Multi-touch attribution: what it is and how to use it

Updated 2026-11-02

Multi-touch attribution splits the credit for a sale across more than one touchpoint. A buyer who finds you through search, returns from LinkedIn and pays after an email gives all three some credit. Single-touch models pick one and ignore the rest.

If people usually visit you several times before they buy, multi-touch attribution gives you a fuller picture. It also has limits, and you should know them before you put it in a board deck. For the basics, start with our guide to marketing attribution.

Single-touch versus multi-touch

Single-touch models give 100% of the credit to one visit. First touch picks the first. Last touch picks the last. They're simple, and they're blind to everything else.

Multi-touch models divide the credit. They come in two kinds: rule-based, where you set the split, and data-driven, where statistics estimate it.

Here's why that matters. Say a made-up company earns $10,000 in a month from 10 buyers, and every buyer visited through search, then LinkedIn, then email. Last touch gives email all $10,000, and LinkedIn and search get nothing. If you trusted that report, you might stop posting on LinkedIn and let your blog go quiet. New buyers would drop off a few months later, and you wouldn't see why.

The main multi-touch models

Linear

Linear attribution gives every visit an equal share. With three visits and a $300 sale, each gets $100. It's the easiest to explain. Its weakness is that it treats each visit as equally important.

U-shaped

U-shaped attribution gives 40% to the first visit, 40% to the last, and splits 20% across the middle. For a $1,000 sale with five visits:

It's a sensible default, because the first visit and the last are often the ones that matter most.

Time decay

Visits closer to the sale get a larger share. It suits short cycles, and it can undervalue the channels that opened a long B2B journey.

W-shaped

This adds a third heavy weight to a middle milestone, such as the moment someone became a lead. You need to record that milestone for it to work.

Data-driven

Statistical models compare paths that converted with paths that didn't. In theory, this finds each channel's real effect. In practice, it needs many conversions, and you often can't check how it reached a figure. For a company with 50 sales a month, it's rarely the right first step.

You can see all of these rules applied to a single sale in marketing attribution models explained.

What you need to run multi-touch attribution

A visit history for each buyer. This needs a stable visitor ID. Headstream, for example, uses a first-party cookie that can last up to 400 days.

A link between visits and people. An identify call at signup joins anonymous visits to an account.

A reliable revenue source. For SaaS, that means your billing system. Pull in refunds and disputes too.

Consistent campaign tags. Your UTM parameters need to follow one format. If LinkedIn appears as "linkedin", "LinkedIn" and "li", it'll be split across three rows.

A sensible lookback window. Pick a lookback window that fits your sales cycle. Visits outside it get no credit.

How to read a multi-touch report

Compare it with single-touch reports on the same data. That's where the insight comes from.

Here's a made-up quarter of $50,000 in revenue:

Reading it row by row:

• Organic search looks stronger the earlier you look. It starts journeys.
• LinkedIn is similar. Last touch alone would make it seem nearly worthless.
• Email closes journeys and rarely starts them. It's useful, and it isn't creating demand on its own.
• Paid search does a bit of both, with a lean toward closing.

So you wouldn't cut LinkedIn. You also wouldn't credit email with building your pipeline. Those are decisions a single model would've gotten wrong.

Limits of multi-touch attribution

It only sees what's tracked. Podcasts, word of mouth, private shares and offline events leave no record. Ad blockers hide some visits. See our guide to dark social.

Credit doesn't equal cause. Multi-touch attribution splits up credit for a sale that happened. It can't say whether the sale would've happened anyway. Channels that reach people already likely to buy will look better than they are. Only a holdout test can measure that.

Rule-based weights are a choice. The 40/40/20 split is a convention. Nobody discovered it. Treat the output as a useful view.

Devices split journeys. A phone visit and a laptop visit look like two people until someone is identified.

Privacy rules and browsers can shorten what you see. Check how your setup handles consent and how long your visitor IDs last.

These limits don't make multi-touch attribution useless. They just mean you should pair it with other signals, like self-reported data.

Add what people tell you

A how did you hear about us question picks up what tracking misses. If your tracking shows a buyer's first touch as Direct and the buyer writes "a podcast", you've learned something no model could have told you.

Should you use multi-touch attribution?

Use it if:

• People usually visit several times before buying.
• You run more than one channel.
• You make budget choices between those channels.

Skip it for now if:

• You have very few sales. With 10 a month, read the journeys by hand.
• Most buyers convert on their first visit. Single-touch will do.

How Headstream handles it

Revenue attribution in Headstream runs linear and U-shaped alongside first touch and last touch, every hour, on your Stripe payments. Connecting Stripe is in beta. The default lookback is 90 days, and you can adjust it.

Credits are allocated to the cent and must add up to the payment ledger before a run publishes. Refunds and disputes are credited back to the original sale, so the same split reverses. You can trace any payment to the visits behind it on your trust pages. The full rules are in our methodology.