bi.004.03.presentation.coupon redemption - power to predict valued shoppers
TRANSCRIPT
Coupon RedemptionPower to predict valued shoppers
Group 3
Akshay KherAdwait Ghule
Priya WaghMidhu C BabyKuldeep Ahir
Agenda• Introduction•Project Motivation•The Challenge•Data Description•Pre - processing Techniques•BI Techniques• Interpretations•Model Comparison•Business Value
Introduction
•Challenge for manufacturers and retailers to acquire and retain customers
•Spend huge amounts in promotions to draw more customers
•Understanding buying behavior of customers of utmost priority
•Coupon promotions – major vehicle used for marketing
Project Motivation
•Marketer’s dilemma: “Who should be offered discount in order to maximize sales and how much?”
•Challenge: To identify loyal buyers amongst the shoppers who redeem coupons
•Use of pre-offer shopping history of shoppers to predict which of them are most likely to repeat purchase after being exposed to a coupon offer
Data•First hand data obtained from the client •350 million rows of transactional data for over
300,000 shoppers•Only one offer per customer
The Challenge
•No ready predictor variables available•To mine this huge data dump and identify variables
that could be used as predictors in the model•Create variables using existing research and our
intuition•Existing techniques : Descriptive – RFM•Pareto Principle•Offer Attractiveness
Variable Creation•No. of transactions by the customer for the product
on offer•Amount spent by the customer for the product on
offer•No. of items bought by the customer for the product
on offer•No. of days since last purchase for the product on
offer •Similar variables for category, brand and company•Offer value for the product on offer
Variable Creation
•No. of pack-sizes for the product on offer•No. of competitors for the product on offer•No. of times the product was bought in the last 30 days•No. of transactions in the store in which the offer is
made•Total units sold of the brand on offer•Average price of the product on offer•Market share of the product vis-à-vis the category •No. of return transactions for the product on offer
Variable Creation
•No. of transactions on discount – discount proneness•Day of the week•Store size•Tenure of the product on offer
SAS Model Diagram
Data Pre - processing
•Used StatExplore node to understand the data statistics
•Used Data Partition node to separate out data in training and validation in order to build our model
•Used Impute node to compute the missing values
Type Percentage of data No of ObservationsTotal data 100 % 160057Train 75 % 120042Validate 25 % 40015
Data Pre - processing•Reduced skewness of input variables by using
Transform Variables node and log as transforming method
Correlation check
Variable Correlated With Correlation Coefficient
No. of transactions for the product on
offer
No. of transactions for the company on
offer
0.93
Amount spent on the category on
offer
Amount spent on the company on
offer
0.87
No. of items bought for the product on
offer
No. of items bought for the company on
offer
0.83
Amount spent in the store on offer
No. of items bought in the store on offer
0.80
Variable Selection•Using default setting for Variable Selection, 37
variables are selected•Target variable ‘repeater’ is a binary variable with
value either t or f (true or false)
BI Techniques
Logistic Regression
Decision Trees
Neural Networks
Logistic Regression Model
Statistical Label Train ValidationMisclassification Rate 0.254353 0.252805Average Square Error 0.177404 0.177001
Decision TreeStatistical Label Train ValidationMisclassification Rate 0.250679 0.249182Average Square Error 0.176605 0.176425
Decision Tree
Neural NetworkStatistical Label Train Validation
Misclassification Rate 0.250421 0.248407
Average Square Error 0.174872 0.174185
Model Comparison•ROC Chart indicates that Neural Network Curve shows
better accuracy
Model Comparison
FindingsVariable Odds
RatioImpact
Total Amount spent by the customer on the brand on offer
2.789 178.93%
No. of items bought by the customer for the category on offer
2.453 145.30%
Average price of the product on offer 1.512 51.20%
No. of days since last transaction for the brand on offer
0.731 -26.86%
Total amount spent by the customer in the store on offer
1.270 27.03%
No. of sizes the product on offer is available in 1.334 33.41%
Total Amount spent for the product on offer 1.000 0.01%
No. of competitors for the product on offer 1.004 0.37%
Implications• Past purchases made in the brand and category for which the
coupon was offered are the strongest predictors of repeat purchase
• More the price of product, more the probability of customer to repeat the purchase
• As the number of days since the brand or category for which the coupon was offered increases, the probability or repeat purchase decreases
• Besides, the store in which the coupon is offered, Product size assortment and Success ratio of a product also plays a small but important role in predicting repeat purchase.
Business Implications
• Loyalty plays a huge part in estimating repeat purchase
• Product pricing and promotion location also plays a significant role
• Better understanding of the customer behavior
• Marketing strategy planning
• Managing the budget effectively