Lesson 22 of 30 · 11 min
Attribution models and their lies
Every model is a rule for assigning credit, and every rule is wrong somewhere.
Attribution is a convention, not a measurement
No system observes causation. Attribution applies a rule — last click, first click, linear, time decay, position-based — to divide credit between touchpoints. Choosing a model is choosing which channels will look good, so the choice deserves more thought than it usually gets.
Last click flatters the harvest
Last-click credits whatever was nearest the conversion, which is almost always branded search or retargeting. It systematically overstates capture channels and understates everything that created the demand. It is the default in most tools, which is why so many plans drift toward the bottom of the funnel.
First click flatters discovery
Flipping to first click simply moves the bias: now the introduction takes all the credit and the work that closed the sale takes none. Neither extreme is more honest than the other — they just favour different teams.
Use models to compare, not to conclude
The useful move is to view the same period through two or three models. Channels that look strong under every model are genuinely strong. Channels that only look good under one are being flattered by that model's rule.
Know what your model cannot see
Attribution misses offline conversations, word of mouth, cross-device journeys, and anyone who blocks tracking. Those gaps are not small, and they are precisely why attribution should inform decisions rather than settle them.
Takeaway
Treat attribution as a lens, not a verdict — and trust channels that hold up under more than one model.
Check yourself
No score, no signup — pick an answer to see why it's right.
Question 1
Which channels does last-click attribution systematically overstate?
Question 2
A channel looks excellent under one attribution model and weak under two others. What is the reasonable read?