Transcript The CRM Textbook: customer relationship training

The CRM Textbook: customer
relationship management training
Terry James
© 2006
Chapter 12: Analytical
1
© 2003 Terry James. All rights reserved
2
Operational vs. Analytical

Operational




transactions, POS point of sale
answer in seconds,
zero failures
Analytical


© 2006 Terry James
learning, analysis, patterns, history
answer in hours or days
3
Data Warehouse vs. Data Mart


More sophisticated than relational database
Data warehouse





Enterprise, huge, standards
Level of granularity
Cube – 3D
Fact tables
Data Mart

© 2006 Terry James
Smaller, departmental, more unique needs
4
Place
Time
Product
Cube 4
© 2006 Terry James
Cube 2
Fact table
5
ETL

Extract


Translate


Data from operational files all over, and
any other useful data source
Standardize the data, clean it, rationalize
Load

© 2006 Terry James
Load up the data warehouse
6
Quality



Major issue
Plan spend 30% of your time for quality
Data dictionary


Most common errors


Missing data, invalid data, out-of-date
Inconsistencies


What is the definition, the data steward, the meaning,
valid values, etc.
Different meanings for the same code, different codes
for the same meaning, multiple data for the same data
element
Meta data

© 2006 Terry James
Data about data
7
OLAP vs. data mining

OLAP



OnLine analytic programming
You start with a question, run reports,
check data, publish results
Data mining


© 2006 Terry James
Start with no question
Wander across the data to uncover
patterns of fraud, buying, selling, etc
8
Data Mining Techniques

Correlation


Regression



Emulates the brain (wetware)
Fuzzy logic
Clustering



Predict the future
Example: Buying = -2.4(price) + 4.1 etc.
Neural network


When prices go down, buying goes up
What things go together in a bundle
If you are like other people who did x, they also did y
Genetic algorithm


© 2006 Terry James
Emulates nature ,evolution, and mutations
If random change to formula provides better predictions, keep it, otherwise
retest and then loop to make new change
Data mining process
9
1.Begin with an important
Learn
New
data
© 2006 Terry James
Take
action
company goal
2. Collect data needed
3. Data quality, ETL
4. Pick technique (genetic,
neural network, …)
5. Build a model
6. Test and validate model
7. Implement model
8. Report results
9. Integrate new learning
10. Go back to step 1
10
Traps

It is so cool, sexy, interesting,…


Yes, but does it put cash on the table?
Prove the obvious

© 2006 Terry James
Don’t burn CPU cycles just to prove
purchase patterns match marketing
campaigns. Go after valuable items, not
motherhood and apple pie.
11
Validating



Does the model work?
Do you have a response equation to the
campaign?
How accurate was the model?




False positive
False negative
Beware Bayes Theory
What about the control group?
© 2006 Terry James
12
Learning is a forever loop

Each worthwhile analysis should be
focused on action





Check ahead if manager is ready for action
and on what topics
Take what you learn and take action
Action will generate data
Take data and learn
Analysis and loop back to step 1
© 2006 Terry James