Attribution modeling

Marketing Analytics for Business

Sarah DeAtley

Principal Data Scientist

What is attribution modeling?

 

image titled "Paths to Purchase" showing path 1 and path 2 ending with dollar sign icons

 

Attribution modeling: determine how much credit to give to marketing efforts upstream from decision

  • Analysts select attribution methodology for their marketing program

  • Attribution influences optimization, ROI modeling, and marketing investments

Marketing Analytics for Business

Last touch attribution

Last touch attribution (LTA): 100% of credit goes to the last channel / tactic / campaign

  • It is very easy to understand this method and explain trends
  • Channels with Direct impact naturally get the most credit
  • Indirect channels get low credit because they are rarely last in the purchase process

icons showing path of search then email then search with 100%, then conversion

Marketing Analytics for Business

Multi-touch attribution

Multi-touch attribution (MTA): giving credit to multiple marketing efforts that led to a customer purchase

  • Multi-touch attribution can account for Direct and Indirect impact
  • MTA makes it challenging to identify which channel is driving marketing trends

path with 30%, 10%, 10%, and 50% credit then dollar sign

Marketing Analytics for Business

Heuristic and algorithmic MTA

MTA can be heuristic (rules-based) or algorithmic (statistical model-based), while adding up to 100%

  • Heuristic example: linear splits credit equally between all channels

icon showing 3-way split

  • Algorithmic example: regression modeling ranks influence of channels based on strength of relationship to KPI

icon showing normal distribution

Marketing Analytics for Business

Time decay attribution

Time decay attribution: heuristic MTA where rules assign gradually more credit based on recency

 

graph of Time Decay attribution showing increasing credit from 10% to 20% to 30% to 40%

Marketing Analytics for Business

Choosing an attribution model

"All attribution models are wrong."

  • No single correct model, but need to evaluate trade-offs of each
  • Consider need for precision, ease of interpretation, customer behavior, channel mix, and data availability
  • Privacy law impact: algorithmic can mitigate when data is not tied to customer directly

icons representing Last Click, Time Decay, Linear, and Algorithmic MTA

Marketing Analytics for Business

Let's practice!

Marketing Analytics for Business

Preparing Video For Download...