1

The Customer Database: Analysis and

Applications
DIRECT MARKETING CHAPTER 3
Ass.Prof.Dr. Tuğçe Ozansoy Çadırcı
The Customer Database: Analysis and Applications
2
What to do with the data stored in a database???

‣Data is meaningless unless it is converted into

something useful.
‣Actionable Intelligence requires not only analysis and
interpretation, but more important integration into
the context of business decision making.
‣Data analysis is a critical part in direct marketing.
The Customer Database: Analysis and Applications
3
What is Data Analysis

‣Analytics is the term used to cover any type of

analysis of consumer data, including retail,
commercial, and purchased demographic data,
using a variety of statistical tools.
The Customer Database: Analysis and Applications
4
Data Mining

‣Data mining is an automated analytical process,

processing large databases searching for global
patterns of behavior, customer attributes, contact
history, or campaign response analysis across
millions of rows of customer behavioral data.
‣ An expensive item.
‣ Can be used for segmentation.
‣ Lacks links to emotional measures of customer interaction (e.g.
attitudes, preferences, satisfaction levels).
‣ Not the best approach for assessing creative ideas.
The Customer Database: Analysis and Applications
5
Statistical Analysis

‣Statistical analysis are most successfully used on

samples of the large databases by incorporating
sampling methods and hypothesis testing.
‣Common tools are:
‣ Statistical Analysis Systems (SAS)
‣ Statistical Package for Social Sciences (SPSS)
‣Provides opportunities like data standardization, consolidation of
data from different sources.
‣Unlike data mining statistical analysis is used to test hypothesis,
and understand and predict quantifiable patterns of customer and
prospect behavior.
The Customer Database: Analysis and Applications
6
How a Marketer Uses a Database

Profile Customers
‣Understanding customer segments that the marketers are
serving.
‣Such analysis helps marketers to understand the needs of the
typical customer
Retain the Best Customers
‣Pareto’s Rule (80/20 rule)
‣With the help of data analysis companies can define their most
valuable customers and make decisions on who to spend
more.
The Customer Database: Analysis and Applications
7
How a Marketer Uses a Database - 2

Capitalize on Cross-Selling and Continuity

Selling Opportunities
‣Cross selling refers to selling your current customers products
and services related to the products and services they
purchased from your organization.
‣By the analysis of purchase history companies can create new
selling opportunities.
Develop a Customer Communication Program
‣Databases create feedback from customers.
‣Such feedback can be used as an input to better serve
customers and maximize profitability.
The Customer Database: Analysis and Applications
8
How a Marketer Uses a Database - 3

Generate New Customers

‣Companies can seek new customer with similar needs.
‣They define prospects and create offers for customer
acquisition.
‣Direct marketing is more effective in consumer acquisition
than mass media.
The Customer Database: Analysis and Applications
9
Data Analysis Issues

‣Retention Analysis
‣Attrition or Churn Analysis
‣Profitability of Prospects
‣Response Rates
The Customer Database: Analysis and Applications
10
The Need for Information

‣Several factors have influenced the way marketing

data is used:
‣ The power of computers increased massively.
‣ The cost of computers has drastically reduced.
‣ Computers have become mobile.
‣ Everyone is becoming more tech-savvy.
The Customer Database: Analysis and Applications
11
How Data Analysis Help Marketers

‣Quantitative decisions
‣ Budgets, finance, RIO, ratios, etc.
‣Qualitative decisions
‣ Why are people loyal, what are your brand strengths, etc.
‣Segmentation, targeting and positioning
The Customer Database: Analysis and Applications
12
The Value of The Marketing Database

‣A marketing database adds value to analysis in two

main areas
‣ MEASUREMENT
‣ SEGMENTATION AND TARGETING
‣There are two reasons why a customer database is a
segmentation professional's dream:
‣ The VOLUME and TYPE of the data
‣ Instead a few hundred of sample units databases provides
hundreds of thousand or even millions sample units.
‣ The database both contains data sets of profile and behavioral
data.
 The Customer Database: Analysis and Applications
13
Uses of the Database

Analysis Descrip on

Internal Informa on Understanding the company’s strengths and weaknesses

Marke ng Research Analyzing consumer markets and buyer behavior

Scanning the wider environment for opportuni es and

Environmental Monitoring
threats

Compe tor Analysis Assessing compe tors in rela on to business

Predic ng market demand; measuring company

Measuring and Forecas ng Markets
performance against objec ves set

Spli ng markets into discrete groups to be treated

Market Segmenta on
di erently

Market Targe ng Selec ng the best markets for the company

ff
tti
ti
ti
ti
ti
ti
ti
ti
ti
ti
ti
ti
ti
ti
 14

THE CUSTOMER DATABASE: ANALYSIS AND

APPLICATIONS

Segmentation
Applications
in Direct
Marketing
PART 1
The Customer Database: Analysis and Applications
15
Overview of Segmentation Approaches

Data$Types: ' 'Customer!Profile ! ! ! ! ! !Customer!Behavior!

Segmenta/on$Op/ons$

Type'of' Number'of'
Geodemographic'' Behavioral' Psychographic'
Purchases' Purchases'

Product'
A@ributes'

Figure'3.1!!Overview'of'segmenta5on'op5ons'
The Customer Database: Analysis and Applications
16
Database Applications
New$$
markets$

Profile$

Marke&ng)Database)
High$£$

Segment$by$
Segment$by$
propensity$to$
value$to$
respond$per$
company$
campaign$

Low$£$
Product$ A$ B$ C$ D$ E$

Segment$by$Product$Category$
Figure$3.2$!Company$Database$ApplicaGons$
The Customer Database: Analysis and Applications
17
The Value of The Marketing Database

‣Behavioral Data
‣ Using behavioral data allows marketers to segment customers
according to their value to the company.
‣ Direct marketers can also use purchase data to segment
customers according to their product needs.
‣Profile Data - Customer Characteristics Data
‣ It is often used to build up “pictures” of customers who are
more likely to respond to a particular campaign.
‣ Profile data is also used to target new customers more
accurately; thereby decreasing the costs of acquisition.
The Customer Database: Analysis and Applications
18
Using Purchase Data to Set Budgets

‣Companies analyze their customers’ purchase data in

order to calculate how much each customer is likely
to be ultimately worth to the company.
‣ This forecast is known as the “CUSTOMER LIFETIME
VALUE”.
‣By predicting how much gross profit we make from
customers, we can construct our budgets on a per-
customer basis. In plain words we can calculate how
much we can afford to spend for each customer.
‣ This calculation is known as the “ALLOWABLE MARKETING
SPEND PER CUSTOMER”.
The Customer Database: Analysis and Applications
19
Segmenting According to Customer Needs

‣This is a common practice.

‣In general marketing we use different research
techniques to obtain data concerning customer
needs.
‣By studying behavioral and psychographic data held
on each customer, marketers can understand the
different benefits that the same product may
provide to customers.
The Customer Database: Analysis and Applications
20
Maximizing Response on a Per Campaign Basis

‣Direct marketers use modeling techniques to identify

the most “responsive” customers.
‣This allows us to allocate budget on a per campaign
basis such that return on investment (ROI) is
maximized.
The Customer Database: Analysis and Applications
21
Targeting New Prospects Based on Existing Customer Descriptions: PROFILING

“Your new customers will have a similar profile

to your existing customers”
‣Similar people like similar products, or will be
attracted to similar brand images.
‣By making the effort to profile existing customers
accurately, marketers can more precisely target new
customers.
‣With the help of modeling techniques, it is possible to
produce a highly accurate quantitative profile of
customers.
22
23
 24

THE CUSTOMER DATABASE: ANALYSIS AND

APPLICATIONS

Analytical
Techniques

PART 2
The Customer Database: Analysis and Applications
25
Analytical Techniques
Applica'on* Technique*
Use&life0me&value&(LTV)&analysis&to&set&
Se#ng&budgets&
allowable&marke0ng&spend&
Use&frequency,&regency,&amount,&category&
Understanding&customers’&value&to&you&
(FRAC)&analysis&and&LTV&analysis&
Database&overlays&with&external&data;&
Understanding&exis0ng&customers&as& profiling/modeling&techniques;&focus&on&
individuals:&what&they&want&from&you& product&categories&link&with&customer&
profiles;&combine&with&market&research&
Targe0ng&your&spend&in&order&to&maximize&
Use&modeling&techniques,&FRAC&analysis&
ROI&per&campaign&
Profiling&exis0ng&customers&in&order&to&
Profiling/modeling&techniques&
target&new&ones&

Table&3.2.&Summary&of&Applica5ons&and&Techniques&in&Direct&Marke5ng&Analysis&
The Customer Database: Analysis and Applications
26
Introduction to Analytical Techniques

‣The amount of data a database holds, and the huge

computing power has heralded a drive towards
heavy use of quantitative techniques.
Predicting (Forecasting) Techniques: Who is likely to buy our
products?
Descriptive Techniques: What customers look like?
The Customer Database: Analysis and Applications
27
Predictive Techniques

‣There are two predictive techniques:

‣Time Series Methods
‣ Make predictions about future levels of customer response.
‣ Projecting forwards from trend, cycle and seasonal patterns contained in
past values of our data.
‣ Time series techniques are the quicker of the two approaches and more
accurate in the short term.
‣Exploratory Methods
‣ They can predict customer response (the dependent variable) by identifying
the factors that drives it (independent variable-s), estimating their effects
and projecting those forward into the future.
‣ Explanatory techniques are generally more useful than time series analyses
when looking at the medium- to long-term future.
The Customer Database: Analysis and Applications
28
Lifetime Value of a Customer (LVC)

‣The lifetime value of a customer can be calculated as

the discounted stream of net revenues that a
customer will generate over the period of his/her
lifetime of patronage with a company.
‣Briefly, lifetime values are the sum of all future net
incomes expected from a customer translated into
present terms.
The Customer Database: Analysis and Applications
29
3 Important Definitions
‣Net Income: It is the income after direct costs have been taken
into account.
Direct costs are costs that can be attributed directly to product/service
production.

‣ Reference to the Present Day: This is an adjustment made to

the future incomes depending on “discount rate” that has been
set. The principle is that money in the future is always worth
less the same amount today.
‣ Net Present Value: If you invest any amount of money it will
worth more in the upcoming years.
‣ Discounted Cash Flow (DCF): If I lend you £100,000, and you pay it back
over five years without interest, DCF calculates how much I will lose at
the end of the period given assumed lost interest rates.
The Customer Database: Analysis and Applications
30
How Long is a Lifetime

‣This varies according to several factors:

‣What type of products you sell
‣ A car has a shorter lifetime than a 20-year mortgage.
‣Service and promise levels
‣ How well you service your customers
‣ How your product lives up to the promises you make.
‣Market stability
‣ In fast moving markets like IT, it is hard to make sound projections about
the length of relationships.
‣Company objectives
‣ Financially strong, strategically orientated companies have more confidence
in taking the long view.
The Customer Database: Analysis and Applications
31
Factors Affecting LTVs

‣Method of recruitment
‣People recruited via word of mouth (‘Member Get Member’ or
‘MGM’) are generally better than those gained via discount
offers.
‣Medium
‣Some newspapers (for example) will generate lower quantity
but higher quality customers.
‣Geo-demographic factors
‣A garden centre will be more interested in owners of large
gardens than those of small gardens.
The Customer Database: Analysis and Applications
32
How to Use LTVs

‣Planning customer acquisition strategies

‣It may be worth using a certain medium which costs more per
advert but returns a higher LTV customer.
‣To allocate acquisition funds
‣If we know a customer’s projected LTV, we can set expenditure
with that rather than initial purchase only
‣This will allow us to make a more generous offer than our
competitors.
‣To select products for customer acquisition offers
‣Just as some media work better than others, some offers do
too – which offers recruit good LTV customers and which just
want the freebie?
The Customer Database: Analysis and Applications
33
Two Ways to Calculate LTVs

‣Historical calculation
‣The simpler method – based on known facts and historical
data rather than projections and estimates.
‣Easier to tie to financial statements (balance sheet, P&L, etc.)
so the accountant will be more persuadable.
‣Projective calculation
‣Uses previously observed trends and projects them into the
future
‣Useful if historical data doesn’t exist
‣More appropriate for doing ‘what if’ analysis.
The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

Example of Simple LTV Calculation – Yearly Basis

Segment Yearly Spending Lifetime (Amount of Years) Total Lifetime Value

Segment A $100 10 years 1000

Segment B $150 15 years 2250

Segment C $125 20 years 2500

The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

Example of Simple LTV Calculation – Monthly Basis

Lifetime (Amount of
Segment Monthly Spending Yearly Base Total Lifetime Value
Years)

x12
Segment A $100 1200 10 years 12000

x12
Segment B $150 1800 15 years 27000

x12
Segment C $125 1500 20 years 30000
The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

Simple LTV Exam Type Question

Lifetime (Amount of
Segment Monthly Spending Yearly Base Total Lifetime Value
Years)

Segment A 1200 10 years 12000

Segment B $150 15 years 27000

Segment C $125 20 years 30000

Segment D $1440 20 years 28800

The Customer Database: Analysis and Applications
34
Calculating Lifetime Value

‣3 pieces of data is needed. These are:

(1)The number of years a customer buys from a company.
(2)The percentage of customers remaining loyal to the
company.
(3)The amount spent per annum.
The Customer Database: Analysis and Applications
35
Steps in Calculating LTVs

‣Identify a cell (customer).

‣Record the sales revenues for this group each year.
‣Calculate costs/profits for each period.
‣Discount each profit figure to arrive at the present
value (NPV) each year.
The Customer Database: Analysis and Applications
36
Example of Historical LTV Calculation

Year%1% Year%2% Year%3% Year%4% Year%5%

A" Customers" 1,000" 721" 536" 410" 323"
B" Average"sales"p.a." £"100" £"105" £"110" £"115" £"121"
C" Total"sales" £"100,000" £"75,705" £"58,960" £"47,150" £"39,083"
D" Net"profit(20%)" £"20,000" £"15,141" £"11,792" £"9,430" £"7,817"
E" Discount"rate"(6%)" 100.0%" 94.0%" 88.4%" 83.1%" 78.1%"
F" NPV"contribuMon" £"20,000" £"14,233" £"10,424" £"7,836" £"6,105"
G" CumulaMve"NPV" £"20,000" £"34,233" £"44,657" £"52,493" £"58,598"
H% LTV"at"NPV% £"20% £"34,23% £"44,66% £"52,49% £"58,60%
I" RetenMon" 72,1%" 74,3%" 76,5%" 78,8%" 81,2%"

Table"3.3!!Example"lifeMme"value"calculaMon"(£)"
The Customer Database: Analysis and Applications
37
Calculating Allowable Marketing Spend Per Customer

(1)The amount we can spend to attract a customer

for the first time (acquisition/recruitment cost)
(2)The amount we can spend on existing customers in
order to keep the with us (retention cost)
Assume that, depending on our budget we set the
acquisition cost to be 20£, and the retention cost to be 3£.
The Customer Database: Analysis and Applications
38
Calculating Allowable Marketing Spend Per Customer - 2
Acquisition marketing budget = allowable recruitment spend
per customer* target number
of customers
=£20*1000 customers
=20.000 £

This figure can also be used for budgeting purposes and to

calculate the response rate for a direct marketing campaign.

Our budget for acquisition is 20.000£ - Suppose that with that

amount we can send out 40.000 mailers. And we need to recruit
1000 customers.
1000 customers/40.000 mail= 0.25
A 25% response rate is needed.
The Customer Database: Analysis and Applications
39
Calculating Allowable Marketing Spend Per Customer - 3
Retention budget for years 2 to 5 can calculated as follows:
The formula
Retention marketing budget = number of customers*allowable
marketing spend for retention.
Year 2= 721*3 = 2163
Year 3= 536*3 = 1608
Year 4= 410*3 = 1230
Year 5= 323*3 = 969

For a single customer the allowable marketing spend is 20£

(acquisition cost) + 3£ (allowable marketing spend for retention) *
number of years
=20£ + 3£*4 years
=32£ - for a customer who stays with the company for 5 years- .
The Customer Database: Analysis and Applications
40
Calculating Expected Return on Investment (ROI)
‣This figure can be calculated by looking at ROI per
customer over the expected lifetime.
‣The formula
Expected ROI per customer = The lifetime value income
per customer - the marketing
spend per customer
= 58.70£ - 32.00£ = 26.70£

The Expected ROI per annum = 26.70£ (the expected ROI per
customer) / 5 years (the expected lifetime of a customer) =
5.34£
The Customer Database: Analysis and Applications
41
FRAC Scoring

‣Frequency, Regency, Amount and Category.

‣Frequency: the time elapsed between purchases
‣Regency: the date of the most recent purchase
‣Amount: the average value of the customers purchases
‣Category: segment which the product bought under.
The Customer Database: Analysis and Applications
42
What is FRAC Scoring

‣FRAC Scoring is used to assess customers on the

database based on their purchase behavior.
‣At the end of the analysis each customer has a number,
and a score attached to them.
‣Direct marketers use these scores to predict the
customers’ likelihood of responding to direct marketing
campaigns.
‣FRAC Scoring is also used as a segmentation tool.
The Customer Database: Analysis and Applications
43
Decisions in FRAC Scoring

‣Category
‣“Past transactions are one of the most important indicators of
future transactions.”
‣This principle highly depends on the market.
‣The Regency Points
‣24 points – purchase in the current quarter
‣12 points – purchase in the last 6 months
‣09 points – purchase in the last 9 months
‣03 points – purchase in the last 12 months
‣Frequency Points
‣Number of points = no. of purchases*4
‣Monetary Points
‣No. of points = 10% of purchase value, with a ceiling of 9 points.
The Customer Database: Analysis and Applications
44
Ranking Customers with FRAC Scoring

‣After each customer is scored according to the their FRAC

data, direct marketers organize them according to that
scores.
‣Customers are assigned to cells (segments) according to
their predicted response levels.
‣The cells know as DECILEs
‣Segment 1 (or Decile 0-10) : the most responsive
customers
‣Segment 10 (Decile 91-100): the least responsive
customers
The Customer Database: Analysis and Applications
45
Ranking Customers with FRAC Scoring
Segment' Score' Decile' Responses' Typical'Marke6ng'Decision'
1" 60+" 0&10" 600" Five"outbound"telephone"contacts"
2" 54&59" 11&20" 400" Four"outbound"telephone"contacts"with"mailer"to"follow"
Three"outbound"telephone"contacts"with"two"mailers"to"
3" 48&53" 21&30" 300"
follow""
4" 42&47" 31&40" 280" Two"outbound"telephone"contacts"with"two"mailers"to"follow"
5" 36&41" 41&50" 120" One"outbound"telephone"contact"with"two"mailers"to"follow"
6" 30&35" 51&60" 100" One"outbound"telephone"contact"with"one"mailer"to"follow"
7" 24&29" 61&70" 80" Three"mailers"
8" 18&23" 71&80" 60" Two"mailers"
9" 12&17" 81&90" 30" One"mailer""
10" 6&11" 91&100" 10" No"acEvity"
Table"3.4!!Gains"Table"DescripEon"

Depending!on!the!data!direct!marketers!make!decisions!about!budget!alloca5ons.!As!
being!the!most!responsive!segment,!the!1 !segment!is!allocated!the!largest!possible!
st

budget.!!
The Customer Database: Analysis and Applications
46
FRAC Scoring in Practice

‣Each segment represents 10% of the total customers.

‣We assume that we start with the most responsive
segment and work our way down.
‣Re-labeling the column – cumulative percentage of
customers to be mailed.
‣Total number of responses are 1980.
The Customer Database: Analysis and Applications
47
FRAC Scoring in Practice - 2

‣4th column ( % of responses received)

‣Segment 1 – 600/1980 = 30%
‣Segment 2 – 400/1980 = 20%
‣Segment 3 – 300/1980 = 15%
‣Segment 4 – 280/1980 = 14%
‣Segment 5 – 120/1980 = 6%
‣Segment 6 – 100/1980 = 5%
‣Segment 7 – 80/1980 = 4%
‣Segment 8 – 60/1980 = 3%
‣Segment 9 – 30/1980 = 2%
‣Segment 10 – 10/1980 = 1%
The Customer Database: Analysis and Applications
48
FRAC Scoring in Practice - 3
! Using this table we can create different graphs to
visualize the data.
Segment' Baseline' Cumula/ve'baseline' %'of'responses'received' Cumula/ve'Responses'
10%$ 10%$ 10%$ 30%$ 30%$
20%$ 10%$ 20%$ 20%$ 50%$
30%$ 10%$ 30%$ 15%$ 65%$
40%$ 10%$ 40%$ 14%$ 79%$
50%$ 10%$ 50%$ 6%$ 85%$
60%$ 10%$ 60%$ 5%$ 90%$
70%$ 10%$ 70%$ 4%$ 94%$
80%$ 10%$ 80%$ 3%$ 97%$
90%$ 10%$ 90%$ 2%$ 99%$
100%$ 10%$ 100%$ 1%$ 100%$
Table$3.5$!Gains$Table$
 The Customer Database: Analysis and Applications
49
FRAC Scoring in Practice - 4

100%#
Percentage)of)all)responses)

80%#

60%#

40%#

20%#

0%#
10%# 20%# 30%# 40%# 50%# 60%# 70%# 80%# 90%# 100%#
Cumula2ve#Baseline# Cumula2ve#Responses#
Percentage)of)Customers)Contacted)
Figure#3.1#!Cumula2ve#Gains#Chart#
The Customer Database: Analysis and Applications
50
FRAC Scoring in Practice - 5
Segment'performance'against'expecta3on'
35%#

30%#
Percentage'of'all'responses'

25%#

20%#

15%#

10%#

5%#

0%#
1# 2# 3# 4# 5# 6# 7# 8# 9# 10#
Segment'
Baseline# %#of#responses#

Figure#3.2#!Performance#of#each#cell#versus#expectaCon#
The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

RFM Analysis
RFM analysis is a marketing technique used to quantitatively rank and group customers based on their
transaction history to identify customers who are more likely to respond to promotions and also to
understand customer behavior.

The methodology is straightforward yet powerful, focusing on three main dimensions:

1.Recency (R): How recently a customer made a purchase. A more recent purchase is generally an
indicator that the customer is more engaged and likely to make future purchases. Recency is often
calculated as the number of days since the last purchase.

2.Frequency (F): How often a customer makes a purchase within a given time frame. A higher frequency
indicates a more loyal or engaged customer. Frequency can be counted as the total number of purchases
made in a defined period.

3.Monetary Value (M): How much money a customer spends over a period. Higher monetary value
customers are often more valuable to the business. This is typically the total spend by the customer in
the same period considered for frequency.
The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

RFM Analysis
To perform RFM analysis, customers are ranked based on each of these dimensions, typically on a scale
(for example, 1 to 5, or 1 to 10), where higher values indicate better performance on that dimension
(more recent, more frequent, higher spend).

The exact method to assign these scores can vary, but here are general approaches:

•Recency: Customers can be grouped based on the time elapsed since their last purchase, with groups
assigned scores from high (most recent) to low (least recent). For instance, you might divide customers
into quintiles based on recency and assign scores from 5 (most recent) to 1 (least recent).

•Frequency and Monetary: Similarly, customers are grouped and scored based on how many times
they've purchased and how much they've spent. Higher scores are assigned to higher frequencies and
monetary values.
The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

Application of RFM Analysis

Businesses use RFM analysis to target specific segments with campaigns that are more relevant to their
behavior.

For example:
➢ High-Value Customers (High R, High F, High M): These customers are likely the most loyal and
profitable, and marketing efforts might focus on loyalty programs, upselling, or cross-selling.

➢ At Risk (Low R, High F, High M): Customers who used to shop frequently but haven't made recent
purchases might be targeted with re-engagement campaigns or special offers to encourage them to
return.

➢ New Customers (High R, Low F, Variable M): Different strategies might be applied to turn new
customers into regulars, depending on their initial spending.
The C u s t o m e r D a t a b a s e : A n a l ys i s and A p p l i c a t i o n s

RFM Analysis
Frequency
Customer ID Recency (Days) Monetary (USD) Segment
(Transactions)

C001 5 10 500 High Value

C002 100 2 150 Low Value

C003 15 5 300 High Value

C004 30 3 200 Medium Value

C005 60 4 250 Medium Value

C006 10 8 450 High Value

The Customer Database: Analysis and Applications
51
Exploratory Analysis Techniques

‣Cluster Analysis
‣Regression Analysis
‣CHAID Analysis
The Customer Database: Analysis and Applications
52
Cluster Analysis

‣Used to create quantitative profiles of customers.

‣Tries to aggregate customers into groups.
‣The customers are assigned to groups based on 2
principles:
‣Inter-group differences
‣Intra-group similarities
‣This means that based on the clustering parameters, the
customer are assigned to same group has similar
characteristics, and the customers in different groups differ
according to their characteristics.
The Customer Database: Analysis and Applications
53
Cluster Analysis - 2

‣Most of the time cluster analysis is conducted with

exploratory purposes.
‣Helps marketers to understand how the mass audience is
divided into various groups.
‣Cluster analysis does not give information about how a
dependent variable will vary across different customer
groups.
‣Further analysis is needed to test such assumptions.
The Customer Database: Analysis and Applications
54
Regression Analysis

‣Most of the time cluster analysis is conducted with

exploratory purposes.
‣Two sets of variables are used:
‣Independent
‣Dependent
‣Tries to formulate the association between independent
and dependent variables.
‣E.g.
‣Independent Variable: Advertising spending
‣Dependent Variable: Sales Volume
The Customer Database: Analysis and Applications
55
CHAID (Chi-Square Automated Identification Detector) Analysis

It is simpler than regression and as a result a very popular

technique in the field.
Regression) CHAID)
Complexity) Complex+ Simpler+
Min+sample+size) 300+ 1,000+
So7ware) Extra+needed+ No+extra+needed+
Data+type) Be>er+with+conAnuous+ Be>er+with+categorical+
Output+type) EquaAon+which+allows+score+ Tree+segmentaAon+
model+
InteracAon+of+variables) Ignores+it+ Detects+it+

Table+3.7+!Regression+analysis+versus+CHAID+
Source:+Portlock+(1992).+
The Customer Database: Analysis and Applications
56
Regression vs. CHAID

‣Regression
‣ Regression analysis takes a dependent and an independent variable
and uses statistical formula to describe how strongly the two are
linked
‣ An independent variable may be advertising
‣ A dependent variable may be sales.
‣CHAID
‣ Each record is put into a discrete cell, generated by the analysis
‣ The essence of the technique is that:
‣ The marketer first chooses the likely discriminating variables, i.e. those best explaining
who responds
‣ CHAID uses a statistical technique to confirm which of these variables best explains the
variance in response.
‣
The Customer Database: Analysis and Applications
57
Steps in CHAID Analysis

‣ STEP 1
‣ Compare the records of ‘responders’ with those of ‘non-responders’.
‣ STEP 2
‣ Computer model asks, ‘Which variable best explains the difference between responders
and non-responders?’ This is done in a systematic rather than an intuitive way.
‣ STEP 3
‣ Split the responders’ file into two cells according to the biggest difference in response
between the cells.
‣ STEP 4
‣ Model then looks at each of the segments separately to see if they can be broken down
any further. Different variables may now be appropriate to segment the cells further.
‣ STEP 5
‣ Continue process until cell sizes are too small for it to be of value to split them further.
‣ STEP 6
‣ Output as a tree segmentation, as illustrated.
 The Customer Database: Analysis and Applications
58
Tree Segmentation for CHAID Analysis
Overall response =4.5%
Total sample size = 100% Split the respondents le into two cells
according to the biggest di erence in response
Cell 1 Cell 2 between the cells (most discrimina ng
Credit card owners Non- Credit cards variable)
Response = 6.3% Response = 6.3%
62% of le 38% of le
Then each cell is examined to discover if they
can be broken down any further
Cell 3 Cell 4
Female Male
Response = 5.8 % Response = 6.8 %
31 % of le 31% of le The process con nues un l cell sizes are too
small of value to split them further
Cell 5 Cell 6 Cell 7
Annual order value Annual order value Annual order value
<50 50 - 100 100+ As a result a tree diagram provides informa on
Response = 3.5 % Response = 8.7 % Response = 11.3 %
15% of people 10% of le 6% of le about customer segments according to their
marke ng poten al.
Figure'3.3!!Tree'segmenta0on'for'CHAID'
fi
fi
fi
fi
fi
fi
ti
ti
ti
fi
ti
ff
ti
ti