Skip to content
Tutorials Center
My learning

Topics

Lesson 25 of 38 Β· Analytics and Attribution

Attribution Models and Incrementality

Understand marketing attribution: first-click, last-click, linear, time-decay, position-based and data-driven models, plus incrementality tests, with a runnable calculator.

A customer rarely buys after a single contact. They may see a social video, read a review, click a search ad, get an email and finally buy after typing your name into a search bar. Which of those gets the credit? Deciding that is attribution, and the answer changes where you spend your budget.

Touchpoints and conversion paths

A touchpoint is any recorded interaction with your marketing. A conversion path is the ordered list of touchpoints before a conversion. For example:

Social video -> Organic search -> Paid search -> Email -> Direct (purchase)

An attribution model is a rule that splits the conversion's credit across those touchpoints.

The common models

  • Last click: all credit to the final touchpoint. Simple, but it favors channels near the end (brand search, email, direct) and undervalues discovery.
  • First click: all credit to the first touchpoint. It rewards discovery, but ignores what closed the sale.
  • Linear: equal credit to every touchpoint.
  • Time decay: more credit to touchpoints closer to the conversion, using a decay curve (for example a half-life of seven days).
  • Position based (U-shaped): 40 percent to the first, 40 percent to the last and 20 percent split between the middle ones. Variants like W-shaped add a milestone such as lead creation.
  • Data-driven: a statistical or machine-learning model estimates each touchpoint's contribution from many converting and non-converting paths. It needs lots of data and is often a black box.

Who gets the credit?

First click Rewards discovery
Last click Rewards closing
Linear Equal share to all
Time decay Recent touches count more

No model is the truth; each is a lens.

1 / 2

Compare models in code

JS Attribution calculator

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Look at how "social" and "display" gain under first-click, linear and position-based models, while "email" dominates under last click. The same sales, a different story.

Windows and platform differences

  • Lookback window: how far back a touchpoint can earn credit (for example 30 or 90 days).
  • Click versus view: some platforms count view-through conversions; others only clicks.
  • Every platform reports its own conversions, and it will claim any conversion it touched. Add the totals from Google, Meta and TikTok and you will often exceed your actual sales. Treat platform dashboards as optimization signals and your analytics or finance data as the check.

Beyond rules: incrementality

All the models above only distribute credit among touchpoints that happened. They cannot answer the question that really matters: would the sale have happened without this marketing? That is incrementality.

Ways to measure it:

  1. Holdout tests: hide the ads from a random group and compare their conversion with the exposed group.
  2. Geo experiments: run ads in some regions only and compare.
  3. Conversion lift studies: offered by some platforms; they use randomized control groups.
  4. Marketing mix modeling (MMM): statistical models using aggregate spend and sales over time, which works without user-level tracking.
JS Incremental lift from a holdout

Press Run (or Ctrl + Enter) to see the console output here.

No output.

To check whether the difference is real or luck, use the test from A/B testing.

Practical advice

  • Use two or three models side by side; where they disagree, investigate.
  • Ask customers how they heard about you ("self-reported attribution"). It finds podcasts, word of mouth and dark social that no link can.
  • Keep UTM tracking clean; every model is only as good as its inputs.
  • Privacy rules and browser limits mean some paths are broken. Accept that attribution is an estimate.

Common mistakes

  • Cutting top-of-funnel channels because last click gives them zero credit.
  • Adding up platform-reported conversions.
  • Treating a model change as a real performance change.
  • Over-trusting a data-driven model with little data.

Check your understanding

Quick check Β· 4 questions

of 4 answered You scored / 4 Β· perfect!

  1. 1. Which model gives all credit to the final touchpoint?

    Last click credits only the last interaction before conversion.
  2. 2. A U-shaped (position based) model commonly gives what share to the first and last touch combined?

    40 percent each, leaving 20 percent for the middle touches.
  3. 3. Why do platform-reported conversions often add up to more than real sales?

    Platforms attribute independently, so the same sale can be counted several times.
  4. 4. What does a holdout test measure?

    Comparing exposed and unexposed groups estimates incremental effect.

Summary

  • Attribution models split conversion credit across touchpoints; none is the truth.
  • Last click favors closers, first click favors discovery; position and time-decay blend them.
  • Platform numbers overlap; use analytics and finance data as a check.
  • Incrementality tests answer whether marketing truly caused the sale.

Next: Dashboards and Reporting