Skip to main content

Combine Media-Mix model attribution with Multi touch attribution at path level

INPUT
  • Customer journeys
  • Channel level attribution for digital channels from a Media Mix model
  • Prior weights for Multi touch attribution (MTA), one for each digital channel. Weights are used to combine Multi Touch attribution (MTA) and Media Mix model attribution (MMMA) at the channel level
OUTPUT
  • Path-level attribution
ALGORITHM

Suppose we have 3 digital channels for which we collect customer journeys:

A > B > A > C > (CONV)
A > B > B > B > (NULL)
...

Suppose the results from a MMMA on the 3 digital channels are:

CHANNELMMMA (# CONVERSIONS)
A1.230
B765
C890

Suppose that prior weights chosen by the user for MTA channels are:

CHANNELMTA WEIGHT
A0.5
B0.3
C0.5

where weigth 0.3 for B means that combined attribution at channel level for B will be 0.3 MTA(B) + 0.7 MMMA(B):

  1. Run an MTA model with ChannelAttribution Pro (heuristic, markov, shapley, logistic regression...) obtaining attribution at path-level

  2. Aggregate path-level attribution at channel level

CHANNELMTA (# CONVERSIONS)
A935
B856
C905
  1. Normalize MMMA and MTA
CHANNELMMMAMTA
A1.230 / (1.230+765+890)=0.43935 / (935+856+905)=0.35
B765 / (1.230+765+890)=0.27856 / (935+856+905)=0.31
C890 / (1.230+765+890)=0.30905 / (935+856+905)=0.34
  1. Combine MMMA and MTA using MTA weights
CHANNELCOMBINED ATTRIBUTION
A0.430.5 + 0.350.5 = 0.39
B0.270.3 + 0.310.7 = 0.29
C0.300.5 + 0.340.5 = 0.32
  1. Find path-level attribution weights from the combined channel-level attribution

θ^:∥CPLA(θ^)−CLA∥=min⁡θ∥CPLA(θ^)−CLA∥\hat{\theta}: \| CPLA(\hat{\theta}) - CLA \| = \min_{\theta} \| CPLA(\hat{\theta}) - CLA \|

where CLA is the normalized channel-level attribution at 4., CPLA() is the aggregated at channel-level and normalized path-level attribution, is the vector of attribution weighs to be used to perform path-level attribution.

  1. Perform path-level attribution using θ^\hat{\theta}