Churn prediction data modeling

Churn prediction data modeling

• 1.
  How to datamodel Churn Real life examples
• 2.
  Quick quizz •  Howmany of you are familiar with Churn issue? •  with Machine Learning? Logistic Regression, Random Forest, Gradient Boosting trees? (Not the subject here) •  With SQL? (we may see some code later) •  What database tech do you use? What about EMC Greenplum or Vertica?
• 3.
  Who I am • Senior Data Scientist at Dataiku (worked on churn prediction, fraud detection, bot detection, recommender systems, graph analytics, smart cities, … ) •  Occasional Kaggle competitor •  Mostly code with python and SQL •  Twitter @prrgutierrez
• 4.
  Churn definition •  Wikipedia: “Churnrate (sometimes called attrition rate), in its broadest sense, is a measure of the number of individuals or items moving out of a collective group over a specific period of time” = Customer leaving
• 5.
  Two types ofChurn •  Subscription models: •  Telco •  E-gamming (Wow) •  Ex : Coyote -> 1 year subscription -> you know when someone leave •  Non subscription models: •  E-Business (Amazon, Price Minister, Vente Privée) •  E-gamming (Candy Crush, free MMORPG) -> you approximate someone leaving Candy Crush: days / weeks MMORPG: 2 months (holidays) Price Minister: months
• 6.
  Two types ofChurn •  Blurred Separation: •  Ex: T-mobile: 1 month subscription -> paying each call •  Ex: Wow: 1 month to 6 month subscription •  Banking? •  Focus : no subscription: •  Can be seen as a generalization where you have to approximate the target •  Bonus : Seller churn •  Market places •  Clients that participate product life •  Forums (Reddit) •  E-gamming (Korean competitions, guilds etc.)
• 7.
  Dealing with churn • Motivations : •  Saturated market -> cost get new client >>> cost keep client •  Ex : http://www.bain.com/publications/articles/breaking-the-back-of-customer-churn.aspx •  Wireline company : 2% to 2.5 % churn rate per month. •  If 5 M customers -> 1.32 M churn per year •  When reducing from 2.5% to 2% lowest estimation : 240 M $ in 18 month
• 8.
  Dealing with churn • Predict churn : •  One model for performance <- our focus, short term, more ML •  One model for understanding <- long term, more Analytics •  Act on it (short term) : •  Special offer (telco call, free in game money, discount coupon … ) •  Does it work? Feedback loop needed! •  Model probabilities of leaving because of offer. A/B tests. Multi arms Bandit? •  Significant LTV for activation? •  Act on it (long term) : •  Is there a problem in my purchasing funnel? •  Is the game too hard at some point?
• 9.
  Dealing with churn • Candy Crush Rumor : •  Change the distribution of probabilities of candies / bombs •  Change the difficulty of the game •  Loosing a lot makes the game easier
• 10.
  Modelling Churn •  Machinelearning model (classification) -> target: •  Known in subscription •  Unknown in general •  Step 1 : Maintain customer status •  Do you care only about your best? •  Anyway churn action won’t be the same •  Has a client churned? -> target = churner = don’t buy / visit since time X -> best = buy / visit more than y since time Y •  Can be refined (“new customer”, several class of best or inactive, reactivated…) •  Storage : maintain only the difference!
• 11.
  Modelling Churn •  Machinelearning model -> features: •  Explicative factors to use as input for the model •  Step 2 : Maintain customer features •  Social (woman, age, etc.) •  Behavioral! •  Utilization / buying rate •  Trend in utilization / buying rate •  Ad hoc features : •  WoW / Social game churn: take into account friend network churn •  Telco: call to call centers •  Beware of time dependence!
• 12.
  Data Model
• 13.
  Computation Dependency diagram
• 14.
  Ex : Trainand predict scheme Time   T  :  present  ,me  T  –  4  month   Data  is  used  for  target   crea,on  :  ac,vity  during   the  last  4  months   Data  is  used  for  feature   genera,on.   Use  model  to  predict   future  churn   Train  model  using  features  and  target  
• 15.
  Ex : TrainEvaluation and Predict Scheme Time   T  :  present  ,me  T  –  4  month   Data  is  used  for  target   crea,on  :  ac,vity  during   the  last  4  months   Data  is  used  for   feature  genera,on   Valida&on  set   Use  model  to   predict  future   churn   Training   Evaluate  on  the  target   of  the  valida,on  set   T  –  8  month   Data  is  used  for  features   genera,on.   Data  is  used  for  target   crea,on  :  ac,vity  during   the  last  4  months  
• 16.
  Thank you foryour attention !