Last-click vs. multi-touch attribution: which should you trust?

Two ways to hand out credit

Say a customer clicks a Facebook ad, reads a blog post on your site a week later, then searches your name on Google, clicks a search ad and buys for $200. Last-click attribution gives the whole $200 to that final Google search ad. Multi-touch attribution splits the $200 across all three touches using some rule.

That's the whole difference. But it has big consequences, because the model you choose decides which channel looks profitable, and most business owners move budget toward whatever looks profitable. If you need a refresher on the basic idea, start with marketing attribution explained.

The case for last click

Last click has real strengths, especially for a small team:

• It is easy to understand. One sale, one source, no math.
• It is easy to check. You can usually see the final click in your own analytics and match it to an order.
• It is stable. It doesn't shift when a vendor updates a hidden algorithm.
• But it is biased toward the bottom of the funnel. Branded search ads and retargeting ads, which reach people who already know you, collect most of the credit. Social, video and display ads that first introduced the buyer often get nothing.

The case for multi-touch

Multi-touch models try to fix that bias by giving every touch some credit. Rule-based versions use a fixed formula. Linear splits credit evenly. Position-based favors the first and last touches. Time decay gives more to recent touches. Data-driven models use software to compare paths that ended in a sale with paths that didn't.

The options have shrunk. In 2023 Google announced it was retiring first click, linear, time decay and position-based models in Google Ads and Google Analytics 4, saying less than 3% of Google Ads web conversions used them. What remains is data-driven attribution, which Google's help center calls the default for most conversion actions, plus last click.

The weakness of multi-touch is that the split is still a judgment. A data-driven model can only see touches it can track, and on a single platform it only sees that platform's ads. You also can't inspect exactly how it reached its answer.

Why neither proves cause and effect

Both models look at what happened before a sale. Neither asks what would have happened without the ad. That second question is the one that matters for your budget.

Research on large ad experiments shows how wide the gap can be. A 2019 study in Marketing Science drew on 15 US advertising experiments at Facebook covering 1.6 billion ad impressions. It found that common observational methods, the kind that underlie most attribution, often failed to produce the same effects as randomized experiments. A 2015 eBay study found brand keyword search ads, a last-click favorite, had no measurable short-term benefit.

So the honest answer to which model to trust is: trust a test. Our guide on testing whether your ads cause sales shows how to run a simple one.

A practical way to choose

For most small businesses, a sensible setup looks like this:

That last check is what Matrix ROI is built around: claimed conversions from each platform beside the sales your Stripe or store records confirm. You can see how it works before deciding whether it fits. The app was made by Smith App Studio, a small studio that builds tools for businesses like yours.

• Use data-driven attribution inside Google Ads if you have enough conversions, since that is what its bidding is built around. Otherwise use last click.
• Pick one model for your own cross-channel reports and keep it fixed so month-to-month changes are real.
• When a channel looks weak under last click, don't cut it straight away. Ask whether it is introducing customers that other channels later close.
• Run a holdout or geo test once or twice a year on your biggest channel.
• Always compare attributed revenue to actual revenue. If attributed totals run well above your real sales, your reports are double counting.