Back to the blog

Attribution in 2026: why last click lies and what to use instead

3D illustration: Imp inspects four glowing relay batons on a cloud running track with a magnifying glass and scales; only the last baton at the finish tape carries the medal.

Last-click attribution hands 100% of the credit for a conversion to a single click, usually the final one, and that is often a branded search or a retargeting banner. Allocate budget on that basis and you feed the channels that harvest demand while starving the ones that create it. Since Google retired its rule-based models in 2023, GA4 and Google Ads essentially leave you with two options: last click or data-driven.

Last-click attribution overvalues channels at the end of the customer journey, such as brand search and retargeting, because it counts conversions that would have happened without any ad. Use data-driven attribution for day-to-day optimisation instead, validate your biggest budget lines with incrementality tests such as conversion lift or geo experiments, and add a marketing mix model once your spend reaches mid-sized budgets. Attribution distributes credit; only experiments measure impact.

What is left of attribution in GA4 and Google Ads in 2026?

In 2023, Google removed the first click, linear, time decay and position-based models from Google Ads and GA4. In Google Ads you could no longer select them for new conversion actions from June 2023, and from September 2023 the remaining conversion actions were switched to data-driven attribution. In GA4, according to Google, the four models have not been available since November 2023.

GA4 now offers three options: data-driven, "Paid and organic last click" and "Google paid channels last click". One detail matters for your analysis: direct visits receive no credit in any GA4 model, unless the path consists only of direct visits. The default lookback window for regular key events is 90 days; 30 or 60 days can also be selected.

In Google Ads, data-driven attribution is the default model for most conversion actions, according to Google. It evaluates interactions across Search (including Shopping), YouTube, Display and Demand Gen. For good results Google recommends at least 200 conversions and 2,000 ad interactions within 30 days. The default click-through conversion window is 30 days; for view-through conversions it is 1 day.

Why last-click attribution lies

Last click answers the question "Who touched the customer last?", not "Who caused the purchase?". A B2B decision-maker who sees your LinkedIn posts for three weeks, downloads a whitepaper and finally googles your company name shows up in the report as a brand search conversion. Retargeting reaches people who were already sitting on a full basket. Both channels look brilliant on paper because they collect demand that was created somewhere else.

Data-driven attribution is better, but it is not a truth serum. Google describes the GA4 approach as counterfactual: the model compares converting and non-converting paths and estimates how much each interaction changes the probability of a conversion. But it only sees what appears in Google's data. A Meta ad that someone viewed but never clicked does not exist for GA4.

Platform self-attribution: everyone counts themselves rich

Meta, Google Ads and LinkedIn each count conversions within their own windows and by their own rules; Google Ads, for example, defaults to 30 days after a click plus one day after a view. If someone buys after a Meta view and a Google click, each platform may well claim that conversion. Add up the platform reports and you quickly end up with more conversions than your shop or CRM actually recorded. That is why every platform dashboard needs a neutral number next to it, such as the marketing efficiency ratio, calculated from total revenue and total ad spend.

Worked example: how last click inflates brand search and retargeting

Assume an online shop spends €20,000 a month and the average order value is €120. Last click distributes 400 orders like this:

Channel Spend Orders (last click) ROAS (last click) Assumed incrementality Incremental orders iROAS
Brand search €3,000 160 6.4 20% 32 1.28
Retargeting €2,000 100 6.0 25% 25 1.50
Generic search €7,000 80 1.37 70% 56 0.96
Meta prospecting €8,000 60 0.90 Lift test: 150 150 2.25

The formulas behind it:

  • Incremental conversions = conversions in the test group − conversions in the control group, scaled to equal group size
  • iROAS = incremental revenue ÷ spend
  • iCPA = spend ÷ incremental conversions

Going by last click, you would cut Meta and scale up brand search. Going by the assumed test results, it is the other way round: Meta prospecting has an iCPA of €53, brand search one of €94. Many of the 150 additional Meta purchases appear in the last-click report as brand search or direct orders. The numbers are illustrative, but the direction of the error is the pattern: last click overvalues the channels whose audiences are already closest to buying.

Incrementality testing: conversion lift and geo experiments

An incrementality test splits your audience into a group that sees ads and a control group that does not. The difference between the two is the causal effect of your advertising.

Meta Conversion Lift: Meta randomises the target audience into a test group and a control group; the control group is a holdout that sees no ads. The comparison covers conversions, converting accounts and revenue.

Google Conversion Lift: Google offers two variants. User-based lift works for Video, Discovery and Demand Gen campaigns and provides breakdowns by conversion action, age or gender. Geo-based lift also covers Search, Shopping and Performance Max and measures online as well as offline conversions, but according to Google it is noisier and usually needs more budget. Conversion Lift is not available for every account; access runs through your Google representative.

Your own geo experiment: The principle comes from Google's research: non-overlapping regions are randomly assigned to a test and a control condition. For a mid-sized company, that could mean pausing brand search for four weeks in one group of regions or states, keeping it running in a comparable group, and comparing total revenue across both.

MMM as a complement, not a replacement

A marketing mix model estimates each channel's contribution from aggregated time series of spend and revenue, without any user-level data. Google has released Meridian, an open-source MMM that has been available to everyone since January 2025. Meridian is Bayesian, can take incrementality test results as priors and accounts for reach and frequency. That is exactly the sensible division of labour: experiments provide hard point measurements, and the MMM spreads them across all channels and time periods.

Decision tree: which attribution setup for which budget?

The thresholds below are guidance, not a standard. What matters is how many conversions and channels you have.

  1. Under roughly €10,000 a month, one or two channels: Data-driven attribution in Google Ads, GA4 as a neutral point of comparison, MER as your steering metric. If you don't reach the 200 conversions in 30 days that Google recommends, treat DDA numbers with caution. A simple regional brand search pause test gives you your first incrementality data.
  2. Roughly €10,000 to €50,000 a month, three or more channels: Add one incrementality test per quarter for your biggest or most disputed budget line. Feed offline and CRM conversions back cleanly so the platforms optimise for real revenue.
  3. Above roughly €50,000 a month, or a large offline share: An MMM, for example with Meridian, calibrated with your test results. Attribution remains in charge of day-to-day bid and creative decisions; the MMM decides how budget is split across channels.

The foundation is the same at every stage: clean tracking with Consent Mode, unambiguous conversion definitions and a single source of truth for revenue. What that foundation looks like is outlined on our tracking & analytics page, and the budget steering built on top of it is part of performance marketing.

FAQ

Should I switch Google Ads back to last click?

Only if you have too few conversions for a stable data-driven model or need a deliberate comparison. Google recommends at least 200 conversions and 2,000 interactions in 30 days for DDA; below that, the model becomes less accurate. But last click is not a neutral alternative: it simply shifts the error towards brand search.

Is data-driven attribution the same as incrementality?

No. DDA distributes credit for observed conversions across clicks and interactions in Google's data. Whether a conversion would not have happened without the ad can only be answered by an experiment with a control group.

Do I need many years of data for an MMM?

An MMM needs enough variation in spend and revenue to separate the effects. Meridian supports geo-level data, which according to Google carries more statistical information. Feeding in test results as priors further stabilises the model.

Sources

Want more of the attention you get to turn into customers? Let’s spend 15 minutes on your setup, free and without a pitch deck. Book your free call

Keep reading

Let’s take a look.

15 minutes, free, no pitch deck. We look at your setup and tell you honestly where the biggest potential is.

Book a free callFree checkPrefer talking? Call us or send a WhatsApp. +49 163 1695917 · WhatsAppor email us at hello@wiro.design
Free check