Last click lies: what attribution modelling is and why it matters
The same picture keeps showing up in reports: search ads seem to bring in most of the sales, and the Instagram row looks nearly empty. The decision looks obvious — cut the weak line, move its budget to the strong one. A few months later total sales don't grow. They drop, even though the "useless" channel is already off.
The problem was never the ad. It was the measurement. The report was telling you who clicked last before the purchase, not who brought that person to you first. The gap between those two questions has a name — attribution modelling — and it decides where the budget flows, usually without anyone questioning the model itself.
"Instagram isn't working" — you're actually shutting down discovery
A typical path looks like this: someone sees your Instagram story ad, doesn't click, just remembers the name. A few days later they search your brand on Google, click the search ad, and buy. The standard last-click model assigns 100% of that sale to search. Instagram's row shows zero, even though without that story ad the search would never have happened.
This isn't a marketer's mistake — it's how the report is built. Last-click gives full credit to the final click before conversion and discards every step before it. Google's own explanation of the logic is that the path to a conversion usually involves several ads, but only the last one gets counted (source).
Do this week: open GA4's Advertising → Attribution → Attribution paths report. It shows the whole sequence of touchpoints before conversion, not just the last-click column — including where Instagram actually sits in that sequence.
Why last-click exists at all — and where it lies
Last-click stayed popular for one reason: it's simple. One conversion, one channel, one number nobody argues with in a meeting. But that same simplicity makes it structurally biased — by design it rates "closing" channels (branded search, retargeting, email) higher than "opening" ones (social ads, display, influencers, content), because a closing channel is almost always the last touch by definition.
The result: the report names the same "winners" every month, because the model is built to name them winners. Splitting budget by those numbers is like reinforcing the roof every single time and never touching the foundation.
Do this week: sort your channels into two columns — openers and closers. When you read a last-click report, consciously discount the openers' numbers and inflate the closers' — the real picture sits somewhere between the two.
What attribution modelling actually is, and the options you have
Every attribution model answers the same question differently: how should credit for a sale be split across several touchpoints. You could give it all to the first touch, split it evenly, or weight the first and last touch more heavily — mathematically, any of these is a valid rule.
Google Analytics 4 narrowed the choice down to three models: data-driven attribution, paid and organic last click, and Google paid channels last click (source). The last two are both versions of the same last-click logic, so everything above applies to either one.
Example: how many touches sit behind one purchase
A real purchase is almost never the result of a single touch. Say the same customer met you in this order: first saw your Instagram story ad and didn't click, a few days later found your blog post through organic search and read it, a week after that ran into a retargeting banner, and finally searched your brand name on Google, clicked the search ad, and bought. Four different channels, four different roles.
First-click would hand all the credit to Instagram, because it started the chain. Even distribution would split it evenly across all four. Last-click — the most commonly used model today — assigns the full credit to the search ad alone, even though it was only the last link in the chain. None of these models is "wrong" on its own; each answers a different question. The problem is that most accounts show only the last one by default.
Do this week: check which model is actually applied to your key events (Advertising → Attribution → Attribution models). Most accounts are still running the default setting nobody has ever opened.
Data-driven attribution: what the algorithm actually does
Data-driven attribution doesn't apply a fixed rule — machine learning compares paths that ended in a conversion against paths that didn't, and calculates each touchpoint's real contribution to the result. Google describes the factors it weighs as "time from key event, device type, number of ad interactions, the order of ad exposure, and the type of creative assets" (source).
The practical consequence: the answer shifts along with your own account's data, so you can't set it once and forget it — as the account grows and new channels get added, the model recalculates credit again.
Do this week: if your account has enough conversion volume, keep data-driven attribution switched on and compare it against last-click over the same period. The bigger the gap, the more skewed your current budget split has been.
Reallocating the budget — a concrete step
The loop closes like this: the opening channel's budget gets cut because last-click marks it "weak"; discovery drops; a while later the closing channel runs dry too, because there's less demand left to close. None of that shows up in a single month's report — it surfaces three or four months in, by which point the cause is hard to trace back to its effect.
The concrete rule: only cut a channel once it's missing from both the last-click column and the attribution paths report. If it still shows up in the paths report, it's doing work last-click simply can't see. If you're still unsure, the cheapest test is to switch the channel off for a week or two and watch total sales, not just its own row.
The channel last-click rewards is usually not the one that created the sale — just the one that closed it.
Running this by hand every month takes time, and setting the model up correctly and turning the result into a real budget decision takes even more attention than that. Our attribution modelling service exists for exactly that — we read the real conversion paths in your account, show where each channel actually earns its keep, and turn the budget split from a guess into a number.
