Marketing Attribution Models: The Definitive Guide

Last updated: June 2026

Every marketer wants the same thing: to know which channels actually drive growth, so they can spend more on what works and less on what doesn’t. Attribution models promise that answer. Most of them fail to deliver it.

This guide is for marketers and marketing leaders who have to report on performance and decide where the next pound goes. It covers what each attribution model does, which ones are worth trusting, the tools that run them, and the most uncomfortable question of all: whether attribution is worth your time at all. What it deliberately avoids is pretending there is one correct model. There isn’t.

What marketing attribution actually means

Marketing attribution is the practice of assigning credit for a conversion across the marketing touchpoints a buyer interacted with on the way there. An attribution model is the specific rule you use to divide that credit, whether that means handing it all to the last click, splitting it evenly, or letting an algorithm decide.

The model you choose changes the story your data tells. Switch from last-click to time-decay and your paid search results shrink while your email and content numbers grow, even though nothing about your actual marketing changed. That is the central trap, and it is why attribution deserves more scepticism than most teams give it.

The state of play

Attribution has been a stated priority far longer than it has been a solved problem. Roughly 84% of marketers list connecting conversions to marketing as a top digital priority, yet only around 10% of companies report having that capability, according to analytics educator Jeff Sauer.

The ground has shifted underneath all of it. Third-party cookie deprecation, iOS privacy changes, and longer, darker buying journeys have made click-based tracking less reliable every year. Google made data-driven attribution the default in GA4 and retired most of the old rule-based models from its reporting. At the same time, two older approaches have come back into fashion precisely because they don’t depend on tracking individuals: marketing mix modelling and incrementality testing. The direction of travel is away from “who do we give credit to” and toward “what would have happened anyway”.

The models, and what I actually think of each

This is the part most guides skip. Here is an honest take on each model, with a link to a deeper explainer for the ones worth your time.

• Last-click. Flawed. It rewards proximity to the conversion, not the work that created demand. Branded search and retargeting look like heroes while the content and brand-building that actually moved the buyer get nothing.
• First-click. Tidy in theory, broken in practice. It assumes a buyer discovers you once, on one device, and travels in a straight line to purchase. Real journeys are fragmented across devices and sessions, so the “first” click you record is rarely the real first touch.
• Linear. Call it the participation-trophy model. Everyone gets equal credit for everything. It feels fair and it is easy to explain, but treating a throwaway impression and a sales-influencing webinar as equals tells you almost nothing about what to do next.
• Position-based (U-shaped). A reasonable compromise: load credit onto the first touch (discovery) and the last (conversion), then split the rest evenly across the middle. Sensible if you believe those two moments matter most, which in many B2B journeys they do.
• Time-decay. Gives more credit to the touches closest to the sale and less to the earlier ones. Of the rule-based models, this is the one that maps most cleanly to how considered purchases tend to build momentum over time.
• Custom. Makes sense and is almost always a mistake. A custom model is only as good as the assumptions you bake into it, and those assumptions tend to favour whoever built the model and wants the credit. You can build one in Google Analytics. About 99% of teams shouldn’t.
• Data-driven (DDA). Algorithmic credit based on your actual conversion paths, often using Shapley value or Markov chain methods under the hood. Genuinely better than the rule-based models, but only if you have the data volume to feed it, and only if you remember whose algorithm you are trusting (see Google, below).

Two models worth knowing that aren’t really about dividing credit at all:

• Incrementality testing. Instead of splitting credit, you run holdouts and experiments to measure what a channel actually caused. The closest thing attribution has to a truth serum.
• Wittgenstein’s attribution model. Not a model so much as a warning: a reminder that a flawed attribution tool measures your own biases as much as it measures reality.

Is attribution even worth your time?

Before you agonise over model shapes, ask whether you should be doing formal attribution at all. The single most useful rule here comes from analytics educator Jeff Sauer, whose talk on the subject is worth twenty minutes of anyone’s time.

Sauer’s framing is deliberately blunt. As he puts it, “Attribution is bullshit. But is it worth your time anyway?” His answer is a volume test:

• Fewer than 100 conversions: no. You don’t have the data, and you’ll be reading noise.
• More than 100 conversions, strong online focus: maybe.
• More than 10,000 conversions, plus online and offline data: yes, this is where attribution earns its keep.

The principle underneath the numbers is the one to tattoo on the wall: “Complexity of attribution strategy = complexity of marketing strategy”. If you are running two channels and a newsletter, a sophisticated multi-touch model is a waste of effort. Match the measurement to the marketing, not to your ambitions.

What actually works

High confidence

• Match model complexity to marketing complexity. The biggest attribution win is usually deciding not to over-engineer it. Simple operations deserve simple measurement.
• Compare models instead of picking one. In your analytics tool, view the same period through several models side by side. The differences between them are more informative than any single model’s “answer”.
• Self-reported attribution. A “How did you hear about us?” field captures dark-social and word-of-mouth influence that no click-based model can see. Cheap, and often the most honest signal you have.
• Incrementality and holdout tests. Turning a channel off for a region or audience and measuring the difference tells you what it actually caused, not what it happened to be near.

Moderate confidence

• Data-driven attribution for high-volume online conversions. Worth it once you genuinely clear the data threshold, with eyes open about its biases.
• Marketing mix modelling. Strong for blending online and offline and surviving a cookieless world, though it needs scale and history.

Emerging evidence

• Unified approaches that triangulate MTA, MMM, and incrementality rather than trusting any one of them. Promising, and still maturing.

What to ignore (for now)

• Building a custom model when you’re not ready. For the vast majority of teams this is effort spent encoding your own bias into a dashboard.
• Treating Google’s data-driven attribution as objective truth. It applies DDA to your data through a Google-centric lens and will tend to over-report the channels Google sells you, namely paid search and, by extension, SEO. Useful, not neutral.
• Obsessing over the model shape. Teams burn weeks debating W-shaped versus linear while ignoring win rates, pipeline growth, and whether the business is actually healthy. The shape is rarely the thing that matters.
• Last-click as a source of truth. Fine as a rough operational signal. Dangerous as the basis for budget decisions.

How to measure it

Start in the tool you already have. Google Analytics lets you compare attribution models against the same data, which is the fastest way to feel how much the model choice moves the numbers. Beyond GA, the right tool depends on your business:

• Tune for mobile attribution.
• Bizible (now Adobe) for B2B organisations with long sales cycles.
• BrightFunnel for sales-led organisations that need pipeline and revenue attribution.
• ConvertPro-style tools for businesses bridging online and offline conversions.

Whatever you use, anchor the work to business KPIs first (pipeline, win rate, revenue), then use attribution to answer specific questions like “why did win rates drop last quarter”, rather than as a standing report nobody acts on.

Action plan

• This week: open your analytics tool and compare last-click, first-click, linear, and time-decay over the same 90-day window. Note which channels swing the most. Add a “How did you hear about us?” field to your highest-intent form.
• This quarter: run Sauer’s volume test honestly. If you clear the bar, pick the simplest model that fits your journey and document why. If you don’t, stop investing in attribution machinery and put the effort into one clean incrementality test instead.

The resource library

Go deeper with our related guides:

• What is data-driven attribution? - the algorithmic approach Google now defaults to
• Everything you need to know about GA4 data-driven attribution - the practical GA4 walkthrough
• What is multi-touch attribution? - distributing credit across the journey
• Markov chain attribution modelling - the removal-effect method behind much of DDA
• Shapley value attribution - the game-theory approach to fair credit
• Incrementality testing - measuring what a channel actually caused
• Wittgenstein’s attribution model - why your ruler measures your biases
• Marketing mix modelling - the cookieless, top-down alternative
• Uncovering the attribution mirage - seeing past brand and media confusion

From around the web:

• The MineThatData Blog - Kevin Hillstrom on measurement, incrementality, and why correlation keeps fooling marketers
• Jeff Sauer: Attribution is Bullshit - the deck behind the talk
• Multi-Channel Attribution Modeling: The Good, Bad and Ugly Models - Avinash Kaushik

People to follow: Jeff Sauer (@jeffalytics), Kevin Hillstrom, Avinash Kaushik, Alex Birkett.

Tools to watch: Google Analytics 4, Tune, Bizible (Adobe), BrightFunnel.

Where this is heading

Attribution is quietly being demoted from oracle to instrument. The privacy changes that broke individual-level tracking have made the old dream of perfectly tracing every buyer to its source look naive, and frankly that is healthy. The teams getting this right in 2026 treat attribution as one input among several: a directional read they cross-check against incrementality tests and mix modelling, not a verdict they obey. The smartest question is no longer “which channel gets the credit”, it is “what would have happened if we hadn’t run this at all”. Models that can’t help answer that are on their way out.

FAQ

The bottom line

There is no right attribution model, only models that are more or less useful for your situation. Compare several rather than crowning one, match the sophistication of your measurement to the sophistication of your marketing, and treat every model’s output as a hypothesis to test, not a fact to act on blindly. If you take one thing from this guide: attribution should answer a specific business question, or you shouldn’t be doing it.

How Growth Method Helps

Growth Method is the growth platform designed for experiment-led and data-driven marketers. We help teams move past vanity dashboards and measure what actually drives growth.

Learn more on our homepage, connect with me on LinkedIn or Twitter, or book a call here.