Last Click vs. Data-Driven Attribution: Which Should You Use?

Last Click vs. Data-Driven Attribution: Which Should You Use?

GA4 now offers three attribution models: data-driven, paid and organic last click, and Google paid channels last click. If you run more than one or two channels, report on data-driven attribution: it shares the credit for each sale across the clicks that led to it, where last click hands all of it to the final one. Keep last click next to it as your baseline. Neither model tells you what a channel caused, since both share out credit for sales GA4 already tracked, so test a channel before you move real budget on either.

GA4 dropped the first click, linear, time decay and position-based models in November 2023, so in GA4 this is now the whole choice. It decides which channels look profitable in your reports, and so where next month's budget goes.

Understanding Last Click Attribution

Last click is the oldest and simplest attribution model: 100% of the credit for a sale goes to the last touchpoint before the purchase. If a customer sees a Facebook ad, later clicks a Google search ad and then buys, the search ad gets all of it. The rule is simple: the final interaction is treated as the only cause.

Its appeal is that it's easy to understand, set up and report on. Many standard reports, such as the session-level acquisition reports in Google Analytics, still work on a last non-direct click basis. It needs almost no data, usually just the last referrer before the sale, which is why most stores start with it, and the reports come quickly, even if they only show the surface. With a short sales cycle and few touchpoints, it can even look good enough.

That simplicity is also the problem. Last click ignores everything earlier in the customer journey, so the channels that introduced the buyer or brought them back get nothing. A social campaign can introduce someone to a product, and if they later buy through a branded search, the social campaign gets no credit. Over time the report pushes budget toward bottom-of-funnel channels and away from the ones that start journeys, simply because those rarely get the last click. It assumes one straight path to purchase, which is rarely how people buy.

Take a typical journey for a beauty brand. A customer finds a new product through an Instagram influencer post (organic social), sees a retargeting ad on Facebook (paid social), searches Google for reviews (organic search), clicks a discount link in the newsletter (email marketing), and finally clicks a brand search ad to buy (paid search). Last click gives the brand search ad all of the credit, and the post, the Facebook ad, the organic search and the email nothing.

Exploring Data Driven Attribution

Data-driven attribution (DDA) swaps the fixed rule for a model. Instead of handing everything to the last click, it uses statistical models and machine learning to estimate how much each touchpoint in the path contributed. GA4's version looks at converting and non-converting paths, and weighs factors such as the order of the ads and the time before the sale.

Underneath, it uses your history to estimate how likely a sale is after a given sequence of interactions. Common methods include Shapley values, Markov chains and the ad platforms' own algorithms, such as Google Ads'. They compare paths that include a touchpoint with paths that don't, or look at the chance of a sale at each step. For illustration: a model gives an awareness ad 20% of a sale, a mid-funnel article 30% and a final retargeting ad 50%. The split is recalculated from your data as it changes.

The main benefit is that it gives some credit back to the channels last click undervalues. Display and content build awareness and demand, and get little or nothing under last click; DDA gives them a share of the sales they appear in. That's a more balanced picture of the whole funnel, not just the moment of purchase.

It also updates itself. As new data comes in, the model re-splits the credit, so it keeps up when your channel mix or your customers change. The price is that it needs more data, and more skill to set up and read, than a one-line rule.

Comparing Last Click and Data Driven Attribution

Side by side, they differ on almost every line:

Read the table as a trade. Last click is cheap, and its bias is predictable: it favors whatever closes. Data-driven costs more to set up and read, and its bias is harder to see, because it depends on what GA4 tracked and on who was going to buy anyway.

The biggest difference is how the credit is handed out: all or nothing under last click, shared under DDA. That flows straight into budget. A last-click report rewards whichever channel tends to take the final click, so budget drifts toward it and away from the awareness and consideration work that feeds it. Reported conversions can hold up for a while as the top of the funnel runs dry. DDA spreads the credit, so the report stops punishing the channels that start journeys.

Take a fashion brand reporting on last click. Branded search looks like its best channel, because it catches people who already know the brand and are about to buy, while the influencer and social campaigns that introduced them look weak. DDA gives those early touchpoints a share of the credit, so they're less likely to be the first thing cut.

DDA also shows more of the path. Last click tells you what the final interaction was; DDA estimates what each step added, which helps you spot where journeys stall and which channels work together. For illustration, it can show that people who read a certain blog post buy more often when they later get an email, which last click can't show. It still shows which steps go together, not which one caused the sale.

When to Use Last Click Attribution

Last click still makes sense in a few cases. The risk is keeping it only because it's the report that was already open.

The first is a store with one or two channels. A local bakery running only Google Ads for online orders has a short, direct path to purchase, and the last click is probably the interaction that mattered. Last click gives that store a quick first read without any analytics setup, and building and maintaining a data-driven model would cost more effort than it could pay back.

The second is a short, purely direct-response push: a flash sale sent in one email or SMS, where you expect people to buy straight from the link. The goal is a click that turns straight into an order, and earlier touchpoints matter little to that goal, so for that campaign on its own, last click is a clear measure. Even then, the brand work last click ignores is part of why people opened the email at all.

The third is as a baseline. Last click is simple enough for anyone on the team to read, which makes it a useful fixed point to compare other models against. Keep it running next to whatever you move to: when a new model credits a channel very differently, last click shows you how big the change is. As you add channels it stops being enough on its own: across several channels, it keeps crediting whichever one closes.

When to Use Data Driven Attribution

If you sell across several channels, use data-driven attribution for your day-to-day reporting. It gives a more detailed view than last click, and that matters for where the money goes. In GA4 it's a setting, not a project: you choose the reporting attribution model in the property's attribution settings.

The main reason is a multi-channel mix. Customers meet a brand through social ads, search ads, email, organic search, content and influencers, and rarely in a straight line. Last click hands the whole sale to the final interaction and ignores how the earlier ones worked together. DDA splits those journeys up and gives each channel a share based on what it seems to add, so you see more than the tail end of the funnel.

Whatever you spend, budget moved on last-click numbers compounds: this month's misread becomes next month's plan. Data-driven gives a fuller picture of which channels contribute along the way, though, as the next section explains, contributing is not the same as causing. For example, DDA can show that branded search still converts well while display does more to introduce new customers than last click suggests. That lets you rebalance from the channels last click overcredits toward the ones it undercredits.

It also helps when winning a customer takes several touchpoints over weeks. That's common in beauty, fashion and supplements, where the first order is often not the last and customer lifetime value matters more than any single sale. Last click only sees the final step, so it says little about the channels that bring people in. DDA credits the earlier steps too, which is closer to how those customers were won.

Privacy limits hit both models. Browsers restrict cookies and each platform keeps its own data, so GA4 sees fewer of the touchpoints either model can credit. DDA is less fragile than a single last touch because it spreads credit across what it does see, but it can't credit a touchpoint GA4 never recorded.

The Limitations of Traditional Data Driven Attribution

DDA is a real improvement on last click, but it shares last click's root problem: it's built on correlation, not causation. Even a well-built model shares out credit for sales that happened; it doesn't test what would have happened without each channel.

Whether they use Shapley values or Markov chains, these models find patterns between touchpoints and sales. They answer: how likely is a sale after this sequence of touchpoints? A channel that shows up in lots of converting paths gets lots of credit, but that doesn't prove it caused those sales. It may just be where people who were going to buy anyway tend to pass through. Someone who already wants the product clicks several ads on the way to buying it, and the model credits those ads, when the real reason was a friend's recommendation or an urgent need that no tag can see.

So budget can still go to the wrong place with DDA. If a channel is always in the path but isn't driving extra sales, spending more on it won't return what the report promises. The model describes what happened in the paths; it can't tell you why. And the budget question is a causal one: if I spend more on channel A, how many additional sales do I get? DDA isn't built to answer that.

Say a supplement brand runs a broad display campaign, and DDA shows that people who saw the ads convert more often. That can mean the ads worked, or that the targeting found people already shopping for supplements who would have bought anyway through another channel. The report looks the same either way. Without a causal test, the brand can keep pouring money into display for very little extra revenue. Every model built on observational data alone has this problem.

DDA is also only as good as the tracking under it. Gaps in the data, broken tags or touchpoints it never sees skew the split: if consent choices or tagging drop part of your sessions, the model shares the credit among the touchpoints that survived, and nothing in the report shows the gap. It also needs plenty of history to train, which is hard for newer stores or ones that change campaigns often. It beats last click, but without causal inference the budget question stays open.

Beyond Attribution: The Need for Causal Inference

Both models describe what happened: the sequence of touchpoints, the final click. Neither explains why a customer bought, or whether they would have bought without a given ad. That's the question causal inference is for.

Causal inference is the statistics of cause and effect. Instead of finding correlations, it estimates the incremental effect of an action: not "this channel was in many converting paths" but "this channel caused X additional sales." It can be applied to a channel, a campaign or a single ad, given the right data or test.

For a store, that's the question that turns attribution into decisions. Not "which touchpoint came last?" but "if we cut this channel, what revenue would we lose?" Attribution models can't answer that, and correlation doesn't pay the bills. Causal inference measures the uplift: the additional sales that wouldn't have happened without the marketing.

The trouble is observational data. Customers aren't randomly assigned to touchpoints: people who click retargeting ads were already more interested than people who scroll past a display ad. That selection bias makes it very hard to isolate what any one touchpoint did from correlations alone. Causal methods such as uplift modeling, synthetic control and randomized controlled trials (A/B tests) get around it by estimating the counterfactual: what would have happened if the marketing had not run?

For a beauty brand, knowing that an influencer campaign caused new customers, over and above those who would have come anyway, is what lets you scale it with confidence and cut the campaigns that only collected clicks. Without that, a campaign that harvested people who were already coming looks just as good in the report as one that brought in new customers.

Causal Attribution for DTC eCommerce

For a store, the causal question is usually about channels: what did Paid Social, Paid Search, Email or Organic Search cause, next to what last click gave each of them? You can estimate that from the data you already have, or test it by changing spend in one place and not in another.

Put the two side by side and the comparison does most of the work. Where they disagree, you've found the channel your reports get wrong, and that's where a test pays off.

The estimate route uses history you already have: a causal model reads it and estimates, channel by channel, what revenue would have looked like without each one. It covers every channel at once and pauses nothing, but it's still an estimate, so treat a big surprise as a reason to test that channel, not as a verdict.

The test route is a regional holdout, the simplest causal test you can run yourself. Pause the channel in one region for two weeks and keep everything else the same. Then compare that region's revenue with the regions where the channel kept running, each against its own two weeks before the test. If revenue in the paused region held up, the channel was getting credit for sales that would have happened anyway; if it dropped, that drop is what the channel was causing there. Two weeks is a floor, not a rule: a store with few orders a day needs longer before the difference stands out from normal noise.

For illustration: a supplement brand's blog gets few last-click sales, because readers come back later through a paid search ad. A causal estimate could show Organic Search causing more than last click gives it, with organic visits early in many of the converting paths. The next step is to put more into content, then check whether revenue follows. A channel-level estimate won't tell you which article did the work.

Choosing the Right Attribution Model for Your Business

Which model to use depends on how many channels you run, what data you have and which decision you're making. There's no single answer, but for every store the direction is the same: toward numbers that say what a channel caused.

With one or two channels and a simple setup, last click is a fine place to start: it's easy to set up and quick to read. Treat it as a report, not as the thing that decides your budget, because budget calls made on it keep favoring whichever channel closes.

Once you run several channels, switch your reporting to data-driven. GA4 has it built in, and it gives a better sense of how your channels work together and which ones last click was shortchanging. It's still correlation, so read it for where the credit moves, not as proof. Keep in mind whose model it is, too: Google's data-driven model is trained and reported by Google, and Google sells some of the ads it's grading. Before you move money on it, it's worth reading how far to trust GA4's data-driven model.

To decide where budget goes, you need the causal step, and buying a tool doesn't guarantee it. More tracking gives a model more touchpoints to share the credit among; a multi-touch layer still assigns credit rather than measuring what a channel caused. You can start testing without buying anything.

Four checks before you move budget on either model:

1. Switching models moves credit, not sales. Your total orders are the same under both. If a channel gains 20 sales of credit under data-driven, the other channels lost 20 between them; nothing new was sold.
2. Compare them on your own data. GA4's attribution models report (Advertising > Attribution > Attribution models) puts the models side by side for the same period. Where they roughly agree, leave that channel's budget alone for now.
3. Test the channel they disagree on most. Hold it out in one region for two weeks and compare revenue there with the regions where it kept running. That's a causal answer for that channel, from your own sales.
4. Watch MER while you test. Total revenue over total ad spend doesn't depend on any attribution model, so switching models can't move it.

FAQs

What is the fundamental difference between last click and data driven attribution?

How the credit is shared. Last click gives 100% of it to the final touchpoint before the sale. Data-driven attribution uses a model to split it across the touchpoints in the path, based on each one's statistical contribution.

Why is last click attribution often considered inaccurate for modern marketing?

Because it ignores every touchpoint before the final click, including the ones that built awareness and intent. That overcredits bottom-of-funnel channels, undercredits the ones that start journeys, and pushes budget the wrong way. GA4's version also ignores Direct, so a buyer who clicks an ad and later types in your URL still credits the ad.

Can data driven attribution tell me the true causal impact of my marketing?

No. It's better than last click, but it finds correlations: which touchpoints often appear in converting paths, not whether they caused the sale. For causal impact you need causal inference or a test, such as holding out one region and comparing revenue.

How does causal inference differ from traditional data driven attribution?

Causal inference estimates what the marketing caused: the sales that wouldn't have happened without it. It gets there by estimating the counterfactual and correcting for the selection bias in observational data. In practice that means a holdout test, or a model that estimates what revenue would have been without each channel. Data-driven attribution shares out credit based on statistical correlations.

What are the benefits of using a causal inference platform like Causality Engine?

Speed and coverage: a holdout tests one channel at a time, and a causal read covers every channel from data you already have. Later, if you want a second opinion on your reports, a causal attribution tool like Causality Engine puts what each channel caused next to what last click gave it, from a GA4 export, with no pixel or tag to install. It works on GA4's channel groups, not individual campaigns.

Is data driven attribution suitable for small businesses?

Yes, once you run more than one or two channels. With a single channel and short, direct journeys, last click is a fine start. As you add channels, data-driven gives a more accurate picture, and it's built into GA4 at no extra cost.

Learn more about marketing attribution

Understand the limitations of traditional attribution

Related Resources

Agency vs In House Attribution Numbers: Who Is Right

Causality Engine vs. Measured: Incrementality Testing Compared

Attribution Model Comparison Template: Side by Side Analysis

Causal Inference Vs Rule Based Attribution