Predicting & Preventing Banking

Customer Churn By Unlocking Big Data.

Case Study on a Bank.

Light Years Ahead.

All Rigths Reserved Rulex, Inc. 2014

CUSTOMERS CHURN.
A key performance Indicator for Banks.

Confidence in the banking industry is on the rise, and trust in customers own
financial services providers is high. But customers are on the move, with
unprecedented access to competing banks and to new types of financial services
providers.
Banks must earn the highest levels of trust in order to retain customers, win more
business and create genuine loyalty.
Customer churn and engagement has become one of the top issues for most banks:
It costs significantly more to acquire new customers than retain existing ones.
It costs far more to re-acquire defected customers.
CHURN IS ONE OF THE BIGGEST DESTRUCTORS OF ENTERPRISE VALUE FOR BANKS
AND OTHER CONSUMER INTENSIVE COMPANIES.
All Rigths Reserved Rulex, Inc. 2014

CUSTOMERS CHURN.
The Key issue: to know customers and predict churn with Rulex.
Pool of customers

ACTIVE

CHURNING

In order to identify early signs of potential churn you

first need to start getting a holistic 360-degree view of
your customers and their interactions across multiple
channels.
RULEX is able to aggregate the customer information
across multiple channels and to focus on several key
indicators that can flag propensity to churn.
If you can easily detect these signs, YOU CAN TAKE
SPECIFIC ACTIONS TO PREVENT CHURN.
RULEX IS THE NATIVE TECHNOLOGY ABLE TO SOLVE
DATA ANALYTICS CHALLENGES POSED BY TRADITIONAL
TECHNOLOGY

CHURNED

Who, When and Why is

going to churn.
All Rigths Reserved Rulex, Inc. 2014

CUSTOMERS CHURN.
Why Rulex is LIGHT YEARS AHEAD?

With RULEX, banks can store, analyze and retrieve a massive volume and variety of
data to aggregate the totality of information about the customer into a single
platform
RULEX allows banks the economical advantage of storing data and scale it elastically
to expand with the data volume growth
RULEX allows banks tap into a real-time data and customer interactions that provide
clear insight into early warning signals to ensure timely retention offers and
preservation of enterprise value

Rulex will build a model which will list the factors resulting in churn in order of
importance in two weeks or less. Rulex will give you the business rules needed to
take action to reduce churn.
All Rigths Reserved Rulex, Inc. 2014

HISTORICAL DATA
Who did / didnt Churn

Bank Dataset:
112984 in the training set
48421 in the testing set

161405 past customers

75 attributes per each customers

Customer State? is the output variable.

It can be Actual or Former.
99961 customers did not churn:
61444 customers churned:

Integer

Nominal

Customer State = Actual

Customer Stare = Former

Continue

Date

All Rigths Reserved Rulex, Inc. 2014

RULEX OUTCOME: THE CHURN MODEL

52 rules explaining the phenomenon
AUTOMATI
-CALLY
INFERRED

RULES

COVERING

ERROR

CONDITION RELEVANCES

All Rigths Reserved Rulex, Inc. 2014

RULEX OUTCOME: THE CHURN MODEL

Details from the GUI
AUTOMATI
-CALLY
INFERRED

Rule # 41
IF (Customer Type is in a given subset) AND
IF (Account Balance SML <= 0.135)
THEN (The Customer Churns)

All Rigths Reserved Rulex, Inc. 2014

RULEX OUTCOME: THE CHURN MODEL

Exploring the Rules Interface

COVERING
Rule#41 is satisfied
by 35.5% of 43083 churning cases

CONDITION RELEVANCES
Removing Cond.1 from
rule#41 increases the error by 41.5%.
Cond.1 is extremely relevant!

ERROR
Rule#41 gets wrong (false positive) in the 4.5% of
the 69900 non-churning cases
All Rigths Reserved Rulex, Inc. 2014

AUTOMATI
-CALLY
INFERRED

ATTRIBUTE RANKING
How are churning customers characterized?

Time since last transaction

has a (positive) relevance of
about 46% for churning
customers (State=Former)
Account balance has a
(negative) relevance of about
37% for churning customers
(State=Former)

AUTOMATI
-CALLY
INFERRED

Customers who churn:

Do not have deposits
Has an old first purchase
Belong to particular categories
(Customer type)
Have a high Time since last
transaction

All Rigths Reserved Rulex, Inc. 2014

BI tools can confirm the simplest conditions

Above 10000 Account

Balance SML, almost
all customers are
Actual

Above 1 Free Saving

Deposit, almost all
customers are actual

but cannot find multi-condition rules. Rulex does,

automatically.
All Rigths Reserved Rulex, Inc. 2014

CONFUSION MATRIX
How good is the churn model?

Customers with a churning

behavior still active
HIGH ACCURACY: the Rulex model fits
about 78.5% of not churning customers,
and 84.2% of the churning ones.

UNBALANCE IMMUNITY: Rulex is

immune to intrinsic unbalances (churning is
less frequent than staying).
All Rigths Reserved Rulex, Inc. 2014

THE RULEX APPROACH

Understand. Forecast. Decide.

Forecast
who is going to churn?
why?
what are their drivers?

All Rigths Reserved Rulex, Inc. 2014

CHURN CANDIDATE LIST

Who is going to churn & who is not
Previsions about new customers (are
they churning?) are made quickly
applying the rules to the available
attributes.

WHO

WHEN

WHY

List of
customers

prevision
confidence

main
applied
rule

Current
state

This customer has already churned (and

Rulex recognized it)

This customer has not churned yet but has a

churn-like behavior.

Automatic
alarm / start
actions

All Rigths Reserved Rulex, Inc. 2014

Prevision

THE RULEX APPROACH

Understand. Forecast. Decide.

Decide
You are the experts in your field.
With the knowledge provided by Rulex,
now you can make effective decisions to
solve the problem of churn.

All Rigths Reserved Rulex, Inc. 2014

THE RULEX APPROACH

Understand. Forecast. Decide.
Creation of the model from the past

Application of the model for the future

Bank Actual Data

Customer info, contract,
transactions.

Bank Historical Data

Customer info, contract,
transactions.
Churn=yes/no.

Forecast
EXPLICIT
MODEL,
DESCRIBED
BY RULES
(IF-THEN conditions)

Churn Model
List of rules and
drivers
describing
who churns

Churn Reduction
Using the rules and attribute relevancies,
the bank defined marketing and sale actions focused to reduce
the phenomenon at the origin.

Application of the
Churn Model
to all customers,
to test if they will
churn or not

Churn
Candidate
List

Understand
AUTOMATIC
ALARM

Decide
Churn Prevention
The bank created a portfolio of actions
to be automatically activated when an alarm is received.

All Rigths Reserved Rulex, Inc. 2014

CONCLUSIONS
Rulex makes Churn Analytics quick, automatic, precise and clear:

Data pre-processing: 1 minute

Automatic model extraction: 20 seconds
Clear view of:
Conditions of churning (rules)
Relevance, for each attribute
Critical thresholds, for each attribute
High accuracy
Confidence of prevision for each customer

All Rigths Reserved Rulex, Inc. 2014

THANK YOU

Light Years Ahead.

All Rights Reserved Rulex, Inc. 2014

USA - 75 Federal Street, Suite 920 - 02110 Boston, MA, 02110 T: +1 617 263 0080 F: +1 617 263 0450
EUROPE - Via De Marini 16, 16th Floor - 16149 Genova (Italy) T: +39 010 6475218 F: +39 010 6475200

Contacts

For more case studies, white papers and further information please go to
www.rulex-inc.com
or follow us on

USA -

or Contact me:
Linda Treiman
Linda.treiman@rulex-inc.com
Linda.Treiman1

USA - 75 Federal Street, Suite 920 - 02110 Boston, MA, 02110 T: +1 617 263 0080 F: +1 617 263 0450
EUROPE - Via De Marini 16, 16th Floor - 16149 Genova (Italy) T: +39 010 6475218 F: +39 010 6475200