Main area
Agency
sea

Understanding Attribution Models

Redaktion ·

Understanding Attribution Models

Attribution answers one question: which click gets the conversion credited to it? Sounds trivial, but it isn’t. A typical customer journey has several touchpoints — someone sees a display ad, later searches for your brand, clicks a search ad, comes back directly two days later and buys. Four touchpoints, one conversion. The attribution model decides how the value of that single conversion is split across the touchpoints.

This isn’t an academic game. The split directly drives two things: your reporting (which channel looks successful) and — much more importantly — your bidding. Smart Bidding strategies learn from the attributed conversions. Credit it all to the last click and the system learns to invest in the last click, underrating everything that warmed the customer up in the first place.

A solid setup builds on clean conversion tracking — without correct conversion data, any attribution model is just math on garbage. For a compact glossary overview, see attribution models.

The classic models

For years, marketers worked with rule-based models — fixed rules that split credit by a fixed scheme:

  • Last Click: the last click before the conversion gets 100% of the credit. Simple, traceable — and systematically biased, because everything before it stays invisible. Details in the glossary under last-click attribution.
  • First Click: the mirror image — the first touchpoint gets everything. Overrates the acquisition channels, ignores the close.
  • Linear: every touchpoint gets an equal share. Seems fair, but unrealistic, because not every contact matters equally.
  • Time Decay: the closer to the close, the more credit. A touchpoint from two weeks ago counts less than yesterday’s click.
  • Position-Based (U-shaped): the first and last touchpoint get the lion’s share (often 40% each), the rest split the remaining 20%.

What they all share: these are assumptions, not measurements. Nobody knows whether “all to the last click” or “40% each at start and end” matches reality. It’s a bet on a rule.

Why Google switched to data-driven

That’s exactly why Google largely abolished the rule-based models. Data-Driven Attribution (DDA) doesn’t split credit by a fixed rule but calculates, with machine learning, the actual contribution each touchpoint made to the conversion. Google puts it this way: DDA “distributes credit for the conversion based on your past data for this conversion action” (source: Google Ads Help, as of 2026).

Simplified, the model compares conversion paths with and without a given touchpoint and estimates its real contribution — instead of setting it by rule of thumb.

The transition was not a suggestion but a shutdown (source: Google Ads Developer Blog, 2023):

  • From June 2023, first click, linear, time decay, and position-based could no longer be selected for new conversion actions.
  • From September 2023, Google automatically switched any remaining conversion actions using these models to DDA.

What this means for bidding and reporting

For Smart Bidding, DDA is the better fuel. Because credit is split more realistically, the algorithm learns to value early and supporting touchpoints too — not just the closing click. Google argues that switching from last click to DDA can help achieve “additional conversions at the same CPA” (source: Google Ads Help). That’s a vendor claim, not a law of nature — but the direction is plausible: reward more than just the last click and you spread budget more broadly and sensibly.

For reporting, the switch mainly means one thing: never compare periods with different models. A campaign looks different under DDA than under last click — not because reality changed, but because the attribution math is different. Keep the model constant when you judge performance over time.

The limits of any attribution

As good as DDA is, attribution only measures what it can see. And that is structurally shrinking:

  • Cross-device: research on the phone, purchase on the laptop. Without a login link, attribution often treats these as two separate users, and part of the path is lost.
  • Cookieless / tracking protection: with the decline of third-party cookies, ITP in Safari and stricter consent rules, all tracking now sees only a slice. Gaps get modeled (Consent Mode, conversion modeling) — modeled means estimated.
  • Walled gardens: what happens on other platforms (Meta, TikTok, YouTube outside Google Ads) is invisible to a Google model. Each platform measures in its own garden and happily credits itself for the success.

In short: touchpoint attribution (often called MTA, multi-touch attribution) is good for steering within one platform but blind to the whole picture.

Distinction: MTA, MMM, and incrementality

Three terms that often get muddled:

  • MTA (multi-touch attribution): splits credit across individual, tracked touchpoints. That’s everything described above — granular, but dependent on seamless tracking.
  • MMM (media mix modeling): a top-down statistical method. Instead of following individual clicks, it correlates aggregated spend per channel with overall revenue over time. It needs no cookies but is coarse and slow to react. In the cookieless landscape, MMM is making a comeback.
  • Incrementality tests: real experiments (geo tests, holdout groups). They don’t measure who “deserves” the credit but the one truly causal question: would the conversion have happened without this advertising? The gold standard, but laborious.

The honest practice: DDA for daily steering, MMM for strategic budget allocation across channels, incrementality tests sampled in for a reality check. No single model delivers the truth on its own.

FAQ

Which attribution model should I choose in Google Ads? For most accounts, data-driven attribution is the right choice — it’s the default for good reason and gives Smart Bidding the best signals. Last click is only a deliberate alternative if you have a very simple, single-step customer journey or have to continue old reporting for comparability.

Can I still use first click or time decay? No. Google abolished first click, linear, time decay, and position-based. Since June 2023 they cannot be selected for new conversion actions, and since September 2023 existing ones were automatically switched to DDA. Only data-driven and last click remain.

How does data-driven attribution work concretely? DDA uses machine learning on your own conversion paths. Simplified, it compares paths with and without a given touchpoint and estimates its actual contribution from that. Instead of a fixed rule, each touchpoint gets a data-based, variable share.

Why do attribution numbers often not match real revenue? Because of the limits of any attribution: cross-device switches, missing cookies and consent, and other platforms’ walled gardens. Attribution sees only a slice of reality and models the rest. For the whole picture you also need MMM and incrementality tests.

What’s the difference between attribution and MMM? Attribution (MTA) works bottom-up on individual tracked touchpoints and is tracking-dependent. MMM works top-down and statistically on aggregated data without cookies. Attribution is granular and fast, MMM is coarse but robust against tracking loss — they complement each other.

Themenuebersicht

Search Engine AdvertisingAttribution