BTYD - BGBB
This model was created by Bruce Hardie of UPenn.
He has an example of how to run the model in Excel here: http://www.brucehardie.com/notes/010/
The model does a good job of predicting future behavior because it matches up well with holdout data (back testing).
This model is better than other models because
- It requires a small amount of data, only RF
- The model has demonstrated good validation on lots of data
- It can be generalized to lots of situations
- Easily implemented in Excel
(1)
We will pass in a matrix of values which are derived from the underlying transaction data. Every unique combo of recency and frequency and the count of customers with those values.
The BG/BB model requires the same information as the Pareto/NBD model, but as it models discrete transaction opportunities, this information can be condensed into a recency-frequency matrix. A recency-frequency matrix contains a row for every recency/frequency combination in the given time period, and each row contains the number of customers with that recency/frequency combination.
https://rdrr.io/cran/BTYD/man/BTYD-package.html
113.020.050.30 BTYD - Pareto NBD
References:
1. https://www.coursera.org/learn/wharton-customer-analytics/home - Week 3 slides, Fader
Source:
- Me