Guide · Revenue attribution

Attribution Models Compared With One Worked Example

First touch, last touch, last non-direct, linear, time decay and data-driven attribution applied to the same customer journey, so you can see how credit moves.

· 3 min read

Traffic source lines meeting at a revenue target

Attribution models are easier to understand with one journey and real numbers. Here is a single $120 sale, run through six models.

The journey

  1. 1Day 1: Google searchReads a comparison article
  2. 2Day 4: X postClicks a link in a thread
  3. 3Day 9: newsletterOpens the launch email
  4. 4Day 10: directTypes the URL and buys
One customer, four visits, one $120 purchase

How each model splits the $120

ModelGoogleXNewsletterDirect
First touch$120
Last touch$120
Last non-direct touch$120
Linear$30$30$30$30
Time decay$12$22$41$45
Data-drivendepends on your data

*Time decay figures are illustrative: recent visits get more credit, and the exact split depends on the half-life chosen.*

What each model is good for

First touch. Rewards discovery. Use it to judge content and SEO that bring new people.

Last touch. Rewards whatever came last. On small sites that's often "direct", which hides the real channels. Rarely useful alone.

Last non-direct touch. Rewards the channel that brought the buyer back when ready, ignoring direct. A practical default for small teams, and what DataFair uses. First touch vs last touch explains why.

Linear. Everyone gets an equal share. Fair-sounding, but it treats a passing glance like the visit that sealed the deal.

Time decay. Recent visits count more. Suits long sales cycles with many touches.

Data-driven. A model learns from many conversions which touches tend to matter. It needs volume, a lot of it, and it's a black box: hard to explain to a co-founder.

Choosing one

  • Fewer than a few hundred sales a month: last non-direct touch. Simple models with honest data beat clever models with sparse data.
  • Many paid channels and plenty of volume: consider data-driven, alongside a simple model you can sanity-check.
  • Whatever you pick: look at revenue per visitor by channel, not just totals.

The model matters less than the data

Two problems hurt every model equally:

  1. Payments that never meet a visit. If checkout doesn't carry the visitor id, the sale is unattributed under any model. Fix the handoff first.
  2. Channels no tracker sees. Podcasts, word of mouth and private messages. A short "how did you hear about us?" question fills that gap better than any model.
Revenue by channel in DataFair
Revenue by channel in the DataFair demo (sample data).

Related guide: Revenue attribution for every payment provider.

Questions

What is the most accurate attribution model?
None is accurate in every case. Each answers a different question. For small teams, last non-direct touch is a practical and explainable default.
What is a multi-touch attribution model?
A model that splits credit for a sale across several visits, such as linear, time decay or data-driven.
Can I compare models in DataFair?
DataFair reports last non-direct touch. For a first-touch view of subscriptions, recurring revenue by source credits each subscription to the source of its first payment.

Related