Attribution models explained, with one buyer's path worked through

An attribution model is a rule for splitting the credit for a sale between the touches that came before it. Every report that says "this campaign drove $X" is using one, whether it says so or not. Change the model and the same sales move between channels.

The clearest way to understand the models is to run one path through all of them.

One path, five answers

A buyer clicks a Meta ad on day 1 and browses. On day 5 they search your brand on Google and click the result. On day 9 they click a link in your email. On day 10 they type your address and buy, for $120.

Illustrative example

Meta ad, day 1Google search, day 5Email, day 9Direct, day 10

First touch$120$0$0$0

Last click$0$0$0$120

Last non-direct click$0$0$120$0

Linear$30$30$30$30

Position-based (40/20/40)$48$12$12$48

Worked example: one $120 order after four touches, split by five rules. Position-based gives 40 percent each to the first and last touch and splits 20 percent across the middle ones.

Model	Meta ad, day 1	Google search, day 5	Email, day 9	Direct, day 10
First touch	$120	$0	$0	$0
Last click	$0	$0	$0	$120
Last non-direct click	$0	$0	$120	$0
Linear	$30	$30	$30	$30
Position-based (40/20/40)	$48	$12	$12	$48

The same $120 goes to Meta, to email, to direct, or gets split, depending only on the rule. None of these answers is "true". Each is a different question:

• First touch asks: what introduced the buyer?
• Last click asks: what happened right before the sale?
• Last non-direct click asks: what was the last marketing touch we know of?
• Linear asks: what was involved at all?
• Position-based asks: what opened and closed the deal, with some credit for the middle?

The rule-based models

First touch

All credit to the first known touch. Good for judging which channels bring new people in, such as prospecting ads, podcasts, partnerships and SEO. It ignores everything that convinced the buyer later, and it depends on how far back your data reaches. A tool that forgets visitors after 7 days will call a returning buyer's day-8 visit their "first touch".

Last click

All credit to the final touch. Simple and easy to check, which is why most tools started there. It overrates channels that sit at the end of the path, like branded search, email to existing customers and retargeting, and underrates whatever created the demand.

Last non-direct click

Last click, but a direct visit never takes the credit if any marketing touch came before. Shopify uses it as its default (Shopify). It fixes the worst flaw of last click, since "typed the address" is rarely the reason someone bought.

Linear

Equal credit to every touch. Fair to every channel involved, but it treats an ad glimpsed once the same as the email that brought the buyer back with a discount.

Position-based

Usually 40 percent to the first touch, 40 percent to the last, and 20 percent split across the middle. A reasonable compromise when you care about both finding and closing. The 40/20/40 split is a convention, not a measurement.

Time decay

More credit to touches closer to the sale, often halving for every 7 days further back. Suits short, considered purchases. It penalizes the touch that started a long sales cycle.

Data-driven attribution

Data-driven models learn the split from your own data: they compare paths that converted with paths that did not, and give more credit to touches that seem to raise the chance of a sale. Google Ads and GA4 use data-driven attribution, and Google retired first click, linear, time decay and position-based models from both in 2023 (Google Ads, GA4). GA4 now offers data-driven, "paid and organic last click" and "Google paid channels last click" (Google).

Two limits to keep in mind:

• It can only weigh what it sees. GA4's model sees the touches GA4 recorded. It cannot see an ad view on Meta, a podcast mention or a conversation.
• It is not transparent. You cannot check its weights by hand, which makes it hard to explain to a client or a finance team.

Every model shares three blind spots

1. Missing touches. Ad views, word of mouth, offline ads, and touches on another device that the tool could not link. A model can only split credit between touches that were recorded.
2. The window. A touch older than the lookback window gets nothing. GA4 looks back 90 days by default for most key events (Google); Meta's default is 7 days after a click.
3. Correlation is not cause. A model says which touches came before sales. It cannot say whether the sale would have happened anyway. A buyer who searched your brand name was probably going to buy.

Platform attribution is a different thing

Each ad platform runs its own attribution, inside its own walls, and it is generous to itself: Meta counts a purchase if the buyer clicked or saw a Meta ad inside Meta's window, even if Google, email and a friend were involved too. Google Ads does the same for Google clicks. That is why the platforms' conversions add up to more than your sales. See why analytics numbers never match.

Measuring what actually caused sales

To answer "would these sales have happened without the ads?", you need an experiment or a model of spend over time, not a path model:

• Lift tests. Hold back ads from a random group and compare. Meta and Google both offer conversion lift studies for eligible accounts.
• Geo tests. Pause or raise spend in some regions and compare with matched regions.
• Marketing mix models. Statistical models of sales against spend over time. Google's Meridian and Meta's Robyn are open source (Meridian, Robyn). They need a lot of history and care to run well.

For most small businesses, a simpler habit works: watch total revenue against total spend (the marketing efficiency ratio) when you change a budget, and use path attribution to decide where inside that budget to shift money.

Which model should you use?

The most useful view is two models side by side. When a channel is strong in first touch and weak in last click, it introduces buyers who convert later through something else. Cutting it because last click undervalues it is the classic attribution mistake.

Common questions

What is the best attribution model?

There is no single best one. Use first touch to judge what brings people in, last non-direct click to judge what closes, and compare the two before cutting any channel.

What attribution model does GA4 use?

GA4 offers data-driven attribution and two last-click models: paid and organic last click, and Google paid channels last click. All of them leave direct visits out of the credit unless the whole path was direct.

Why did Google remove first click and linear attribution?

Google said few advertisers used them and moved those conversion actions to data-driven attribution in 2023. Last click remains available.

What is multi-touch attribution?

Any model that splits credit across more than one touch: linear, position-based, time decay and data-driven are all multi-touch.

Sources

• Google Ads Help: First click, linear, time decay and position-based attribution models are going away. Checked October 2026.
• Google Analytics Help: Get started with attribution, Select attribution settings. Checked October 2026.
• Shopify: Marketing insights. Checked October 2026.
• Google Meridian and Meta Robyn.