Attribution models compared: last-click, data-driven and incrementality
Last-click, data-driven, incrementality — a clear comparison of attribution approaches, what each gets wrong, and how to allocate budget on evidence not illusion.
Why attribution matters more than it sounds
Attribution decides where you put your budget. Get it wrong and you’ll systematically over-fund the channels that merely harvest demand and starve the ones that create it — while believing you’re being data-driven. You don’t need perfect attribution (it doesn’t exist); you need attribution honest enough to allocate budget well. Here’s how the main approaches compare.
Last-click: simple, and usually wrong
Last-click gives all the credit to the final touch before conversion. It’s the default in most tools because it’s simple — and it’s misleading precisely because it’s simple. It over-credits bottom-funnel channels (brand search, retargeting) that were often just the last step of a journey created elsewhere, and it renders your top-of-funnel activity invisible. Optimising to last-click quietly shrinks your funnel.
Data-driven / multi-touch: better, but not truth
Data-driven and multi-touch models distribute credit across the touchpoints in a journey, which is more realistic than last-click. They’re a reasonable default for day-to-day budget allocation. But they’re still models built on observed correlations within tracked journeys — they can’t see everything (offline, cross-device, un-consented), and they still infer, rather than prove, what caused the conversion.
Incrementality: the closest thing to truth
Incrementality testing measures cause directly: you deliberately turn a channel or audience up or down (a holdout or geo test) and measure what actually changes in conversions. It answers the only question that really matters — what would have happened without this spend? It’s more effort and you can’t run it on everything, but used a few times a year on your largest budget lines, it keeps every other model honest and catches channels taking credit they didn’t earn.
How to use them together
Don’t pick one — layer them. Use a data-driven model for everyday allocation; sense-check it with periodic incrementality tests on your biggest lines; and never trust last-click as your optimisation target. This layered approach is pragmatic, affordable and far more accurate than any single model. (It sits inside the broader measurement stack.)
Frequently asked questions
Which attribution model should we use?
A data-driven model for daily allocation, sense-checked with incrementality on big lines. Avoid optimising to last-click.
What is incrementality testing?
Deliberately varying spend (a holdout or geo test) to measure the causal effect — what the spend actually added, versus what would have happened anyway.
Isn’t multi-touch attribution enough?
It’s a good default, but it infers from correlation. Incrementality proves causation and keeps the models honest.
Allocating budget on attribution you’re not sure you trust? Let’s fix it. Book a discovery call →
Leave a Reply