Attribution Model Simulator
Build a customer journey, then watch five attribution models fight over the credit for one conversion.
Interactive · MTA vs. LTA
Who gets the credit?
Build a customer journey. Compare how five attribution models split a single conversion.
How it works
What it computes
You build an ordered journey. Each model assigns weights across positions that sum to 1, and a channel's credit is the sum of the weights at its positions:
- Last-touch / first-touch: all weight on the final / opening position.
- Linear: 1/n to each of n touches.
- Time-decay: position i gets weight proportional to 2^(−(n−1−i)/1.5), a half-life of 1.5 touches counted back from the final touch.
- U-shaped: 40% first, 40% last, the remaining 20% split evenly across the middle (100% for a single touch, 50/50 for two).
Assumptions
- One journey, one conversion of fixed value. Position order is the only input: no timing, no spend, no channel quality.
Where it breaks
- All five are bookkeeping rules over observed touch order. None measures incrementality: a channel can collect most of the credit while driving zero incremental conversions.
- Touch order is itself selected. Retargeting and brand search show up late because bidding puts them in front of people already about to convert, and last-touch rewards exactly that.
- Real platforms decay by time (a 7-day half-life, say), not by touch count. Data-driven attribution (Shapley, Markov) needs thousands of journeys, and even those stay correlational.