Multi-Touch Attribution Is Dead. Here’s What Replaced It (2026)

For a long time, multi-touch attribution seemed like the answer to a genuinely hard question.

The question was: a customer saw a YouTube ad on Monday, clicked a Facebook post on Thursday, searched your brand name on Saturday, and bought on Sunday. Which ad gets the credit?

Last-click attribution gave it all to the Google search. First-click gave it all to YouTube. Neither felt right. So the industry built more sophisticated models that distributed credit across every touchpoint in the journey. Linear models split it evenly. Time-decay models weighted recency. Data-driven models used machine learning to assign fractional credit based on which touchpoints correlated with conversion.

These models got more technically impressive every year. The problem is that a more sophisticated version of a flawed idea is still a flawed idea.

Multi-touch attribution is not just outdated. It was built on an assumption that was always shaky and is now almost completely false. And the brands still treating MTA as their primary measurement framework are making budget decisions on data that ranges from incomplete to actively misleading.

The promise of MTA was visibility. If you could track every touchpoint a customer had before converting, you could understand which channels and campaigns were contributing to revenue. You could allocate budget toward what was working and pull back from what was not.

That promise was compelling because the alternative, last-touch attribution, was obviously wrong. Giving all the credit to the final click before purchase systematically undervalued brand awareness, upper funnel media, and any channel that introduced customers to a brand before they were ready to buy. MTA at least attempted to correct for that.

For a window of time, roughly 2012 to 2018, the mechanics mostly held together. Third-party cookies were functional. People used one device most of the time. Cross-site tracking worked well enough that you could stitch together a customer journey with some reliability. The data was messy but it was there.

That window has closed.

Third-party cookies are effectively gone. Safari and Firefox blocked them years ago. Chrome has been in the process of deprecating them, and while the exact timeline has shifted, the direction has not. Third-party cookies were the scaffolding on which most MTA systems were built. Without them, cross-site tracking breaks down. You cannot attribute across channels if you cannot follow the user across channels.

iOS 14 changed mobile tracking fundamentally. Apple’s App Tracking Transparency framework, released in 2021, required apps to ask users for permission to track them. Most users declined. Overnight, the signal loss on Meta, Snap, and other mobile-heavy platforms was severe. Some advertisers reported losing visibility into 40 to 60 percent of iOS conversions. MTA systems built on this data became structurally incomplete.

Walled gardens never shared their data anyway. Google, Meta, Amazon, and TikTok each have their own measurement systems, their own attribution windows, and their own incentives. They do not share raw user-level data with third-party attribution vendors. They share summary data at best. So any MTA model that claims to give you a unified view across all your channels is, in practice, working with significant gaps where the biggest platforms sit.

Put these three things together and you get a measurement framework that might be tracking somewhere between 30 and 60 percent of actual customer touchpoints, depending on your channel mix, audience demographics, and how much of your traffic is mobile. The rest is either guessed at or missing.

Even if the data problems were solved tomorrow, MTA would still have a fundamental issue that no amount of engineering has ever addressed.

Attribution models measure correlation, not causation.

When a data-driven attribution model says display advertising contributed 12 percent of your revenue last quarter, what it actually means is that people who converted were 12 percent more likely to have seen a display ad than people who did not. That is a correlation. It does not tell you whether the display ad caused those conversions, whether it just happened to reach people who were already in the funnel, or whether those customers would have found you regardless.

This is not a minor methodological quibble. It is the difference between a channel that is genuinely driving new customers and a channel that is efficiently reaching customers who were already on their way to buy.

You cannot answer that question with attribution. Attribution, by design, looks backward at what happened and assigns credit. It does not and cannot run the counterfactual, the version of the world where that ad never ran, and compare the outcomes.

Without the counterfactual, you do not know what the ad actually did.

The shift has been away from user-level tracking and toward methods that can actually establish cause and effect. Three approaches have gained serious traction.

Media mix modeling takes a statistical approach at the aggregate level. Rather than tracking individual users, it analyzes the relationship between historical spend across channels and business outcomes over time. A well-built MMM can tell you that, controlling for seasonality and external factors, each additional dollar of spend in connected TV has historically produced roughly X dollars in incremental revenue, while paid social has produced Y. It does not require cookies, pixels, or user-level data. It works with the data you already have, mostly spend and revenue logs.

MMM was the dominant measurement approach in the 1990s and early 2000s, before digital tracking made user-level attribution seem possible. It fell out of fashion when MTA emerged. Now it is back, rebuilt with better statistical methods and modern computing power, and it is more credible than it was the first time around.

Incrementality testing is the experimental approach. Instead of modeling historical data, you run controlled experiments where you withhold advertising from a segment of your audience or geography and measure what happens to their conversion rate compared to the group that kept seeing ads. The difference is your incremental lift. This is the same logic as a randomized controlled trial in medicine. It is not an inference from correlation. It is a direct measurement of cause and effect.

Geo holdout testing is the most common form of incrementality testing for brands running national campaigns. You split your markets into two groups, run advertising in one and hold it out of the other, then compare revenue. When markets are well-matched and the test is run cleanly, the results are as close to ground truth as marketing measurement gets.

MMM and incrementality testing are complementary, not competing.

MMM gives you a portfolio-level view across all your channels simultaneously. It is continuous, meaning you can update it as new data comes in, and it handles the full complexity of your media mix. The limitation is that it is statistical. The confidence intervals are real and the models need to be built carefully to avoid overfitting.

Incrementality testing gives you high-confidence, causal answers for specific channels or campaigns. The limitation is that experiments take time to run and you cannot test everything at once.

The combination works well in practice. You use MMM to understand your full budget allocation and identify where returns appear to be diminishing. You use incrementality tests to validate specific channels, set iROAS benchmarks, and calibrate the MMM. Over time, the two methods reinforce each other.

What you are not doing is asking an attribution model to tell you which ads deserve credit. You are running experiments that directly measure what would have happened without the advertising.

The risk is not that MTA gives you zero signal. It is that it gives you confident-sounding signal that systematically points in the wrong direction.

Companies optimizing on MTA data tend to over-invest in retargeting and branded search because those channels have the most touchpoints close to conversion and therefore capture the most attributed revenue. They tend to under-invest in prospecting, upper funnel, and brand awareness because those touchpoints are early in the journey and get less credit in most attribution models.

The irony is that prospecting is often what is actually filling the funnel that retargeting is harvesting. When you cut prospecting because its MTA numbers look weak, the retargeting pool slowly depletes. Efficiency looks fine until the day it does not.

This is one of the mechanisms behind the pattern where brands that scale paid social quickly eventually hit a ceiling they cannot explain. The attribution numbers still look good. The revenue growth has stopped. The MTA model cannot surface the problem because it is part of the problem.

Transitioning off MTA as your primary measurement framework does not mean throwing out everything and starting over. It means layering in better evidence where it matters most.

Start with a single incrementality test on your highest-spend channel. Geo holdout tests are typically the easiest to run at scale and produce the most defensible results. Budget four to six weeks for the test. Compare the incremental revenue lift to what your attribution model claimed that channel was driving. The gap between those two numbers will tell you a lot about the state of your measurement.

If you are spending significantly at scale, getting an MMM built is worth the investment. A good model will give you a cross-channel view of where your spend is actually generating returns versus where you are reaching diminishing returns or baseline traffic.

Neither of these things is as fast or as granular as pulling an attribution report. But fast and granular and wrong is worse than slower and right.

This is the problem Measured was built to solve.

Measured runs incrementality tests at scale across every channel, using a geo holdout methodology that does not depend on cookies, pixels, or platform reporting. The platform gives you iROAS benchmarks by channel so you know what each part of your media mix is actually earning, not what the attribution model assigned to it. And because the tests are run continuously, you get an ongoing read on incrementality rather than a one-time snapshot.

For brands that have been running on reported ROAS and platform attribution, the first tests often produce surprises. Some channels that looked efficient turn out to be mostly capturing baseline conversions. Others that looked average on ROAS turn out to be generating strong incremental lift. The budget decisions that come out of that information are usually quite different from the ones the attribution model was pointing toward.

If your current measurement stack cannot tell you what would have happened if you had not run a given campaign, it is worth asking whether it is actually telling you what you need to know.

What is multi-touch attribution? Multi-touch attribution is a measurement approach that distributes conversion credit across multiple ad touchpoints in a customer’s path to purchase. Unlike last-click attribution, which gives all credit to the final interaction before a sale, MTA models like linear, time-decay, and data-driven attempt to reflect the contribution of each touchpoint along the way.

Why is multi-touch attribution no longer reliable? MTA depends on tracking individual users across websites, apps, and devices using third-party cookies and mobile identifiers. The deprecation of third-party cookies, Apple’s iOS tracking restrictions, and the closed nature of major ad platforms have made this tracking increasingly incomplete. Many MTA systems today are missing between 30 and 60 percent of actual customer touchpoints, which means the credit they assign is based on partial data.

What replaced multi-touch attribution? The primary replacements are media mix modeling and incrementality testing. Media mix modeling analyzes aggregate spend and revenue data to estimate channel-level contribution without relying on user tracking. Incrementality testing uses controlled experiments, such as geo holdout tests, to directly measure the causal impact of advertising by comparing outcomes in groups that did and did not see ads.

What is the difference between attribution and incrementality? Attribution assigns credit for conversions that happened. Incrementality measures whether the advertising caused those conversions to happen. A customer who sees a retargeting ad and buys might be attributed to that ad, but if they were already going to buy regardless, the ad generated zero incremental revenue. Incrementality tests establish causality by measuring what would have happened in the absence of advertising.

Is MTA completely useless? Not completely. At the campaign management level, MTA can still provide useful directional signals for creative testing and tactical optimization within a single platform. Where it breaks down is as the basis for cross-channel budget allocation and strategic measurement. For those decisions, incrementality testing and media mix modeling produce more reliable results.

What is a geo holdout test? A geo holdout test is a type of incrementality experiment where you divide your geographic markets into a test group and a control group, run advertising in the test group but not the control group, and compare revenue outcomes between the two groups. Because the test and control groups are statistically matched before the experiment begins, the difference in outcomes can be attributed to the advertising with a high degree of confidence.

Measured helps performance marketing teams replace attribution with incrementality. If you want to know what your media spend is actually doing, rather than what your platforms claim it is doing, learn more at measured.com.