Explain how last click and data-driven attribution give credit, and which channels each one tends to favour.
Farah runs a clothing shop and writes its newsletter. A customer paid for a pair of work trousers on a Thursday night, and Farah is sure the email made that sale.
The shop's data shows more than that. Two weeks earlier, the same customer had clicked a Meta ad, viewed the trousers and gone away without buying. A week later, they typed the shop's name into Google and browsed for a few minutes. On Thursday they opened the newsletter, clicked its button and paid. The freelancer who runs the shop's ads would argue that the ad found this customer in the first place. Farah and the freelancer both have a fair case.
Attribution means the rules a report uses to divide credit for a sale or enquiry among the touches that led up to it. If you switch the rule, a different channel can come out on top, so any channel's ranking in a report depends partly on the rule that produced it.
A cheap item bought quickly may have only one touch: the person searches, clicks and pays. Bigger decisions work differently. Someone choosing work clothes, a tuition centre or a sofa will usually come back several times, through different channels, before acting. Lesson 2.1 of Digital marketing foundations: strategy before tactics, "The journey from noticing a problem to telling a friend", followed that journey as the customer experiences it. Attribution follows the same path as a report records it.
GA4 sees each touch as a click or visit that has a source, for example a Meta ad, an organic search result or a link in a newsletter. It keeps the order of those touches for each user, though only while it can track them on a single browser. Once a key event happens, a model decides how the credit is divided.
The last click model hands the full credit to whichever channel brought the person in just before the key event. For Farah's trousers, the newsletter takes the entire sale, and the Meta ad and the branded search get zero.
Its strengths are that it is simple and you can verify it yourself. Its weakness is consistent: it rewards channels that usually turn up late, when someone has already decided to act. In most businesses those are branded search, where people type your name, and email, which goes to people who know you already. Channels that usually come first, such as social ads, display ads and content that puts you in front of strangers, score poorly under last click even while they are bringing in new customers.
There are two versions to know about. GA4's last click option skips direct visits when the user had an earlier source. Google Ads has a separate version that credits only Google's own paid channels.
GA4 uses data-driven attribution by default. It draws on your own property's data to estimate what each touch contributed, where last click follows a fixed rule. In broad terms, it sets the paths of people who completed a key event against the paths of people who did not, and touches that seem to raise the chance of a key event earn a bigger share. Farah's sale could be divided between the Meta ad, the branded search and the newsletter, in proportions set by the patterns in the shop's data.
You can't work the split out by hand, since Google keeps the detailed formula private. It is still a sensible default, because it doesn't assume in advance that one position in the journey counts most. What holds it back is data. A small site with few key events gives the model little to learn from. It can also divide credit only among touches GA4 recorded, and lesson 5.3 deals with the touches it misses.
Older guides list four more rule-based models. First click puts all the credit on the opening touch. Linear splits it equally across every touch. Time decay weights recent touches more heavily. Position based weights the first and last touches more heavily than those in between. In 2023 Google took these out of GA4 and Google Ads, so only data-driven and the last click options remain. An agency report or a course that shows you a linear or first click view from GA4 is using outdated material.
The practical step is to compare models against each other, since there is no single correct one to find. Go to the advertising section of GA4 and open the key event attribution models report (once called model comparison), which compares key event metrics by channel under two attribution models side by side. Data-driven against last click is the obvious pair. Look at which channels pick up credit and which give some back as you move off last click. Then open the key events attribution paths report (once called conversion paths), which shows whether a channel tends to appear early, midway or late in journeys. A channel that gains under data-driven is probably starting journeys, and one that loses credit is probably finishing them.
Farah ran this comparison for one month. Under data-driven, Paid Social took on more credit and Email gave some up. The likely reading is that the Meta ads were opening journeys the newsletter later completed. Cutting the ads because of a weak last click figure would probably have reduced what the newsletter brought in a few weeks later, so check the paths before you drop any channel that only looks poor under last click.
When numbers in GA4 disagree, check where each one comes from. The model set in GA4's attribution settings affects figures in the advertising section and in some other places. The traffic acquisition report from lesson 4.1 ignores that setting and credits every session to its own source. This difference is one reason separate parts of GA4 can show different totals.
Every property has its own pattern. The shift you see when you move away from last click, with some channels gaining credit and others losing it, tells you more about the job each channel does than any one number in a report.
In GA4, compare key events by channel under data-driven and under last click attribution, then note which channels gain credit and which lose it.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).