Data Mining for CRM: Some Relevant Issues
Introduction
• It provides relevant intelligence to the CRM initiative
• The aggregate or macro level, finds general customer behavior,
trends and preferences from a large data base eg market
basket purchases from a retail store
• The individual or micro level, tracks a particular customer to
act on his/her preferences
Aggregate or Macro level data
• Eg If customers of a retail store are segmented by profitability
criteria, we obtain clusters who are profitable
• Further mining gives us characteristics of each of these
segments
• The natural segments might change if we change the focus of
our exercise
• Segmentation by demographics instead of profitability would
give completely new clusters
Aggregate or Macro level data
• This is useful in situations where:
• Targeting new set of customers- if a retail chain opens a new
store, it can use data from the similar current store to predict
the behavior of the new prospects
• Individual tastes and preferences cannot be met- layout of a
retail store
• New customer for a product not tried by him/her- basis
existing similar profile
Individual or Micro level
• For this, we need to track the customer and mine at the
individual or the micro level
• Specific info about a customer eg the preferred colour of his
shirt
• Separate data base for profitable customers will save time and
money
Individual or Micro level
• It is useful in situations where:
- The firm wants to customize offerings- bundling
- Push new products based on past purchases- discount on
matching tie with suit recently bought
- Take advantage of personal events- cement the precious
relationship
- Sudden switching of brands- analyze
The Framework: 3 components
• Operational or process management technologies- involves
customer touch points
• Analytical or performance management technologies- analyses
data using technologies to feed the result back to operational
or collaborative technologies
• Collaborative or business collaboration management
technologies- facilitate interaction with customers
• There is an overlap as well as feedback between efforts
Interaction between the CRM technologies
Operational Analytical Collaborative
CRM CRM CRM
Customer
Data mining methods relevant to CRM
• Can be predictive or descriptive
• Predictive analyses assign value to some variables related to
other related ones
• Descriptive focuses on essential features or characteristics of
data
• CRM uses both to understand the customer better
1. Regression
• It needs sufficient data to be reliable and valid
• Linking time spent in a store and the value of purchases is an
example- the RFM model
• This can be modeled on Regression and can be used to predict
future purchases
2. Link Analysis
• Seeks to establish relationship between items or variables in a
data base record to expose patterns and trends
• Can also trace connections between items of records over time
• Eg Market basket analysis- seeks relationship product items
and buyer preferences
• Can help store layout design, coupons etc
3. Segmentation
• Aims to identify a finite set of clusters or categories to describe
data
• Eg a retailer can segment his customers based on buying habits
• He might find two distinct clusters a) spend less time, buy less
expensive items b) more time, buy more expensive items
• Or on basis of profitablity
4. Deviation detection
• It focuses on discovering the most significant changes in data
from previously measured and/or expected parameters
• Eg a retailer finds that sales from a particular section of a store
has been less than expected
• Deviation detection may point to the non-stocking of a popular
brand- change in fundamentals
• Or it may sometimes be a result of a random occurrence
sometimes or change
Some Tools & Techniques
• Case Based Reasoning: CBR method
• Tries to simulate how a human being thinks
• When an example is presented to a CBR
solution, it tries to match the current example
with other examples that it has in its
repository
• It then retrieves the case that is most similar
to the current case- extrapolation
• It depends largely on the indexing methods
used to store cases
Visualization technique
• Allows the user to view data from various angles using graphic
display techniques like charts, diagrams, multi dimensional
data etc
• It is used in conjunction with other data mining techniques
Nearest Neighbor technique
• It uses a set of examples to approximate a classification model
• A similarity measure is used to find the closest example in
terms of certain parameters
• Then it assigns the new case to the class that has the maximum
representation among its neighbors
• Eg a new retail store that has been set up is matched with an
existing store with similar parameters
Conclusion
• Choice of a data mining tool depends not only on the capabilities of
the tool but also on the question we seek to answer
• Multiple tools might answer the same question but to different
degrees of satisfaction and completeness
• Data is dynamic- new data coming in real time into the data
warehouse
• Tight integration between front end CRM solution and the back end
data mining solution helps to ensure consistency & timely information
sharing throughout the firm- indispensable to any CRM initiative