Multi-touch attribution (MTA) is a measurement method that divides credit for a conversion - a purchase, a sign-up, a lead - among the several ads and channels a person encountered beforehand. A shopper might watch a video ad, see a social post, open an email and finally click a search ad. Last-click reporting gives that final search ad 100% of the sale. MTA spreads the credit across all four, either by fixed rules or by statistical models. It exists because single-touch reporting systematically rewards the channels that sit closest to the checkout and starves those that create demand earlier.

How a path is assembled

Every MTA system begins by reconstructing paths: the ordered sequence of ad exposures and interactions tied to one person or device, ending either in a conversion or in nothing. Building them requires three ingredients.

The first is an identifier that links events. Historically this was a third-party cookie set by an ad server, or a mobile advertising identifier such as Apple's IDFA. Today vendors increasingly rely on hashed emails, logged-in user IDs and identity graphs that match signals across devices and households.

The second is event data. Impression and click logs come from ad servers and demand-side platforms (DSPs), the software buyers use to purchase programmatic inventory. Click identifiers such as Google's gclid or Meta's fbclid carry information through landing-page URLs. Conversion events arrive from a tag on the advertiser's site, a server-side feed or an offline upload from a customer relationship management (CRM) system.

The third is a lookback window, the period before a conversion during which a touchpoint can still earn credit. Thirty days for clicks and one day for views are common settings, but each platform sets its own defaults.

Rule-based models

The oldest MTA approaches apply fixed weights chosen in advance. Google's Campaign Manager 360 documentation and Adobe's Customer Journey Analytics documentation describe the standard set:

  • Linear "gives equal credit to each channel interaction on the way to a conversion," according to Google.
  • Time decay gives more credit to touchpoints closer to the conversion. Both vendors use a default half-life of 7 days, meaning "a touch point 7 days before a conversion will get half the credit," according to Google. Adobe expresses the weight as 2^(-t/half-life).
  • Position-based, which Adobe calls U-shaped, assigns 40% each to the first and last interaction and splits the remaining 20% among the middle.
  • J-shaped gives 60% to the last touch, 20% to the first and 20% to the middle; inverse J reverses the first and last weights, according to Adobe.

A worked example shows how much the choice matters. Take a $100 sale preceded by a display ad 20 days earlier, a social ad 10 days earlier, an email 3 days earlier and a search click on the day of purchase. Linear gives each touch $25. Position-based gives display and search $40 each, and social and email $10 each. Time decay with a 7-day half-life gives display about $6, social $17, email $33 and search $44. Last click gives search $100. The same path, run through four rules, produces a display value ranging from zero to $40.

Rule-based models are transparent, but the rules encode the analyst's assumptions about buyer behaviour rather than measuring it.

Algorithmic models

Algorithmic, or data-driven, MTA learns weights from data, and crucially it uses non-converting paths as well as converting ones. Comparing the two reveals which touchpoints are associated with a higher probability of conversion.

Three families dominate. Shapley value models, drawn from Lloyd Shapley's 1953 game-theory work, average each channel's marginal contribution across every possible order in which it could have appeared. Adobe's algorithmic model uses "a generalization of the Shapley value," according to its documentation.

Markov chain models treat a path as a sequence of states and estimate transition probabilities between channels. A channel's credit is derived from its removal effect - how far the overall conversion probability falls when that channel is deleted from the graph. Eva Anderl, Ingo Becker, Florian von Wangenheim and Jan Hendrik Schumann set out the framework in the International Journal of Research in Marketing in 2016, applying it to four customer-level datasets with at least seven online channels each. Their results differed substantially from last-click heuristics and identified carryover and spillover effects between channels.

Regression and machine-learning models, from logistic regression to neural sequence models, predict conversion from path features such as recency, frequency and creative type, then translate coefficients into credit.

Who runs it

On the buy side, MTA is operated by advertisers' analytics teams, media agencies and independent vendors. Analytics suites such as Adobe offer it inside reporting tools. Independent providers, including TransUnion, which absorbed Neustar's marketing business when it completed a $3.1 billion acquisition on December 1, 2021, according to a TransUnion filing, sell cross-channel products. Direct-to-consumer brands often use newer vendors such as Northbeam, Rockerbox and Triple Whale.

Publishers and platforms control the supply of exposure data, and increasingly withhold user-level logs.

Origin and evolution

The commercial MTA market formed in the late 2000s. Adometry was founded in 2006 and merged into Click Forensics, which took the Adometry name on March 1, 2011, according to TechCrunch. Google's Analytics team introduced Multi-Channel Funnels reports in August 2011, exposing assisted conversions to ordinary marketers for the first time.

Consolidation followed. On May 6, 2014, AOL acquired Convertro, "a leading provider of multi-touch attribution modeling technology," for about $101 million including a $10 million earn-out, according to AOL's filing with the US Securities and Exchange Commission. The same day, Google announced its acquisition of Adometry. Nielsen completed its purchase of Boston-based Visual IQ on October 3, 2017, according to Nielsen.

The data supply then began to contract. Google told buyers it would stop populating the encrypted user ID field in DoubleClick log files for European Union users from May 25, 2018, the day the General Data Protection Regulation took effect, according to MarTech. Without that ID, impressions and clicks could no longer be joined across logs outside Google's Ads Data Hub.

Browsers removed more raw material. Apple blocked all third-party cookies in Safari from March 24, 2020, and App Tracking Transparency arrived with iOS 14.5 in April 2021, requiring opt-in before apps could access the IDFA. Google announced in January 2020 that Chrome's third-party cookies would go within two years, then switched to a user-choice plan on July 22, 2024 and retired most Privacy Sandbox technologies in October 2025.

Platforms also narrowed the menu. On April 6, 2023, Google said it would remove first click, linear, time decay and position-based models from Google Ads and Google Analytics 4, noting that fewer than 3% of Google Ads web conversions used them. Data-driven attribution and last click were left.

Why it matters for marketers

MTA determines which channels appear to work, and therefore where budgets move. It remains widespread. A Winterberry Group survey of 121 US executives, sponsored by measurement vendor Lifesight, found 69% of enterprise brands use multi-touch attribution, against 71% for marketing mix modelling and 52% for incrementality testing.

Confidence lags usage. A Coalition for Innovative Media Measurement (CIMM) and 4As study of 197 senior marketers, published in March 2026, described attribution as a "black box" for advertisers.

Limitations and criticisms

The fundamental objection is that MTA measures association, not causation. A model can see that people who viewed a retargeting ad converted more often, but cannot tell whether the ad persuaded them or simply followed people already about to buy. Brett Gordon and Florian Zettelmeyer of Northwestern University, with Facebook's Neha Bhargava and Dan Chapsky, compared observational methods with 15 randomised experiments at Facebook in Marketing Science in March 2019. Observational methods "often fail to accurately measure the true effect of advertising," Gordon said, according to INFORMS, and generally overestimated it. Genentech's Karan Dhir argued in September 2026 that such methods miss experimental results "frequently by a factor of 3 or more", though the underlying papers state no single multiplier.

Coverage is the second problem. Walled gardens rarely release user-level exposure data, and offline media such as linear television, radio and out-of-home leave no path at all. MTA therefore over-credits measurable digital channels by default.

Signal loss compounds both. Alliance Digitale, the French digital advertising trade body, warned in January 2026 that multi-touch models are "increasingly difficult to apply" as signals fragment.

Ron Berman of the Wharton School found that attribution creates a "team compensation problem" among publishers.

Not the same as

Data-driven attribution is the algorithmic subset of MTA, and in Google's products the name of a specific default model.

Last-click attribution is single-touch: one interaction receives all credit. It is what MTA was built to replace.

Marketing mix modelling (MMM) works top-down on aggregated weekly spend and sales, needs no user-level paths and can include television, price and seasonality.

Incrementality testing uses randomised control groups to estimate causal lift, often reported as incremental return on ad spend. It answers whether an ad worked, not how to split credit.

Recent developments

Meta has reshaped the inputs most MTA systems ingest. It removed 7-day and 28-day view-through windows from its Ads Insights API from January 12, 2026, and in March 2026 limited click-through attribution to link clicks while naming Northbeam and Triple Whale as partners for its view data. Meta's own measurement framework places MTA in the middle of a "ladder" of incrementality, below randomised experiments.

On May 20, 2026, TransUnion and Google made TransUnion the only MTA provider for YouTube, after a pilot with more than 15 customers. As of October 2026, the direction of travel is towards hybrid systems that calibrate path-level attribution against mix models and experiments rather than trusting any single method.

Timeline

  • 1953 - Lloyd Shapley publishes the value concept later used in algorithmic attribution
  • 2006 - Adometry founded
  • March 1, 2011 - Click Forensics merges with Adometry and adopts its name
  • August 2011 - Google Analytics introduces Multi-Channel Funnels reports
  • May 6, 2014 - AOL acquires Convertro for about $101 million; Google announces the acquisition of Adometry
  • 2016 - Anderl, Becker, von Wangenheim and Schumann publish the graph-based Markov attribution framework
  • October 3, 2017 - Nielsen completes its acquisition of Visual IQ
  • May 25, 2018 - DoubleClick log files stop carrying user IDs for EU users
  • March 2019 - Gordon, Zettelmeyer, Bhargava and Chapsky publish their Facebook field-experiment comparison
  • January 14, 2020 - Google announces plans to phase out Chrome third-party cookies
  • March 24, 2020 - Safari blocks all third-party cookies
  • April 2021 - App Tracking Transparency ships with iOS 14.5
  • December 1, 2021 - TransUnion completes its $3.1 billion acquisition of Neustar
  • April 6, 2023 - Google announces retirement of four rules-based models in Google Ads and GA4
  • July 22, 2024 - Google drops Chrome cookie deprecation in favour of user choice
  • October 2025 - Google retires most Privacy Sandbox technologies
  • January 12, 2026 - Meta removes 7-day and 28-day view windows from its Ads Insights API
  • January 2026 - Alliance Digitale publishes its hybrid measurement white paper
  • March 2026 - Meta limits click-through attribution to link clicks
  • May 20, 2026 - TransUnion becomes the sole MTA provider for YouTube
  • September 30, 2026 - Winterberry Group report finds 69% of enterprise brands use MTA

Summary

Who: Advertisers, agencies and analytics teams on the buy side; vendors such as Adobe, TransUnion, Nielsen's Visual IQ, Northbeam and Rockerbox; platforms and publishers that control the exposure data.

What: A method that splits conversion credit across multiple ad touchpoints using fixed rules (linear, time decay, position-based) or algorithms (Shapley value, Markov chains, machine learning).

When: Grounded in 1953 game theory, commercialised from the late 2000s, consolidated by acquisitions in 2014 and 2017, and constrained by identifier loss from 2018 onwards.

Where: In analytics suites, ad servers and independent measurement platforms, drawing on cookie, device ID, click ID and first-party identity data.

Why: Last-click reporting misstates the role of earlier ads; critics counter that MTA is correlational, blind to offline and walled-garden media, and weakened by privacy changes.