Data Insights and Decisions COMM1190

The Head of Management Services of Freshland, a large grocery store chain in Australia, has made use of your

updated report (from Assessment 1) to deliver a presentation to the Senior Executive Group.

→ Access this presentation via your Moodle course site.

Based on this initial analysis and recommendations, approval has been given for further analysis of customer loyalty

and churn using an expanded data set. The core task will involve a comparison of predictive models and subsequent

recommendations on how to use and improve these to inform future retention policies.

The analysis in the presentation to the Senior Executive Group was based on the initial pilot data set which was used

by the intern to produce the initial report and was part of the data provided to you with Assessment 1. These data

have now been extended, with extra variables being added. These extra variables are:

ltmem

=1 if 𝑚𝑒𝑚𝑏𝑒𝑟 ≥ 3

mamt1

Average monthly expenditure ($) in first 6 months of previous year (2023)

mamt2

Average monthly expenditure ($) in second 6 months of previous year (2023)

fr1

Frequency of monthly transactions in first 6 months of previous year; 1 (low) 2 (medium), 3 (high)

fr2

Frequency of monthly transactions in second 6 months of previous year; 1 (low) 2 (medium), 3 (high)

rind

XYZ risk index in the form of a predicted probability of customer churn

You and your team have been tasked with investigating alternative algorithms for predicting customer churn. Given

the structure of data that has been made available, you have been advised to define churning to be when a customer

has previously had non-zero transactions for at least 6 months but then has zero transactions in the next six-month

period. The outcome of interest is the binary variable churn . Given the available data, an observation for a customer

will have 𝑐ℎ𝑢𝑟𝑛 = 1 if 𝑚𝑎𝑚𝑡1 > 0 & 𝑚𝑎𝑚𝑡2 = 0  and 𝑐ℎ𝑢𝑟𝑛 = 0 if 𝑚𝑎𝑚𝑡1 > 0 & 𝑚𝑎𝑚𝑡2 > 0.

The Head of Management Services has given you authority to use your expert judgment to make the necessary

modelling choices but has outlined an overarching research plan for you and your group to follow:

• Currently, Management Services has a basic regression model (details below) that can be used to predict

future customer expenditure for members of the rewards program. It has been suggested that this could be

used to generate a risk index where those with predicted expenditures that are low relative to actual

expenditures being deemed as high risk of no longer shopping at the store.

• However, there were suggestions that the existing model could be improved as a predictor of expenditures

and your group has been asked to evaluate a range of model extensions.

• The current focus is on predicting churn. Based on the performance of the alternative models in predicting

expenditures, choose one and analyse whether it also performs well in predicting churn.

• An analytics firm, XYZ, that uses proprietary predictive methodology has offered a trial of their products by

providing a predictor of churn. Your evaluation of predictive performance should include a comparison of this

predictor with that generated by your chosen regression-based predictor.

• Based on this analysis, make recommendations on using such algorithms in initiatives targeting customers at

risk of churning with the aim of retaining them as loyal customers.

o Notice that any recommendation to employ the predictors of the analytics firm would involve

additional cost compared to a method produced in-house by Management Services.

o In addition, any decision to employ the predictors of XYZ will not include documentation of the

methodology used to generate the predictions.

o It might also be that you conclude that neither predictor is adequate and that it would be appropriate

to explore alternative predictors or approaches. You are not expected to explore such alternatives.