Ecommerce Attribution Models: Which One Should You Use?

No attribution model is correct in every case. Last click is simple and consistent but ignores earlier touchpoints. Data-driven attribution, Google's recommended default in GA4, shares credit using your conversion paths but only sees tracked touchpoints. Rule-based multi-touch models (first click, linear, time decay, position based) are no longer available in GA4. For decisions about large budgets, use incrementality experiments, and at scale, marketing mix modelling. Pick one model for trend reporting and test the big assumptions.

An attribution model is a rule for dividing credit. If a customer clicks a social ad, then an email, then a search result before buying, the model decides whether the email gets all the credit, none of it, or a share. The model doesn't change what happened; it changes which channels look successful in reports, and therefore where budget goes.

This article compares the models. For the wider practice of attribution, including tracking, reconciliation and surveys, see ecommerce attribution.

Google Analytics 4 changed its attribution options. According to Google's documentation, the available reporting attribution models are data-driven attribution, paid and organic last click, and Google paid channels last click. First-click, linear, time-decay and position-based models were deprecated and removed (Google Analytics Help). Google recommends data-driven attribution as the default (Google Analytics Help).

Rule-based multi-touch models still appear in some other analytics and marketing tools, and they can be built in a warehouse from path data. Check your own tool's current documentation, as these options change.

Last click remains useful because it's predictable. It answers "what was the final tracked step before purchase?" and gives stable week-to-week comparisons. Its bias is well understood: channels that introduce customers (social, display, content) look weaker, and channels that close (branded search, email, retargeting) look stronger.

Use last click for operational reporting and quick checks, knowing the bias. Don't use it alone to cut discovery channels.

Data-driven attribution compares the paths of users who converted with those who didn't, and estimates how much each touchpoint increases the probability of converting. It adapts to your data and usually gives discovery channels more credit than last click.

It still works only on tracked, consented touchpoints, and it's correlational: a channel that tends to appear in converting paths gets credit even if those customers would have converted anyway. It's also less transparent, so explain to stakeholders that it's a model estimate. See customer journey analytics for path analysis.

Incrementality experiments measure what attribution models can't: the sales that wouldn't have happened without a channel. A platform lift study randomly withholds ads from a control group; a geo test compares regions with and without spend; an email holdout withholds a campaign from a random group. The difference between groups is the incremental effect.

Experiments cost money (you give up some exposure) and time, and they need enough volume. Use them for the channels where spend is highest or where attribution is most suspect, such as retargeting and branded search.

Marketing mix models estimate each channel's contribution from aggregate time series (spend, sales, pricing, promotions, seasonality). They can include offline media and aren't limited by user-level tracking. They need enough history and variation in spend, and their assumptions (how long advertising effects last, how returns diminish) strongly influence results. Calibrating MMM with incrementality experiments improves trust in its estimates.

An illustrative scenario, not a client case: under last click, a store's retargeting campaign shows the highest return and paid social the lowest. Switching the primary view to data-driven attribution moves some credit to paid social. A two-week geo holdout on retargeting then shows a smaller incremental effect than either model suggested. The team shifts part of the retargeting budget to prospecting and schedules a follow-up test.

Model choice isn't the only setting that moves results. The lookback window (how long before a conversion a touchpoint can receive credit) changes which channels get credit; shorter windows favour closing channels. Whether view-through conversions are counted, and which events count as conversions, also matter. Record these settings with the model, and keep them stable when comparing periods.

Teams with path data in a warehouse sometimes build their own rule-based or statistical models, for example to keep a first-touch view for acquisition reporting after it left GA4. This gives control and transparency, but it inherits the same tracked-only blind spot and needs maintenance. Document the logic, compare results with your analytics tool's model, and still validate major decisions with experiments. See ecommerce data warehouse.

Budget owners often want one number per channel. A clearer way to present attribution is a range: the credit under last click, under data-driven attribution and, where available, the incremental effect from a test. When all three agree, confidence is high. When they diverge, that's where a test should go next. This framing prevents endless arguments about which model is right. See ecommerce dashboard design.

Attribution models divide credit; they don't prove cause. Choose one model for consistent reporting (in GA4, usually data-driven), understand its biases, and check large decisions with experiments or MMM. Related: ecommerce analytics, customer analytics and KPI dashboards.