Marketing attribution: how to assign credit for a conversion and why most teams pick the wrong model.

What marketing attribution actually is

Marketing attribution is the practice of assigning credit for a conversion across the touchpoints that led to it. A customer sees a TikTok ad, searches Google, reads a review, and finally clicks an email. Attribution decides how much of the eventual purchase to credit to each touchpoint. The rule the team picks determines which channels look profitable, which look wasteful, and which budgets get cut next quarter.

The work matters because budget decisions follow the credit assignment. If your model gives 100 percent of the credit to the last touchpoint, paid search will look enormous and your top-of-funnel channels will look worthless. Cut the top-of-funnel and the bottom-of-funnel volume collapses six months later. The attribution model is a lens; the lens shapes every decision the team makes.

Seven main attribution methods exist. Heuristic models (first-touch, last-touch, linear, time-decay, position-based) are simple rules that split credit by a formula. Data-driven models (Markov chains, Shapley value, Google's data-driven attribution) use observed user paths to estimate causal contribution. Incrementality testing is the gold standard but requires controlled experiments and longer time horizons.

The seven attribution models that matter

Heuristic models are easy to compute and easy to communicate but assume credit by rule, not by data. Data-driven models are harder but closer to truth. Incrementality testing is the actual causal answer. Modern measurement uses a combination: data-driven attribution as the day-to-day lens, incrementality tests as the quarterly truth-check.

The seven main attribution methods, what each does, and when each is right

Model	How it works	Best fit
First-touch	100% credit to the first touchpoint	Awareness-stage attribution; honest about what introduced the customer
Last-touch	100% credit to the last touchpoint before conversion	Closing-stage attribution; honest about what closed the deal
Linear	Equal credit to every touchpoint in the path	When you have no opinion about which stage matters more
Time-decay	More credit to touchpoints closer to conversion	Short consideration cycles; impulse purchases
Position-based (U-shaped)	40% first, 40% last, 20% middle	Both ends of the funnel matter; middle is supporting
Markov chain	Probabilistic removal-effect modeling	Multi-touch paths with sufficient data; data-driven approach
Shapley value	Game-theoretic credit allocation	The mathematically fair allocation; computationally expensive

Data-driven attribution and Google DDA

Data-driven attribution (DDA) uses observed user paths to estimate each channel's incremental contribution. Google's DDA model, available in Google Ads and GA4 since 2021, applies a Markov-chain approach to your account's actual conversion paths. The output is per-channel credit based on what would happen if that channel were removed. DDA is the default attribution model in GA4 since October 2023.

The math behind DDA is removal-effect modeling. The system builds a transition probability matrix from your observed conversion paths. For each channel, it recomputes the conversion probability with that channel removed. The drop in conversion probability is the channel's causal contribution. The math is grounded in Markov-chain theory from probability and is far more honest than heuristic models.

DDA has limits. It only credits channels that appear in converting paths. It cannot measure incremental lift from non-converting traffic. It requires substantial conversion volume (typically 1,000+ conversions per channel) for stable estimates. Pair it with incrementality testing for cross-validation.

Claim: Google made data-driven attribution the default model in GA4 in October 2023, replacing the previous last-click default. The change shifted how every GA4 user sees channel performance. Source: Google Analytics Help: Attribution models in GA4. Context: The shift moved tens of millions of marketers from a 25-year-old last-click view to a probabilistic data-driven view. Many teams discovered their paid-social channels were credited more than they realized once last-click stopped being the default lens.

Why incrementality testing is the gold standard

Incrementality testing is the only attribution method that produces causal answers. The team holds a geographic or audience-level holdout, measures the difference between treated and untreated cohorts, and the gap is the true incremental contribution of the channel. Heuristic models guess. Data-driven models approximate. Incrementality tests measure.

Geo holdouts are the most common method. The team turns off a channel in 10 of 50 U.S. designated market areas (DMAs) for 30 days, measures conversions in the held-out DMAs vs. the running DMAs, and the difference is the channel's incremental contribution. The method works because DMAs are roughly comparable in size and demographics, so the holdout serves as a natural control group.

Conversion lift studies are the platform version of incrementality testing. Meta, Google, TikTok, and LinkedIn all offer built-in conversion lift studies that randomly assign users to test and control cells. The platforms then report the lift. The studies are cheap to run but vulnerable to selection bias because the platforms control both the assignment and the measurement.

How iOS ATT and cookie deprecation broke attribution

Apple's App Tracking Transparency framework launched in iOS 14.5 in April 2021. The framework requires explicit user consent for cross-app tracking. Around 75 percent of iOS users opt out. The change broke mobile-app attribution overnight. Combined with cookie deprecation in browsers, attribution accuracy has declined materially since 2021. Server-side tagging and conversion APIs are how the industry is rebuilding.

The pre-2021 attribution model assumed advertisers could track users across apps and websites with persistent identifiers (IDFA on iOS, third-party cookies on web). Apple's ATT framework killed the IDFA assumption. Browser ITP (Intelligent Tracking Prevention) and ETP (Enhanced Tracking Protection) killed the third-party cookie assumption for Safari and Firefox. Chrome's Privacy Sandbox is doing the same for Chrome, gradually.

The post-2021 measurement stack relies on first-party data. Server-side tagging (sGTM), conversion APIs (Meta CAPI, Google Enhanced Conversions, TikTok Events API, LinkedIn Conversions API), and customer data platforms (CDPs) replace the lost browser-side and device-side tracking. The cost is operational complexity. The benefit is durability against further privacy regulation.

Three common attribution mistakes

Three measurement failures appear in almost every audit. Each one causes systematic misallocation of budget. Catching them early is the difference between a profitable program and one that quietly burns capital.

Mistake 1: relying on a single attribution model

The team picks last-click (or DDA, or first-touch) and runs the whole business off it. Every channel decision gets made through one lens. The fix is to look at the same data through multiple models — first-touch, last-touch, DDA, and an MMM if available — and pay attention when they disagree. Disagreement is signal that some channels are being misvalued.

Mistake 2: not running incrementality tests

The team trusts attribution models without ever testing them. The fix is quarterly incrementality tests on the largest channels. A 30-day geo holdout on Google Search or Meta Ads is cheap to run and tells you whether the channel's attribution credit reflects real incremental value.

Mistake 3: ignoring the iOS / cookie attribution decline

The team's attribution accuracy degraded materially after 2021 and the team has not adjusted. Reports look the same shape but the underlying data has gaps. The fix is implementing server-side tagging and conversion APIs for every paid channel, plus running periodic incrementality tests to catch where the attribution layer is leaking.

Quick answers

What is marketing attribution in plain English?

Figuring out which marketing channels deserve credit for a sale. If a customer saw a TikTok ad, then a Google ad, then bought after an email, attribution decides how much of the sale to credit to each channel.

What is the best attribution model?

There is no single best model. The honest practice is to look at the same data through multiple models and pay attention when they disagree. Disagreement is signal.

What is Google data-driven attribution?

Google's data-driven attribution (DDA) uses observed user paths to estimate each channel's incremental contribution. The math is Markov-chain removal-effect modeling. DDA became the default in GA4 in October 2023.

What is incrementality testing?

Holding a portion of users or geographies as a control group, running the channel only for the rest, and measuring the difference. The gap is the true incremental contribution. Incrementality is the gold standard but takes 30+ days per test.

How did iOS ATT change attribution?

Apple's ATT framework (iOS 14.5, April 2021) requires explicit consent for cross-app tracking. Around 75 percent of iOS users opt out. The change broke mobile-app attribution overnight.

What replaces lost attribution accuracy?

Server-side tagging (sGTM), conversion APIs (Meta CAPI, Google Enhanced Conversions, TikTok Events API, LinkedIn Conversions API), and customer data platforms. The new stack relies on first-party data rather than browser cookies and device identifiers.

Frequently asked

What is marketing attribution?

Marketing attribution is the practice of assigning credit for a conversion across the touchpoints that led to it. The model the team picks determines which channels look profitable and which budgets get cut next quarter. There are seven main models and a separate gold-standard method (incrementality testing).

What are the main attribution models?

First-touch, last-touch, linear, time-decay, position-based (U-shaped), Markov chain, and Shapley value. The first five are heuristic models. The last two are data-driven models that use observed paths to estimate contribution.

What is data-driven attribution (DDA)?

DDA uses observed user paths to estimate each channel's incremental contribution. Google's DDA model applies a Markov-chain removal-effect approach. DDA became the default in GA4 in October 2023, replacing 25 years of last-click defaults.

What is incrementality testing?

Holding a control group (often a set of geographic markets) where a channel is paused, then measuring the difference vs. the running geos. The gap is the channel's true incremental contribution. Incrementality is the only causal attribution method.

How did iOS ATT change attribution?

Apple's App Tracking Transparency framework (iOS 14.5, April 2021) requires explicit user consent for cross-app tracking. Around 75 percent of iOS users opt out. The change broke mobile-app attribution overnight and degraded multi-touch attribution accuracy across the board.

What is server-side tagging?

Server-side tagging routes analytics and conversion data through a server you control rather than the browser. The result is more reliable conversion measurement that survives cookie restrictions, ad blockers, and ITP. Google's sGTM, Tealium, and Segment are the main implementations.

What is a conversion API?

A server-to-server API that sends conversion events directly to ad platforms (Meta CAPI, Google Enhanced Conversions, TikTok Events API, LinkedIn Conversions API). The APIs replace browser pixels for conversion measurement and are required for accurate attribution in the post-iOS-ATT world.

How long does an attribution rebuild take?

Implementing server-side tagging plus conversion APIs across the major paid channels typically takes 6 to 12 weeks for a mid-market business. Adding incrementality testing as a quarterly practice takes another quarter to establish baselines.