Marketing attribution models, compared side by side

You switch your report from last click to linear and a campaign you were about to pause suddenly looks like one of your best. Switch to first touch and your email program, which looked like a hero a second ago, barely registers. Nothing about your orders changed. Only the rule did.

That rule is the attribution model, and choosing one is one of the quieter decisions that ends up steering your whole ad budget. This guide lays the common marketing attribution models side by side on one shopper's journey, so you can see exactly where each one puts the credit, what question it's good at answering, and where it will mislead you.

What is an attribution model?

An attribution model is the rule that decides how credit for a sale is split across the marketing touchpoints that led to it. A touchpoint is any interaction a shopper has with your brand before buying: clicking an ad, arriving from a search, opening an email.

Most people don't buy on their first visit. They see an ad, come back later from a search, click an email, then buy. The model decides whether the ad, the search, the email, or some mix of them gets the credit for that order. If you want the bigger picture of why this matters, start with what is multi-touch attribution .

Models fall into three families:

• Single-touch models give all the credit to one touchpoint.
• Rule-based multi-touch models split credit across several touchpoints using fixed percentages.
• Algorithmic models (often called data-driven) let software estimate the split from patterns in your data.

The journey we'll use for every model

To compare models fairly, we'll run all of them on the same made-up order. Meet Sam, who buys a $200 jacket:

Four touchpoints, one $200 order. Each model below hands out that $200 differently.

Single-touch attribution models

Last touch (last click)

How it works: the final touchpoint before the purchase gets 100% of the credit.

Sam's order: Email gets $200. TikTok, Meta and search get nothing.

Good at answering: "What closes sales?" It's simple, easy to explain, and it's the default in a lot of tools for that reason.

Where it misleads: it ignores everything that created the demand in the first place. Ads that introduce new people almost never get the last click, so they look worthless and get cut. We go deeper in the pros and cons of last-click attribution .

First touch

How it works: the first recorded touchpoint gets 100% of the credit.

Sam's order: TikTok gets $200.

Good at answering: "What brings new people in?" It's the clearest view of which channels start journeys.

Where it misleads: it ignores everything that happened after discovery, including the retargeting and email that got the person over the line. It also depends on how far back your tracking goes: if the real first touch happened before tracking started, a later touchpoint gets mistaken for the first.

Last touch, paid only

How it works: the last paid touchpoint gets 100% of the credit. Organic visits, email and direct visits are skipped.

Sam's order: Meta gets $200, because it's the last ad Sam clicked.

Good at answering: "Which ad campaign gets credit for this sale?" This is useful when you're judging ad spend specifically and don't want your own email list or brand searches to soak up credit your ads earned.

Where it misleads: it pretends your free channels don't matter. If your email program is doing real work, this model hides it.

First touch, paid only

How it works: the first paid touchpoint gets everything.

Sam's order: TikTok gets $200 (the same as first touch here, because Sam's first touchpoint happened to be an ad).

Good at answering: "Which ads are actually finding new customers?"

Where it misleads: the same blind spots as first touch, plus it ignores organic discovery entirely.

Rule-based multi-touch models

Linear

How it works: every touchpoint gets an equal share.

Sam's order: $50 each to TikTok, Meta, search and email.

Good at answering: "Which channels show up in buying journeys at all?" It's a fair, neutral starting point when you don't have a strong opinion about which stage matters most.

Where it misleads: it treats a casual click the same as the one that convinced someone. In long journeys, credit gets spread so thin that nothing stands out.

Position based (U-shaped)

How it works: the first and last touchpoints each get 40%, and the remaining 20% is split evenly across the touchpoints in the middle.

Sam's order: TikTok $80, Meta $20, search $20, email $80.

Good at answering: "What starts and what finishes journeys?" It respects both discovery and closing, which suits most stores selling through ads. Full detail in what is the U-shaped attribution model .

Where it misleads: the middle of the journey gets very little, even when that's where the real persuasion happens.

J-shaped and inverse J-shaped

These are lopsided versions of position based. Exact splits vary by tool, but a common version is:

• J-shaped: 60% to the last touchpoint, 20% to the first, 20% shared across the middle. It leans toward what closes the sale.
• Inverse J-shaped: 60% to the first touchpoint, 20% to the last, 20% shared across the middle. It leans toward what started it.

Sam's order: J-shaped gives email $120, TikTok $40, Meta $20 and search $20. Inverse J-shaped flips it: TikTok $120, email $40, Meta $20, search $20.

Good at answering: the same questions as first or last touch, but without completely ignoring the rest of the journey. If your gut says closing matters most but you don't want discovery to drop to zero, J-shaped is a reasonable compromise.

Time decay

How it works: touchpoints closer to the purchase get more credit than older ones. Most tools use a half-life: the number of days it takes for a touchpoint's weight to halve. With a 7-day half-life, a touchpoint from a week before the sale counts half as much as one on the day of the sale, and one from two weeks before counts a quarter as much.

Sam's order (7-day half-life): email about $85, search about $57, Meta about $35, TikTok about $23.

Good at answering: "What's driving sales right now?" It fits short buying cycles and promotions, where recent touchpoints really do matter more.

Where it misleads: it consistently undervalues discovery. The ad that introduced the customer is, by definition, the oldest touchpoint. Pick the half-life to match how long your customers usually take to buy: a short half-life on a product people research for weeks will bury your top-of-funnel ads.

W-shaped and Z-shaped

These models give extra weight to milestones in the middle of the journey, such as the moment someone signs up for your email list or adds to cart, as well as the first and last touchpoints. They make the most sense when you can reliably track those milestones and the journey is long enough to have them. We cover them in what is W-shaped attribution and what is Z-shaped attribution .

Participation

How it works: every touchpoint gets full credit for the order. It doesn't split anything.

Sam's order: TikTok, Meta, search and email each get $200. Total credit: $800 for a $200 order.

Good at answering: "How often is this channel involved in sales?" It's a reach measure, not a revenue measure.

Where it misleads: the totals add up to more than your actual revenue, so you can't use it to calculate ROAS or compare channels by money earned. Treat it as a separate lens, never as your main report.

Algorithmic (data-driven) attribution

Data-driven attribution uses statistics or machine learning to estimate how much each touchpoint changes the chance of a purchase, based on comparing journeys that converted with journeys that didn't. Google Analytics 4 uses a data-driven model as its default.

Good at: adapting to your actual data instead of fixed percentages.

Watch out for: it usually needs a lot of conversions to be reliable, and you often can't see why it gave a touchpoint the credit it did. If you can't explain a number to your team, it's hard to act on it with confidence. Also remember that a model built inside one platform still only sees that platform's view of the journey.

All the marketing attribution models on one page

Here's Sam's $200 order across every model:

Look at the Meta column. Depending on the model, the same ad on the same order earned $0, $20, $35, $50 or $200. Now imagine that across a month of orders and a real budget. That's why "which model?" is never a technicality.

How to choose an attribution model

There's no single correct model. Each one is a different question. Here's a practical way to pick:

1. Write down the decision you're making. Cutting a prospecting campaign? Look at first touch or inverse J-shaped. Judging retargeting or email? Look at last touch or J-shaped. Setting overall channel budgets? Start with linear or position based.
2. Match the model to your buying cycle. Impulse products with one- or two-visit journeys: the models will mostly agree, and last touch is fine. Considered purchases with weeks of research: use a multi-touch model, and if you use time decay, give it a longer half-life.
3. Compare at least two models. Put first touch and last touch next to each other. Channels that look strong in both are safe bets. Channels that only look good in one are where you need to think.
4. Pick one as your default and stick with it. Switching models every week makes trends impossible to read. Choose a default for regular reporting, and use the others for spot checks.
5. Make sure the totals add up. Any model you use for revenue and ROAS should split each order once, so total credited revenue equals real revenue. Participation and the ad platforms' own reports don't.

Common mistakes when comparing models

• Comparing models on different data. If you compare last touch in one tool against linear in another, you're also comparing two different sets of tracked visits. Compare models on the same data, or the differences mean nothing.
• Adding up platform reports and calling it a model. Each ad platform credits sales using its own rules and only sees its own touchpoints. Summing them double counts orders that touched more than one channel.
• Picking the model that makes your favorite channel look best. It's tempting, especially when reporting up. It also leads to the worst budget decisions.
• Forgetting that attribution isn't proof of cause. Every model here shares out credit among touchpoints that were present. None of them prove a sale wouldn't have happened without a given ad.

How Soran handles it

Switching models in Soran (shown on a demo store). Every number in the report updates to the model you pick.

Soran lets you switch between ten attribution models on the same data, including first touch, last touch (the default), first and last touch paid only, linear, U-shaped, J-shaped, inverse J-shaped and time decay with a half-life you can change. Every number in the report updates when you switch.

Each order's credit is split once across the visits that led to it, so totals add up to what your store actually took. Channel, campaign, ad set and ad views all load together, and spend is matched to the exact ad behind each order, so you can see how an individual ad's ROAS changes from one model to the next.

Frequently asked questions

What are the main types of marketing attribution models?

There are three families. Single-touch models, like first touch and last touch, give all the credit to one touchpoint. Rule-based multi-touch models, like linear, position based and time decay, split credit using fixed rules. Algorithmic or data-driven models estimate the split statistically from your data.

What is the most common attribution model?

Last touch (last click) is one of the most common, mostly because it's simple and has been the default in many tools for years. Simple isn't the same as accurate, though, because it ignores every touchpoint before the final one.

Which attribution model is best for ecommerce?

For most stores running paid ads, a multi-touch model such as position based or linear gives the most balanced view for budget decisions, with first touch and last touch as spot checks. If your customers usually buy within a visit or two, the models will mostly agree and last touch is fine.

What's the difference between first-touch and last-touch attribution?

First touch credits the touchpoint that introduced the customer. Last touch credits the one right before the purchase. On the same order, they often point to completely different channels, which is why comparing them is so useful.

Can I change my attribution model later?

Yes, and in a good tool it's instant, because the model is applied when the report is calculated rather than baked into the data. Just keep one model as your default for reporting so trends stay comparable over time.

Want to see your own orders under every model at once? Book a demo and we'll show you on your store's data.