This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Banks face the challenge of leveraging large volumes of disparate data for increasing customer engagement (consumers & Bank affiliated merchants)
Merchants find it challenging to provide new and existing customers with target offers at the right place & time and increase sales
Customers need to manage personal finances, monitor spend and save money on their purchases through relevant offers
3
SpendWise. Be wise
Cognizant believes that Banks can address these questions by leveraging advanced Cloud Based Data Analytics and the Digital Banking–driven ecosystem ( including Social Media Feeds )
This submission is an attempt to solve these business problems by leveraging the power of Cognizant’s Pioneering SMAC (Social - Mobile - Analytics - Cloud) Framework
Introduction“SpendWise Genie”, is a multipurpose Mobile App that gives consumers rich insights into benchmarking their spend behavior with people of similar demographic profile( by location and merchant category). The App also uses the power of real time predictive algorithms for relevant offer presentment and empowered decision making
BBVA App For Customers
Value Proposition
SpendWiseGenie
Spend Comparison with
Peers across segments Responsible
Spending (for Consumers*)
+Increased Sales
(for Banks & Merchants)Best Offers
Prediction supported with real-time data ( Maps/ratings)
“People like me” AppWith “People Like Me” learn how consumers spend in your same category
People Like Me
Rich interactive charts represent the spend pattern across time cohorts
The consumer also picks spend category and subcategoryThe user provides his age, gender & location information and also the time period for which he wants to view spend patterns
Objective of the model is to predict user spend category and spend amount based on:- The historic spend behavior of user segment (age group & gender);- Location (zip code) based spend patterns as well as seasonality trends – based on day-of-the-week and time-of-the-
dayThe base data* used for model training is based on the BBVA transaction data for:- Top 10 spending Zip Codes using BBVA APIs: Cards Cube and Consumption Pattern- Data filtered for Top 5 Merchant Categories with the highest spend in these 10 zips
Input data* for predictive model training is prepared by combining data from above 2 BBVA APIs to calculate the most probable spend on merchant category for a user segment at a weekday, at a particular hour: - By calculating merchant spend probability at an hour of a weekday using weighted contribution of spend at that hour
and spend by that user segment- For the sake of simplicity, spend behavior on a particular weekday (e.g a Tuesday) across months is assumed to be
similar
Using Advanced Predictive Models for a given Zip Code, User Segment , Day, Hour of Day:- Model 1: Prediction of Merchant Category with the Highest Probability of next User spend; Model Type: Classification Model; Model Accuracy: 89%- Model 2: Prediction of Spend Range for a user for selected/predicted Merchant Category; Model Type: Regression
Model; Refer MSE (Mean Square Error Values in Appendix)
Model Output i.e. the predicted category and the predicted spend is used to pull offers from offers database filtered for :- Selected Zip Code, Merchant Category, Predicted Spend (Range)- Pull rating/reviews from Google Places for displayed filtered offers
Detailed Evaluation of Models explained in next 2 slides
12
CRISP – DM Methodology Followed
* - All local copies of data made to test model accuracy have been purged
Prediction of the Merchant Category with the Highest Probability of next User spend in a given:- Zip Code- User Segment (Age Group, Gender)- Week Day- Hour of Day
1
Spend Prediction
Category Prediction
Prediction of the Spend Range for a user for selected/predicted Merchant Category in a given:- Zip Code- User Segment (Age Group, Gender)- Week Day- Hour of Day
2
Model
Model
Model Limitations:Since the current model uses the costumer transaction data as input data from BBVA APIs , which is available only at a segment level (age-group, segment) and not at a customer ID level, the models output is only valid for segment/cohort level predictions, assuming all the customers within that cohort behave in a similar spending manner.
Prediction of Merchant Category with the Highest Probability of subsequent user spend in a given:- Zip Code - User Segment (Age Group, Gender)- Week Day - Hour of the day
Assumptions:- Training* and prediction limited to Top 5 Spend Merchant Categories (except mx_others):
{(mx_basrsandrestaraunts, mx_food, mx_fashion, mx_auto, mx_hyper (mall)}- High Accuracy prediction for Top 10 spending zips for current model
Model Type: Classification Model; Model Accuracy: 89%Test Cases:
Prediction of Spend Range for a user for selected/predicted Merchant Category for:- Zip Code - User Segment (Age Group, Gender)- Week Day - Hour of Day
2Model
Assumptions:- Training* and prediction limited to Top 5 Spend Merchant Categories (except mx_others):
{(mx_basrsandrestaraunts, mx_food, mx_fashion, mx_auto, mx_hyper (mall)}- High Accuracy prediction for Top 10 spending zips for current model
Model Count: Separate Model for each of the Top 5 Merchant Categories
Model Type: Regression Model; Test Cases: E.g. For merchant Category – “mx_auto”
Zip Code
Age Group Gender Week Day Hour Actual Spend Predicted Spend
"11590" "19-25" "Female" "Fri" 7 147.33 483.6822
"11320" "36-45" "Male" "Mon" 17 1114.3 1611.875
"11520" "46-55" "Male" "Sat" 12 2723.5 1415.509
"64000" "56-65" "Male" "Wed" 18 168.15 169.7097
"11590" "46-55" "Male" "Wed" 9 1000 1294.2
15
How it works….Evaluation of the Data Mining Model for Spend Range Prediction
* - All local copies of data made to test model accuracy have been purged