data science for crm in banks
TRANSCRIPT
Data Science for CRM in BanksA non-exhaustive overview
Peter Koglmann
Vienna Data Science GroupKnowledgefeed Vol 13, 2016-09-16
Agenda
What‘s the aim of CRM* ?
What‘s special to CRM in banks ?
How can data science help in solving CRM problems in banks ?
* Customer relationship management (CRM) is an approach to managing a company's interaction with current and potential future customers. (wikipedia.org)
The Aimof CRM
• Create and keep valuable customers
• By offering to the customer– the right product
– at the right time
– in the right way
• Reduce costs and increase sales
Distinctivefeatures ofCRM in banks
• High expectation on privacyrequire careful handling andusage of sensitive data
• Banks were among the firstsectors heavily using IT
• Plethora of data
• Rather long terms contracts• Service oriented• Life cycle driven
How can data science help in solving CRM problems in banks ?
• Create and keep valuable customers
• By offering to the customer– the right product
– at the right time
– in the right way
• Reduce costs and increase sales
• Use data science to– better understand the customers and predict
their future behaviour
– push prescriptive actions to take the mostrelevant and timely step
Predictive Modelling in CRM
Explainingvariables Log
Regr
RF
SVM
low
high
Classification
Socio-demographic
Behavioural (accounts, buying, contacts)
External (pricing, competitors, reputation)
propensity
Customer account record
Change in propensitycaused byintervention
Uplift model
Log Regr
RF
SVM
low
high
conversion
Keep valuable customers
• Predict churn– Classification models (Log. Regr., RF, SVM, …)
– Uplift models to assess the effect of intervention
– Survival models (how long will customer stay?)
– Recommender systems (which actions?)
• How many to target? – Cost-benefit & ROI
analysis based on confusion matrix
predicted
yes no
actuals yes TP => gain FN
no FP => -costs TN
• Who is how valuable?– Customer Lifetime Value: present value of
future revenues (e.g. Semi Markov models)
The right productat the right timein the right way
• Predict propensity to order a service/product– Classification models (Log. Regr., RF, SVM, …)
– Age, behaviour, etc. of customer as explainingvariables lead to information of „right time“
• Predict uplift & conversion rates for eachchannel
– Uplift models, A/B tests
• Next best offer– Recommender systems, Multinom. Log. Regr., …
Reduce costs andincrease sales
• Target only clients with high propensity andconversion rate. How many?
– Cost-benefit analysis (confusion matrix)
• Profile top clients and identify currentunderperformers within that group
– Cluster analysis
– Customer Lifetime Value
• Cross-sell and up-sell– Next best offer techniques
• Improve models on a continuous basis– Validation, benchmarking, trial and error