Touch sequence — credit by attribution method
All five touches are measurable: ad clicks, email opens, postcard QR scans, and event RSVPs. Brand awareness (TV, billboards, branch signage) is real and matters — but it's not in this table because nobody captures the impression event.
How the custom model assigns weights
For each touch, the model computes its marginal contribution to the conversion across every possible combination of touches — then averages.
v(∅) = 0 // no touches → no conversion baseline
v({FB}) = 0.05 // facebook alone → 5% lift over base
v({FB, Email}) = 0.18
v({FB, Email, Chamber}) = 0.71
... 32 combinations total for 5 touches ...
Email's marginal contribution =
Σ over every combination S excluding Email:
[ v(S ∪ {Email}) − v(S) ] · weight(S)
Email's credit = the average across all those combinations.
The "fair" allocation — but expensive. Brute force on 5 touches is fine; 20+ touches needs an approximation.