Glossary - Acquisition and attribution

Attribution model

An attribution model is the rule a measurement system uses to assign credit for a sale across the marketing touchpoints that preceded it: last click, first click, linear, time decay, position based, or data driven. Ecommerce operators use the model to decide which channels appear to drive revenue and where budget moves next.

How it is measured

There is no formula. An attribution model is a credit-allocation rule applied to the touchpoints a system observed, and every rule produces a different answer from the same journey.

Last click	100% of credit to the final click before purchase. The default in most ecommerce reporting.
First click	100% of credit to the first recorded touch.
Linear	Credit split equally across every touch.
Time decay	More credit the closer a touch sits to the purchase.
Position based	Weighted to the first and last touch, the remainder spread across the middle.
Data driven	Algorithmic weighting fitted by the platform. Still correlational, not causal.

Worked example

One journey: a TikTok video view, then a Meta retargeting click, then a Google branded-search click, then an order$200 order

Last clickGoogle $200, Meta $0, TikTok $0

First clickTikTok $200

Linear$66.67 each

Position based (40/20/40)TikTok $80, Meta $40, Google $80

Same order, four different channel reportsevery ROAS changes, the facts did not

Example numbers. Whichever model you pick rewrites each channel's apparent revenue, and therefore its ROAS and per-channel CAC.

What is a good attribution model?

None is accurate in a causal sense. Every model allocates credit among touches that were observed; no model can say whether the sale needed any of them. The useful choice depends on the question: last click over-rewards demand capture like branded search, first click over-rewards discovery channels, and the multi-touch rules in between are opinions about weighting, not measurements.

The workable practice is to compare several models side by side, treat the gaps between them as information about journey shape, and settle causal questions with incrementality testing. Remember that some numbers need no model at all: MER and blended CAC divide real totals and are immune to attribution entirely.

Attribution model vs related methods

Method	What it is	Relation to attribution
Incrementality	A holdout experiment measuring conversions ads actually caused	Causal where attribution is correlational; episodic rather than always-on.
Marketing mix modeling (MMM)	Statistical estimation from aggregate spend and sales history	Estimates channel contribution with no user-level tracking or credit rule.
MER	Total revenue ÷ total marketing spend	Needs no attribution at all; cannot double-count.
ROAS	Attributed revenue ÷ ad spend	Consumes the model's output; change the model and ROAS changes with it.

Common mistakes

Treating platform-reported attribution as ground truth. Each platform grades its own homework with its own model and window.
Summing attributed revenue across platforms. Overlapping windows let several channels claim the same order, so channel totals can exceed what the store actually took.
Switching models and reading the shift as performance. Moving from last click to position based reshuffles credit; nothing about the business changed.
Judging upper-funnel channels on last click. Discovery channels rarely hold the final click, so last click systematically defunds them.
Assuming data-driven means causal. Algorithmic models still fit weights to observed, platform-visible touches; they do not run an experiment.

Frequently asked questions

None is accurate in a causal sense: every model is a credit rule over observed touches, and each rule gives a different answer from the same data. Questions of true channel worth are answered by incrementality testing, not by model choice.

Because attribution windows overlap: a buyer who touched TikTok, Google and Meta can be claimed as a conversion by all three. Summed platform-attributed revenue routinely exceeds real store revenue for exactly this reason.

No. Blended CAC and MER divide real spend totals by real customer or revenue totals, so they need no attribution. Only per-channel CAC and per-channel ROAS depend on a model's credit assignment.

Inside Blufire, S5 Acquisition & Attribution shows attribution models compared side by side, so the model choice is visible instead of silently baked into one dashboard number.

Updated July 2026