Marketing attribution models explained, with one worked example

Updated 2026-11-02

A marketing attribution model is a rule for splitting credit for a sale across the marketing that came before it. Each model answers a slightly different question, so the same sale can produce very different reports.

The easiest way to understand them is to run one sale through all of them. This post uses a single $1,000 example, then explains when each model helps and when it misleads. For the bigger picture, see our guide to marketing attribution.

The example journey

Here's a made-up buyer named Sam. Sam's company pays $1,000 for an annual plan, after four visits over five weeks:

1. Day 1: Organic search. Sam finds a blog post on Google.
2. Day 12: LinkedIn. Sam clicks a post from your founder.
3. Day 24: Newsletter. Sam clicks a link in a newsletter you sponsored.
4. Day 35: Paid ad. Sam clicks a search ad, goes to pricing and pays.

Same buyer, same four visits, same $1,000. Now let's apply the four models.

First-touch attribution

First-touch attribution gives all the credit to the first visit.

• Organic search: $1,000
• LinkedIn: $0
• Newsletter: $0
• Paid ad: $0

The question it answers: what introduces people to us?

Use it when you want to judge awareness work, or when you need to know which channels bring in new people.

Watch out for channels that find people who never would have bought without later help. First touch ignores everything after the introduction.

Last-touch attribution

Last-touch attribution gives all the credit to the last visit before the payment.

• Organic search: $0
• LinkedIn: $0
• Newsletter: $0
• Paid ad: $1,000

The question it answers: what was the final step before people paid?

Use it when you're looking at conversion work, or your buying cycle is so short that most people need only one or two visits.

Watch out for channels that collect credit for demand they didn't create. Someone who already wants you may search your name and click an ad. Last touch hands that ad the full sale. In our example, it says the blog, the founder's post and the sponsorship were worth nothing.

Linear attribution

Linear attribution splits credit evenly across every visit. With four visits, each gets 25%.

• Organic search: $250
• LinkedIn: $250
• Newsletter: $250
• Paid ad: $250

The question it answers: what was involved in the journey at all?

Use it when you want a simple, fair view of every channel that took part.

Watch out for frequent, low-value visits. A person who returns through email ten times can load one channel with credit. Linear treats a quick glance the same as a visit that changed someone's mind.

U-shaped attribution

U-shaped attribution gives 40% to the first visit, 40% to the last visit, and splits the other 20% across the middle. It's also called position-based attribution. With two visits, each gets 50%.

• Organic search: $400 (40%)
• LinkedIn: $100 (half of the middle 20%)
• Newsletter: $100 (half of the middle 20%)
• Paid ad: $400 (40%)

The question it answers: what opened and closed this journey, and what helped along the way?

Use it when buyers make several visits and you want a single balanced default.

Watch out for journeys where the middle did the real work. A demo or a webinar in the middle gets a small share under this rule.

All four side by side

Look at the paid ad row. It's worth anything from $0 to $1,000 depending on the rule. So if your team argues about the value of a channel, check that everyone's using the same model.

Notice the total row too. Each model must add up to the sale. If credits don't equal the revenue, something has gone wrong. Headstream checks this on every run and won't publish one that doesn't match its ledger.

Other models you'll hear about

Time decay. Visits closer to the sale get more credit. It suits short sales cycles.

W-shaped. It adds a heavy weight to a middle milestone, such as when someone became a lead.

Data-driven. Statistics estimate each channel's effect from your own conversion paths. It needs a lot of conversions, and the output is hard to check by hand.

Custom. You set the weights yourself. It's flexible, and it's also easy to tune until the result matches what you hoped for.

These can be useful. Start with the four above, though, because you can explain them to anyone in a sentence.

How to choose a model

You don't need to pick just one. Each model is a lens, and comparing them tells you more than any single report:

• Strong in all four: the channel is probably working. Keep it.
• Strong in first touch only: it brings people in. Judge it on new visitors and pipeline. Don't cut it for a weak last-touch number.
• Strong in last touch only: it may be capturing demand that already exists. Test what happens when you reduce it.
• Weak in all four: look at cutting it.

If you need a single default for reporting, use U-shaped when buyers make several visits and last touch when most buy in one visit. Then stay with it, so month-to-month changes reflect your marketing and not your model.

What no model can tell you

Every model divides up credit for what happened. None of them proves what would've happened without the marketing. Sam might have bought without the ad.

Models also rely on what tracking recorded. Word of mouth and dark social don't appear, so add a how did you hear about us question to fill the gap. To test cause and effect, you need an experiment, such as pausing a channel in one region.

Compare the models on your own data

Revenue attribution in Headstream runs all four models hourly on your Stripe payments, so you can switch between them on the same data. Connecting Stripe is in beta. Linear and U-shaped are covered further in our guide to multi-touch attribution, and you can see how we calculate each model in our methodology.