Multi-touch attribution, explained: a RevOps playbook

Every B2B deal I've ever seen involved more than one touchpoint. A prospect reads a blog post in January, clicks a retargeting ad in February, joins a webinar in March, books a demo in April, closes in May. Single-touch attribution forces you to pick one of those five moments and give it all the credit. That's not analysis — that's a coin flip with extra steps.

Multi-touch attribution (MTA) is the category of models that distribute credit across every touchpoint in the customer journey. It's the only honest way to think about B2B marketing ROI, and it's also where RevOps teams quietly lose their minds because the models contradict each other.

This post walks through what MTA actually is, which models exist, where each one shines and fails, and — the thing most articles skip — how to pick one without drowning in vendor marketing.

What multi-touch attribution actually is

Multi-touch attribution answers one question: when a customer finally converts, how should we split credit across every marketing and sales touch that contributed?

There are three moving parts:

1. The journey: every trackable interaction a prospect had before they converted. A click on an ad, a page view, a newsletter open, a webinar attendance, an SDR email, a sales call.
2. The model: the set of rules for distributing credit. This is where the arguments happen.
3. The conversion event: what "counts" as the win you're crediting. Opportunity created? Deal won? First payment? Renewal? The choice materially changes the answer.

The journey data is the hard part technically — CRMs, ad platforms, analytics, and marketing automation tools all store touchpoints in different schemas, and stitching them together is where 80% of attribution projects die. The model is the easy part, relatively, because there are only a handful of widely-used ones.

The five models you need to know

Linear

Every touchpoint gets equal credit. Five touches in the journey? Each gets 20%. Simple to explain, simple to implement, and philosophically defensible: you can't prove any touchpoint mattered more than any other without running an experiment.

When it works: short sales cycles (under 30 days), high-frequency conversion events (e-commerce), small number of touchpoints per journey.

When it breaks: long B2B cycles with dozens of touches where the 3am blog post read genuinely didn't carry the same weight as the discovery call.

U-shaped (position-based)

First and last touches get 40% each. The remaining 20% is distributed evenly across middle touches. The intuition: the introduction and the close both matter disproportionately.

When it works: most B2B scenarios. It's the model I'd default to for mid-market SaaS unless something argues otherwise.

When it breaks: if the very first touch was tangential — a meme shared in Slack, a random LinkedIn impression — 40% credit is too much. If the last touch was just an SDR calling to close a warm deal, same problem inverted.

Time-decay

Touches closer to the conversion get more credit. Exponential decay, usually with a half-life of 7 days. So a touch 3 weeks before close gets roughly 1/8 the credit of a touch the day before close.

When it works: impulse purchases, short cycles, or journeys where the final push is genuinely what drove the decision.

When it breaks: B2B journeys where the first touch was "saw a case study six months ago that planted the seed." Time-decay heavily discounts early-funnel content, which punishes your best-performing organic search and brand marketing.

W-shaped

Like U-shaped, but with an extra 30% credit assigned to the "opportunity creation" midpoint. First touch 30%, opportunity 30%, last touch 30%, other middle touches share 10%.

When it works: B2B with a clear MQL-to-opportunity moment that's a genuine funnel milestone (not just a CRM field being set).

When it breaks: when your opportunity-creation moment is fake — created automatically by a workflow when a lead hits a score — rather than a real qualification.

Algorithmic / data-driven

A machine-learning model (usually Markov chains or Shapley values under the hood) that assigns credit based on what the data shows actually influenced conversion, calibrated on your own journeys. Google Analytics and a handful of attribution vendors offer this.

When it works: high-volume businesses with thousands of conversions per month. The model needs data to be any good.

When it breaks: mid-market B2B. If you close 40 deals a quarter, no ML model will find statistically significant signal in your journey data. You'll get a model that's marginally better than linear and harder to explain.

The four questions to ask before picking a model

1. What conversion event are we crediting?

"Opportunity created" and "deal won" produce wildly different attribution results. Opportunity-created credits demand-gen. Deal-won credits late-funnel sales. Pick the one that maps to the decision you're making. If you're budgeting marketing spend, credit opportunity. If you're evaluating the sales team, credit closed-won.

See our walkthrough of first-touch vs last-touch vs multi-touch for the conversion-event trap in more detail.

2. How long is your sales cycle?

Under 14 days: linear or time-decay work fine, because the journey is short enough that distribution doesn't matter much. Over 60 days: you need something like U-shaped that explicitly recognizes the first touch, because half your journeys start with a blog post read six months before the deal.

3. Do you trust your touchpoint data?

If your UTMs are a mess, your GA4 is half-implemented, and your CRM sales activity logging is spotty, no attribution model will save you. Garbage in, garbage out — and attribution models are particularly unforgiving because they amplify data quality problems.

Fix the data pipeline first. Our post on why attribution numbers don't add up covers the five most common data hygiene problems to fix before you touch a model.

4. How will you handle organic search keywords?

Organic search rarely shows up in first-party touchpoint data as a specific keyword — it shows up as "google.com / organic" with no further detail. For businesses where keyword-level attribution matters (SEO-heavy verticals, content marketing teams), you need a probabilistic matching layer that pairs GSC click data with CRM touchpoints.

We covered this in depth in keyword-level attribution for organic search. Short version: if organic is your top channel, a standard attribution model isn't enough.

Where multi-touch attribution still fails

Even a perfect MTA model doesn't solve the fundamental problem: you're measuring correlation, not causation. A customer who saw 12 touches before converting probably would have converted with 11 touches too. Or 10. Or 3. Attribution tells you what touched, not what caused.

The only real way to measure causation is controlled experiments — holdouts, geo tests, incrementality studies. Most B2B teams don't have the volume to run them rigorously. So attribution is the second-best tool, and the honest version is: "this model produces a repeatable ranking of channel contribution that we use to make budget decisions, and we know it's directionally correct, not literally correct."

A CMO who can say that out loud will not be caught off guard by the CFO's questions. A CMO who insists the multi-touch numbers are "the truth" will eventually learn otherwise.

Getting started

If you're starting from scratch:

1. Pick a conversion event. Most teams should credit opportunity creation AND closed-won separately — two reports, two stories.
2. Pick U-shaped as your default model.
3. Clean up UTM discipline and CRM source tracking before building anything.
4. Get the model running. Don't optimize it for three months — optimize your data for three months.
5. Once it's running, run a comparison view against time-decay and linear. If they all agree on channel ranking, you're good. If they diverge wildly, your data has a problem.

And if you want multi-touch attribution that actually works out of the box — including the probabilistic keyword engine we mentioned — book a 20-minute walkthrough and we'll show you what the models look like running on your data.

TL;DR

Multi-touch attribution splits conversion credit across every touch in the journey. There are five mainstream models — linear, U-shaped, time-decay, W-shaped, algorithmic. For most B2B RevOps teams, U-shaped is the right default. The model is the easy part. Clean touchpoint data is the hard part. And no model measures causation, no matter what the vendor says.