tools use different half-lives, so treat it as indicative. The point is the pattern: the same sale produces a very different ranking of channels depending on the rule.

Each model in one paragraph

Last click. All the credit goes to the final touch before the purchase. It is simple, stable and answers "what closed the sale". It undervalues the channels that brought the buyer in, which is usually prospecting on social.

First click. All the credit goes to the first touch. It answers "what brought this buyer to us" and undervalues everything that happened after.

Linear. Every touch gets an equal share. It is a fair description of the journey and a weak guide to which touch mattered.

Time decay. Recent touches get more credit than old ones. It suits short consideration cycles, where the last days are the decisive ones.

Position-based. Usually 40% to the first touch, 40% to the last and 20% split among the middle. It values both discovery and closing.

Data-driven. An algorithm assigns credit based on the paths of people who did and did not convert, estimating how much each touch changed the probability of a sale. It adapts to your data, and its reasoning is not fully visible.

What the platforms offer now

Google Ads no longer supports first click, linear, time decay or position-based; conversion actions that used them were moved to data-driven, which is the default, and last click is still available. GA4 retired the same four in November 2023 and offers data-driven and last click. Meta works mostly through attribution settings (how many days after a click or a view it credits itself) and has added an incremental attribution setting that estimates which conversions its ads caused, which is Meta's own model of its own effect.

So in practice most advertisers read data-driven in Google, a click-and-view window in Meta, and something else again in their store. That is one of the reasons the numbers never agree, covered in why Meta, Google, GA4 and your store never match.

What no model can fix

A model splits the credit for the sales a platform can see. Each platform applies its own model to its own touches, so Meta can give a sale 100% to its ad while Google gives the same sale 100% to its search ad. Changing models inside each platform does not stop them from both claiming the order. Only a model applied once, across all channels, to the orders your store recorded, can split a sale without counting it twice.

Also, a model is not the window. The window decides which sales a platform counts at all; the model decides how to split what it counts. We explain windows in what attribution windows actually mean.

Which one to use

• For daily optimization of bottom-funnel campaigns: last click. It is stable and it matches how most bidding works.
• To see which channels bring new buyers: first click, read next to last click.
• To describe the whole journey to leadership: linear or position-based, with a sentence explaining the rule.
• Inside Google's products: data-driven is the default and generally sensible there, as long as you remember it only sees Google's touches.
• Before moving budget: compare two models side by side. If a channel looks good under every model, it is good. If it only looks good under one, you have learned something about the model, not the channel.

Models versus measuring impact

Every model distributes credit among touches that happened; none of them tells you what would have happened without the ad. That question belongs to incrementality tests and media mix modeling, which we compare with attribution in attribution vs MMM vs incrementality. For the concept of multi-touch attribution in depth, see our guide (in Spanish). Adray computes last click, first click and linear across all channels from each order's own journey, so a sale is split once and never claimed twice.