“We are extremely thankful to Lityx for the valuable guidance and results provided to our client.”
– Project Manager, International Consultancy
Our client was a retail industry consulting firm delivering analytics and modeling to predict customer behavior. Collaboratively, we engaged with an international online grocery business which was currently successful but facing stiff competition and new challenges.
The grocery firm had compiled a large detailed data warehouse with information such as customer demographics, line item transactions, marketing activity and responses, customer survey data, and more. The data was not being leveraged beyond simple business intelligence.
Recently, it was noticed that a growing number of customers were making a single visit and not returning. Loyalty was decreasing and churn was increasing. The goal was to increase sales and profitability through increased customer visits (loyalty) and increased spend per visit. In addition, it was increasingly important to identify at-risk customers early in their lifecycle so that action could be taken to make them more loyal.
Data: LityxIQ was used to bring together approximately twenty client data tables into a single comprehensive view of customer activity and behavior that could be used for predictive modeling. A major challenge was to create a view of past customer behavior at any historical point in time. This is necessary to ensure that the data being used to model behavior has a clear path from the past information and insight to future behavior. In addition, the grocery firm did not previously have definitions of certain key behaviors such as churn. We tested multiple ways of evaluating churn by looking at different options in the lapse in time between purchases.
Predictive Models: A series of predictive models was developed for the grocery firm. These included:
- Multiple churn models to predict the likelihood of winning back a lapsed customer.
- Prediction of a prospect’s potential future value.
- Prediction of a customer’s likelihood to re-purchase and their future value based on first-visit data.
Among these models, the last was quite innovative and unique for the client. They were able to identify key segments of their new customer population who were at risk of not re-purchasing based only on simple demographics and first transaction metrics. Information on time-of-day, transaction amount, and basket categories from the first transaction were found to be highly predictive of future potential.
Churn models with 330% lift in identifying at-risk customers
265% lift in predicting re-purchase behavior following first visit
Easy capability to build modeling dataset for any historic point-in-time