Marketing Attribution · 2026

Attribution Modeling, Visualized

A customer taps an Instagram ad, googles you, opens an email, watches a review, then buys. Which of those gets the credit for the sale? That's attribution modeling. Switch between the models below and watch the same 100% of credit redistribute across one journey — then learn how to pick a model, and why none of them is ever perfect.

0 / 6 sections
1
The Credit Problem
Touchpoints, a conversion, and the question of who deserves the credit

Almost no one buys on the first click. A customer collects touchpoints — an interaction with your brand at a moment in time — until one of them tips into a conversion (the action you wanted: a purchase, a signup). Attribution modeling is the process of deciding how to split the credit for that conversion across the touchpoints that led to it.

Get it wrong and you'll over-fund the channels that look decisive and starve the ones that quietly did the work. So before any model, get the journey straight.

Demo — hover each touchpoint for its role, then play the path to the sale
Five touchpoints, one conversion. Every model on this page splits credit across these five.
What does an attribution model actually decide?
2
Single-Touch: First- and Last-Click
All the credit to exactly one touchpoint — the simplest (and most common) approach

Single-touch models hand 100% of the credit to one touchpoint. First-touch credits whatever created awareness; last-touch credits the final interaction before the sale — and last-touch (last-click) is the most common model in the marketing world because it's so easy to measure.

The catch is obvious once you see it: everything else in the journey gets a zero.

Demo — flip between first- and last-touch and watch 100% jump
🛒 Conversion — the purchase being attributed
Last-touch attribution assigns 100% of the credit to…
3
Multi-Touch: Linear & Time-Decay
Spread the credit across every touchpoint — equally, or weighted toward the finish

Multi-touch attribution (MTA) distributes credit across all the touchpoints. The simplest split is linear — everyone gets an equal share. Time-decay keeps multi-touch but gives more weight the closer a touch sits to the conversion, on the theory that recent nudges mattered more.

It's a real shift in thinking: a 2024 MMA report found 52% of marketers use multi-touch attribution, and they report more confidence in how they allocate budget than single-touch users.

Demo — compare an equal split with a recency-weighted one
🛒 Conversion — the purchase being attributed
Compared with linear, time-decay attribution…
4
Position-Based: U, W, and Z Shapes
Reward the milestone moments — discovery, conversion, and the key steps between

Position-based models concentrate credit on the touchpoints that matter most and treat the rest as supporting cast. The shapes are named for where the credit piles up:

U-shaped rewards the first and last touch (≈40% each), splitting the remainder across the middle — like a book whose beginning and end you remember while the middle goes fuzzy. W-shaped adds a third milestone in the middle; Z-shaped recognizes four milestones.

Demo — switch shapes; the highlighted bars are the milestones
🛒 Conversion — the purchase being attributed
In U-shaped (position-based) attribution, which touchpoints get the most credit?
5
Data-Driven & the Wider Toolkit
When a fixed rule isn't enough: let the data decide — and look beyond clicks

Every model so far applies a fixed rule. Data-driven attribution (DDA) instead uses machine learning over your historical conversions to learn how much each touchpoint really contributed — including incremental credit for touches a conversion wouldn't have happened without. It's the most accurate option, but it needs a large volume of data to work.

And attribution is bigger than clicks. Last non-direct click credits the last non-direct touch (a newsletter or ad) rather than someone typing in your URL. Cross-channel models weigh every channel together. Marketing Mix Modeling (MMM) zooms all the way out to include offline — TV, print, in-store — and seasonality. Vendors layer on proprietary blends of first- and zero-party data on top.

Use the master allocator below to flip through all the headline models on the same journey. Watch how dramatically the same five touchpoints get revalued.
Demo — the master credit allocator: every model, one journey
🛒 Conversion — the purchase being attributed
What makes data-driven attribution different from the rule-based models?
6
Choosing a Model — and What No Model Can Fix
Match the model to your reality, and stay honest about its blind spots

There's no universally "correct" model — the right one depends on your sales-cycle length, journey complexity, how much conversion data you have, and whether offline channels are in play. Answer a few questions and see where that points.

Demo — answer four questions for a starting-point recommendation
1 · How long is your sales cycle?
2 · Which channels do you run?
3 · How much conversion data do you have?
4 · How complex is the typical journey?
Recommendation
Answer the four questions above…
Pick one option in each row and a starting-point model will appear here.
What no model can fix. Attribution is built on incomplete data. Cookies are disappearing in favor of first- and zero-party data; offline and word-of-mouth go untracked; people switch devices mid-journey; assisted conversions get undervalued; and different models produce different answers, so the numbers can be nudged. There is no model that captures 100% of every customer's data — the best one is simply the one that gets you closest to the truth for the decisions you need to make.
Why can no attribution model be 100% accurate?

🎯 You can read an attribution model now

Touchpoints → a conversion → credit. Single-touch (first/last) hands it all to one touch; multi-touch spreads it (linear, time-decay, position-based U/W/Z); data-driven learns it from your data; MMM zooms out to offline. Pick for your cycle, data, and channels — and remember no model captures everything.

Learning Reference · Triple Whale — Attribution Modeling

Get new posts in your inbox

No spam — just new posts and learning pages when they ship.

Share X LinkedIn Reddit Hacker News