Attribution models: first click, last click and data-driven

Most customers visit your shop more than once before they buy. They might click an ad, come back from a search, and finally buy from a newsletter link. An attribution model decides which of those visits gets the credit for the order. AxiTrace gives every order exactly one source, and the model decides which one.

The three models

First click

Credits the visit that brought the customer to your shop for the first time in the 365 days before the order. It answers the question "what introduced this customer to us?". Use it to judge the campaigns that find new customers. First click is the default.

Last click

Credits the most recent visit before the order that came from a campaign, an ad or another website. Visits where the customer typed your address or used a bookmark are skipped. It answers the question "what brought them back to buy?". Use it to judge reminders such as newsletters and retargeting.

Data-driven attribution

Ranks acquisition visits by recorded engagement and shopping actions. Model dda-2026.10.3 excludes direct returns after a known external source. An initial direct visit can still win. Later orders inherit that acquisition until a new engaged visit from any known external source renews it. This ranks recorded evidence of influence; it does not establish causation.

One journey, three answers

Here is one customer who buys a lamp. Every model looks at the same three visits:

When	How the customer arrived	What they did
Day 1	Clicked a Facebook ad	Browsed for about 5 minutes, looked at 6 products including the lamp, and added the lamp to the cart
Day 12	Clicked a Google search result	Left after a few seconds without opening a second page
Day 13	Clicked a link in your newsletter	Went from the cart straight to checkout and paid

Model	Credited visit	Why
First click	Facebook	It was the first visit.
Last click	Newsletter	It was the most recent visit before the order.
Data-driven	Facebook	It had by far the strongest engagement, and the customer looked at the lamp they bought during it.

Why data-driven picks Facebook, visit by visit:

• Facebook, day 1: the longest visit, the most products viewed, the lamp added to the cart. This visit provides the strongest recorded evidence of shopping intent.
• Google search, day 12: a visit of a few seconds on one page barely counts. It keeps a small recency contribution, but no behavioral credit.
• Newsletter, day 13: paying is not persuasion, so the time spent in checkout is not counted. Email receives no blanket discount; meaningful browsing before checkout competes normally.

When data-driven agrees with last click

Data-driven attribution does not always pick the first visit. When the most recent visit is the one where the customer did the real shopping, it wins:

When	How the customer arrived	What they did
Day 1	Clicked a TikTok ad	Looked at two pages for about half a minute and left
Day 20	Clicked a Google ad	Browsed for about 7 minutes, looked at 5 products including the one they bought, added it to the cart and bought it in the same visit

First click credits TikTok. Last click and data-driven both credit Google: the Google visit was recent and had much stronger engagement. The short TikTok visit three weeks earlier did introduce the customer, and first click is the model that shows that.

What data-driven looks at

For every eligible visit, including Direct, the new model looks at:

• How much time the customer actively spent on your site
• How deep into the shop they went before checkout: pages, distinct products, internal search and cart actions
• Trusted clicks and scrolling
• Whether they viewed or added to cart a product they later bought
• How recent the visit was

Engagement matters most. Recent visits count more than old ones. A few more rules keep the result fair:

• Initial direct visits can compete. A direct visit after a known external source cannot take attribution credit. Its activity is not transferred to that source.
• Acquisition carries across orders. A new engaged visit from ads, organic search, email, a comparison site or another known external source can renew it. Direct returns and visits without qualifying engagement do not. Inheritance has no 90-day expiry; each new acquisition comparison uses up to 90 days of visit evidence.
• Checkout and earlier purchases are boundaries. Timestamped activity must belong to this purchase cycle and precede checkout. Cumulative measurements spanning a boundary are excluded from time and interaction comparisons.
• Behavior receives 90% of the weight. Active time 30%, funnel depth 35%, interactions 10%, purchased-product actions 15%. Recency stays 10%, even when measurements are missing.
• Meaningful actions count. Internal searches and adding the purchased product to cart contribute; repeated views of the same product, repeated searches and a key press or touch alone do not earn extra bonuses.
• No source or position bonus. Email and purchase visits are not discounted. An untagged organic return after an earlier advertising visit retains the 0.7 multiplier.
• Missing measurements remain unknown. Ranking uses comparable signals across visits and reallocates unavailable weights within the behavioral budget. A single candidate or incomplete evidence never automatically produces high confidence.

What you see in AxiTrace

Each order is first scored about an hour after it is placed. Stored verdicts remain fixed; background jobs do not recalculate historical results. The new model applies only to purchases placed after its activation. When data-driven attribution is selected, the Orders list can show these badges next to the source:

Badge	What it means
Pending	The order has not been scored yet. The first-click source is shown until then.
No journey data	No visits are linked to the order, so the first-click source is shown.
All visits direct	A historical verdict with only direct candidates. The new model can select and credit a particular direct visit.
Counts only	Time and interaction measurements were incomplete, so the ranking uses comparable browsing and shopping actions with limited recency influence.
Low confidence	The ranking is close or qualifying behavioral evidence is missing. Treat the result as uncertain.

Open an order to see its Customer Journey. The credited visit is marked Data-driven winner in the timeline, and Why this visit lists the reasons, the top visits and the runner-up. The Customer Journey shows this for every scored order, whichever model you use by default, so you can always compare.

Orders placed before data-driven attribution was enabled are not scored. They keep their first-click source and show no badge.

How to switch

• For the whole workspace: open your workspace, go to Workspace settings → Attribution and pick a model on the Default attribution model card.
• For one view: use the Attribution model dropdown on the Orders list in Orders & journeys.

You can switch back at any time. All three models are worked out for every order in the background, so switching is instant and nothing is recalculated or lost.

Where the model applies

• Applies to: every report that credits orders or revenue: the Orders list in Orders & journeys, Recent orders, the Facebook, Google Ads and TikTok campaign cards and platform pages, the Campaigns module, Products, the Traffic Sources breakdown, the sources table on the Overview, Profit by platform, the Customer Journey, the public API and the AI assistant connection. Each order is credited to exactly one source or campaign, so an order that two ad platforms touched is never counted twice, and the campaign cards, Traffic Sources and Orders & journeys agree order for order. The Orders list also keeps its own Attribution model dropdown for a single view.
• Totals stay the same: revenue, orders, sessions and spend for the whole workspace are the same under every model. The model only decides which source, campaign or ad an order is credited to.
• One first click: first click means the same everywhere, the customer's first visit in the 365 days before the order, including a direct visit.
• Does not apply to: clicks and visits, which count the visits inside the selected date range, whatever model you pick. It also does not change the conversions sent to your ad platforms: Facebook, Google and the others receive the same data as before and apply their own attribution rules.

Visits from AI assistants such as ChatGPT, Perplexity, Gemini, Claude and Microsoft Copilot are a source like any other: every model can credit them, and they are grouped as AI assistants. Traffic sources, Orders & journeys and Products can be filtered to that group. Which assistants are recognised and how is listed in AI assistant traffic.

On the Business and Custom plans you can also put two models side by side for the same days: see Comparing attribution models side by side.

Not the same as Google's data-driven attribution

Google Ads and Google Analytics have a model with the same name, but it works differently. Google compares customers who bought with customers who did not, needs a lot of conversions to do it, and splits the credit between its own touchpoints. AxiTrace looks at each order on its own with fixed, published rules, works from your first order, covers every channel you track and gives the whole order to one visit.