Now Reading
Developing a Predictive Model to identify Festive Shoppers


Developing a Predictive Model to identify Festive Shoppers


festiveStudies say it’s ten times more expensive to acquire a new customer than to retain an existing customer. The retailers put their marketing efforts to attract more visits from already enrolled members. This year they approached us to plan their Diwali festive season marketing campaign with a mandate to increase the sales from non-festive shoppers and to remind festive shoppers about the new product line. Various approaches were discussed, paper highlights below showcase how we developed a predictive model to predict the repeat customers who will shop during the festive season.

Festive seasons exhibit high seasonality for retail industry. This particular retailer’s sale too increases dramatically by 20-30% during festive seasons. At any particular time, sales happening through repeat members ranges between 40-50% of the total sale. We planned to increase the sales happening through repeat customers during the forthcoming festive season. We’ve developed a predictive model to predict those customers who are most likely to churn during the festive period. So that separate marketing actions can be put in for churn customers and non–churn customers.



Research Methodology

Sample data of 5.75 lakh customers containing variables such as their lifetime sales, lifetime visits, tier classification, whether they’ve shopped during last festive season or not, their latency, frequency, number of stores they shopped at, etc. were used to understand the customer behaviour pattern. We ran a logistic regression model to predict the shoppers who visit during this festive season.

Key Insights

70% of the data is set for training and the remaining 30% is set aside for testing. Based on these factors, the summary of logistic regression for training is as follows:
Raymond Churn model example

The confusion matrix which is shown above is used to describe the performance of the model on the training data versus the true values. The columns here represent predicted class, whereas the rows represent the actual values. So in our example, out of the actual 98.3k customers who’ve shopped during the festive season, our model is able to successfully predict 79k customers.

Model Insights: Probability to shop vs Model Variables

Members who were less recent are more likely to come back to shop during the festive season. 

Raymond Churn model example

Propensity to visit during the festive season decreased with increase in latency. 

Raymond Churn model example

Members who purchased across different formats are more likely to visit during the festive season. 

Raymond Churn model example

Members who purchased in more stores are more likely to visit 

Raymond Churn model example

Members with high vintage are more likely to visit 

Raymond Churn model example

Members who visited in the last festive seasons are high likely to visit again during this festive season 

Raymond Churn model example

 

Members with high visits are more likely to visit during the festive sale 

Raymond Churn model example

See Also

Business Impact

By now, we’ve two broad classifications of customers as festive shoppers who are more likely to come during the festive season and the non-festive shoppers. We’ve to drive additional revenue by sending personalized campaigns to these customers. Few of the strategic marketing activities are listed below.

Raymond Churn model example

Validation

We have applied the model on actual shoppers who visited during the festive season in October. Out of the 70K repeat customers who have visited during this year festive season, our model is able to successfully predict 49K customers. It is showing a true positive rate of 70%.

Assumption

Our model predicts only the repeat customers who will shop during the festive season or not. For the new customers, we can use past trends and growth rate to arrive at their number.

Conclusion

This model can be replicated to other peak seasons where seasonality factor is high, such as, End of the Season Sales (EOSS) which happens in January/July or during the wedding seasons.


Authors

Nishit Mittal

Nishit Mittal is working with the Consulting-econometrics and analytics research team at Hansa Cequity and is based out of Mumbai office. Nishit is a data science enthusiast   and   is   trained   in   data   science   and   business   analytics   from   IIM Bangalore and LSE. He has worked as an economist with RBS and UBS and as a consultant   with   NCAER   before   joining   Hansa   Cequity.   He   can   be   reached   at nishit.mittal@cequitysolutions.com

Varun Gampa 

Varun Gampa is an alumnus from IIM Trichy and currently working with the analytics team at Hansa Cequity. Prior to joining Hansa Cequity, he worked with Infosys. He leverages his skills to explore more meaningful avenue for analytics with the clients Varun.Gampa@CequitySolutions.com



Register for our upcoming events:


Enjoyed this story? Join our Telegram group. And be part of an engaging community.


Our annual ranking of Artificial Intelligence Programs in India for 2019 is out. Check here.

Provide your comments below

comments

What's Your Reaction?
Excited
0
Happy
0
In Love
0
Not Sure
0
Silly
0
View Comments (0)

Leave a Reply

Scroll To Top