Loyalty Mining Datamining Loyalty Card Data with RapidMiner Gábor NAGY [email protected].
-
Upload
darren-ramsey -
Category
Documents
-
view
223 -
download
2
Transcript of Loyalty Mining Datamining Loyalty Card Data with RapidMiner Gábor NAGY [email protected].
Loyalty MiningDatamining Loyalty Card Data with RapidMinerGábor [email protected]
Loyalty Program
• Offered by retailers and manufacuterers to stimulate continued patronage of consumers, often used to give additional privileges to best/loyal customers
• Systematic collection of customer data in return for rewards and benefits
Benefits overview• Understand customers• To upsell or cross-sell products• Give discounts to best customers ONLY• Adjust real price for certain customers• Find valuable customers and link them to engaged employees• Increase customer frequency• Retain existing customers• Better targeted product mix and category management
• Use data to negotiate with partners• Sell data/insight to FMCG retailers• Improve inventory management• Informed ranging and geographic location decisions
Understand customers• Segmentation• Primary Retail Segmentation• Recency, Frequency, Value (RFV):
• High value Frequent Shoppers• Low value Frequent Lapsed
• Lifestage: Young Adult, Young Family, Older Family• Lifestyle: Vegetarian, Bio-freak, Fast Food, Home Cooking, Ethnic Dining
• Marketing material, message targetting• Use data collected to personalize vouschers/coupon offers• Discounts are personalized based on shopping habits
• RFV• Lifestage• Shopper type
• Advantages:• Increased response rate
Geo-marketing• Store catchment areas in overcrowded markets need to be
defended from competitors• Use data to run targeted promotions on areas of highest
competition
Customer retention• Cheaper by a magnitude to retain existing customers than to
acquire new ones• By building loyalty we retain customers and keep them from
shopping elsewhere
Driving sales• Most valueable customers are invited for store events• High quality presentation create conducive enviroment for
offering high end grocery products• Regular communications
Effects of Cards• Increased customer loyalty = lower price sensitivity• Access to important information on consumers and trends• Higher average sales (due to cross-selling or up-selling)• Greater ability in tartgeting special segments
Costs• Loyalty program has an explicit cost of learning about
customers
• Costs include but not limited• Freebies• Discounts• Data storage and technical costs
Data management• Data stored on card• Data stored on multiple databases • Data stored onsite at network units• Aggregation takes place during closed hours
• Database for storing customer transactions• Huge datasets• Tesco Clubcard UK: 12 million subscribers• cca. 20 items bought per visit• 2.16 * 1010 transactions per year
• Adds up to massive hardware and software costs
Type I.• Discount cards:• Points collection• No differentiation between consumers• No targeted communication• No information base on customers• No customer segmentations
• Participants are not identifiable• Everybody receives a discount• Differentiation between customers and communication not possible
• Easy to implement• Low cost of maintenance
• Examples• Stamp collection (Spar), point gathering (DM)
Type II.• Free product/service after certain quantity• Participants are not identifyable• Discount based on cumulative puchases• Basic consumer information• Reactive analysis• No targeted communication
• Examples• Multiplex cards• McDonalds Cafe
Type III. – Mineable• Cummulative spending makes points and discounts• Identification is implemented• Customer segmentation• Targeted communicaiton• Add-on services• Datawarehouse/database needs• More complex
• Examples:• Airtravel (Air Miles)• Supershop
Type IV. – Mineable • Customers are identified• Segmentation of customers• One to one communication could be implemented• Integration with communication, targetting, supply chain
management, product placement etc• Other sofisticated analytical tools and reports
• Examples• Tesco Clubcard• Harrah’s• Sainsbury’s
Loyalty card mining process• Capture all data relevant to customer
relationship• All customer transactions• Point redemtions• Online transactions• Calls to touch points• Demographic
• Segment data according to:• Shopping behavior• Lifestage• Lifestyle• Attitude• Shopp habit• Geographic-demographic• Basket• Interest• Etc.
Type III. and Type IV. program tipology
Customer identification
based programs
One participantMultiple
participants(co-branded)
Program owned by the company
Outsourced
Excercise 1
• If You were to be asked to build a loyalty card database, what attributes would You store?
• How would it be possible to aquire new data?• What other datasources could you use?• What kind of segmentation could You build from the
database?• Who are „best” customers?
Loyalty Marketing industry US• US $6 billion• 90% of Americans participate in at least 1 loyalty program• The average is 12
The Tesco Clubcard 1• One of the leading clubcard in the world• 12 million households in Tesco Clubcard database (UK)• Most sophisticated data management and datamining• Partners with dunnhumby• Longitudinal databases:• Trends of purchase, repurchase, store visits
• 1Ł spent = 1 point• Points are stored and built up• Double points on special offers• 150 mŁ invested in development• Result:• Exapnsion into new consumer needs: Insurance, Tesco Bank,
Telecommunications• Leading online retailer• New format (Fresh and Easy)• Data is used for new endavours
Example use of data• Tesco used loyalty card data to develop Finest private-label,
because loyalty card data showed that higher spending customers were not purchasing wine, cheese and fruit from Tesco
• After Wal-Mart purchased ASDA Tesco used the DB to identify 300 items that price sensitive shoppers typicaly bought. Tesco lowered these products prices to retain customers.
• Products are labeld with 42 attributes (DNA)• Bio pproduct• 2 in 1• Largest format• Lowest price• Etc
• Customer baskets have aggregated DNA-s• This helps segmentation
Example use of data• 15 customer segments from food preferences• Segments are targetted with differnet newsletters, coupons.• Industry average of coupon retention is between 1-2% • Tesco’s coupons retention = 15-20%
Ethics of Loyalty CardsOpen discussion• Loyalty card data invader of privacy or efficient way to boost
sales?• Ways to keep data private while still redeeming coupons?
Knowledge of loyalty programs
Sonda Ipsos (N=1500)
Total Spontanous response
Participation
Shell Smart 70% 42% 16%
Multipont 58% 27% 12%
Tesco 35% 14% 9%
Cora 26% 13% 7%
Spontaneous knowledge is higher in• Budapest• 18-29 years old• AB state consumers (refers to ESOMAR lifestyle segments)• Higher spending• Higher educated
Attitude towards loyalty programs• Not participating because (28% of total sample)• Not interested (49%)• Does not make any sense to participate (39%)• Not confident in providing data (21%)• No time to take part in loyalty program (8%)
• Passive group description• Lower educated• Lower ESOMAR class• Older• Rural enviroments
SuperShop• Owned by: Plus, OMW• 60% of cardholderes active• Partners: OBI, Kaisers, PhotoHall, PatikaPont• Registration needed• Registration data:• Address• Date of birth
• Every transaction is logged• One transaction = 1 point• 1 liter gas = 1 point• Technology• Offline and online bankcard
Shell SMART• Owned by: SHELL• General consumers• Partners: McDonalds, Citibank• Multicountry• Registration data:• Name• Address• Date of birth• Mothers name• Phone number
• All transactions are logged
Multipont• Partners: OTP, MOL, CBA• General consumers• Registration data• Name• Address• Date of birth• Mothers name• Phone number
• Bankcard
Hipermarkets• Tesco card registration data• http://tesco.hu/clubcard/regisztracio
• Cora card registration data• http://cora.hu/az-en-coram/bizalomkartya-adatmodositas/
Class dataset• Dunnhumby shopping challange • http://www.kaggle.com/c/dunnhumbychallenge• Tesco Clubcard data (undisclosed but most probably)• 100 000 Clubcard customers• USA data• For descriptives see process
• Download class files from:• http://troja1.tmit.bme.hu/CA/Dunnhumby.zip
• Software: RapidMiner• http://rapid-i.com/content/view/26/84/• Import repository• Details in readme.txt
Loyalty card data examination• Working set sampled (N=1000)• Transformed dataset• Attributes:• Customer ID• Spend• Visit date• Previous visit• Previous spend• Day name of visit date• Day name of previous visit date• Training and test data flag
Aggregation of data• Reporting • Monthly reports
• Summation• Daily reports
• Average
• What different aggregates tells us?• Sum• Average• Median
• Reason for using median spend over average spend?
Seasonal effects• Daily effects• Sum spending• Average spending
• What kind of seasonal effects can You identify in the dataset?• Aggregate by date
• What is the meaning behind the seasonalities?
Study of shopping patterns• What kind of irregularities can we find in the dataset?• Hypothesis: People tend to follow patterns in shopping• How can we „measure” or identify patterns?
1. Aggregate spending, day difference between successive visits• Sums• Averages• Standard deviation• Median• Mode
2. Let these aggregates define a customers• Segmentation: how shops regularly (eg. k week between successive
visits)• Who spends more than the median spending
Modelling• Prediction of next visit• Based on stats (average, median, mode)• Based on models (Decision tree, Random Forest, K-NN)
• Prediciton of spending• Based on stats (average, median, mode)• Based on model (W-RepTree, Linear Regression)