22. Evaluation of an Analytical Model
YEAR MONTH DISCREPENCY HEADLINES REFERENCE
2015 6 SALES >> FORECAST
2015 7 SALES >> FORECAST
2015 8 SALES >> FORECAST
2015 9 SALES >> FORECAST
2015 10 SALES >> FORECAST
2015 11 SALES >> FORECAST
2015 12 FORECAST >> SALES
2016 1 FORECAST >> SALES
2016 11 FORECAST >> SALES
2016 12 FORECAST >> SALES
2017 1 FORECAST >> SALES
2017 2 FORECAST >> SALES
"Indian automotive industry shows modest recovery in 2015"
https://www.businesstoday.in/topics/year-roundup:-2015/indian-automotive-industry-shows-modest-recovery-in-2015/story/2
https://www.livemint.com/Industry/cxtwVL8m1rjKunGEsYimyK/Passenger-vehicle-sales-rise-1429-car-sales-up-449-in-N.
2017 11 SALES >> FORECAST Passenger vehicle sales pick up in November
Ten events that dominated auto Inc in 2016. 2016
has been a year of mixed fortunes for the Indian
automobile industry. There have been launches
which raised hopes for manufacturers to regain
https://auto.economictimes.indiatimes.com/news/industry/top-10-events-that-dominated-indian-auto-inc-in-2016/5626
23. What is
Analytics?
1-23
Analytics is the use
of:
data,
information technology,
statistical analysis,
quantitative methods,
and
mathematical or
computer-based models
to help managers gain improved
insight about their business
operations and
make better, fact-based decisions.
24. What is Analytics?
1-24
Importance of Business
Analytics
• There is a strong relationship of BA
with:
• - profitability of businesses
• - revenue of businesses
• - shareholder return
• BA enhances understanding of data
• BA is vital for businesses to remain
competitive
• BA enables creation of informative
reports
25. Model:
An abstraction or representation of a real system, idea, or object
Captures the most important features
Can be a written or verbal description, a visual display, a
mathematical formula, or a spreadsheet representation
Decision Models
1-25
26. Decision
Models
1-26
Predictive Decision Models often incorporate
uncertainty to help managers analyze risk.
Aim to predict what will happen in the
future.
Uncertainty is imperfect knowledge of what
will happen in the future.
Risk is associated with the consequences of
what actually happens.
27. Decision Models
Prescriptive Decision Models help decision makers identify the best
solution.
Optimization - finding values of decision variables that minimize (or
maximize) something such as cost (or profit).
Objective function - the equation that minimizes (or maximizes) the
quantity of interest.
Constraints - limitations or restrictions.
Optimal solution - values of the decision variables at the minimum (or
maximum) point.
1-27
28. Scope of
Business
Analytics
1-28
Retail Markdown Decisions
• Most department stores clear seasonal inventory by reducing prices. The question is:
When to reduce the price and by how much?
• Descriptive analytics: examine historical data for similar products (prices, units sold,
advertising, …)
• Predictive analytics: predict sales based on price
• Prescriptive analytics: find the best sets of pricing and advertising to maximize sales
revenue
41. Data for
Analytics -
Categorical
(nominal) Data
Data placed in categories according to a
specified characteristic
Categories bear no quantitative
relationship to one another
Examples:
- customer’s location (America, Europe,
Asia)
- employee classification (manager,
supervisor,
associate)
1-41
42. Data for Analytics - Ordinal Data
Data that is ranked or ordered according to some relationship with one another
No fixed units of measurement
Examples:
- college football rankings
- survey responses
(poor, average, good, very good, excellent)
1-42
43. Data for
Analytics -
Interval Data
Ordinal data but with constant
differences between observations
No true zero point
Ratios are not meaningful
Examples:
- temperature readings
- SAT scores
1-43
44. Data for
Analytics -
Ratio Data
Continuous values and have a
natural zero point
Ratios are meaningful
Examples:
- monthly sales
- delivery times
1-44
45. A common reorder policy in a supply chain is to produce enough to have a
low percent chance, say, a 5 percent, of running out of a product.
Suppose the retailer wanted to know how many cakes to order to reduce
stock out chances.
For cakes, demand for cakes that exceed x or a 95 percent chance of
demand materialising. This would imply that you should produce x cakes,
where there is a 5 percent chance of stock out situation.
46. What if you high
sales in a day and
you want to know
whether you
should increase
your production
levels ?
• The percent rank function provides the rank of
an observation compared to all values in data
set. The ranking returned is percentage rank
and not absolute rank
• On December 27 2015, the store sold ____
cakes. What is the chance that they will again
sell those many cakes. Use the
PERCENTRANK(range, x) function.
• How would you interpret the answer ?
47. Cookies - More computing with fn
R3=COUNTIFS(daywk,Q3,promotion,"p
romotion")gives num of promotions
each day
U3=AVERAGEIF(daywk,T3,Cookies)give
s average sale.
Now lets use sumif for each product
each month
R14=SUMIFS(INDIRECT(R$13),Namem
onth,$Q14). Why indirect ?
R29=AVERAGEIFS(INDIRECT(R$28),Na
memonth,$Q29)
49. Summarizing data
with Subtotals feature
• Second step : put cursor
anywhere and go to data
ribbon
• Go to outline tab and select
subtotals.
• A dialog box will open. Fill as
shown here
50. Summarizing data
with Subtotals feature
• Third step : put cursor anywhere
and go to data ribbon
• Go to outline tab and select
subtotals.
• The next subtotal dialog box is to
create a nested row of average
sales during promotion or no
promotion
• Do not check replace currect
subtotals and summary below
data, because we want to build
additional lines of analysis.
51. Spreadsheet Quality
Verification is the process of ensuring that a model is accurate and
free from logical errors.
Below are three approaches to spreadsheet engineering that can
improve spreadsheet quality:
1. Improve the design and format of the spreadsheet itself.
2. Improve the process used to develop a spreadsheet.
3. Inspect your results carefully and use appropriate tools available in
Excel.
Spreadsheet Modeling and Engineering
2-52
52. Are daily cookie sales symmetric or skewed ?
• Use the skew function
• = =SKEW(Cookies)
• The answer is 0.440287
• Since the value is between +1 and -1 hence the cookies sales are
symmetric
53. By the rule of thumb you would expect on 95 percent of all days daily smoothie
demand will be between _______ and _____.
• Answer
• =AVERAGE(Smoothies)-2*STDEV(Smoothies) gives you the lower
range
• And
• =AVERAGE(Smoothies)+2*STDEV(Smoothies) gives you the upper
range
• So the answer is between 70 and 367
54. There is a 10 percent chance daily demand for smoothies will exceed _________
• Answer: At 90th percentile the sale of cookies is 317. Function
=PERCENTILE.EXC(Smoothies,0.9).
• The inference that can be drawn is that there is only a 10% chance
that the daily demand for cookies will exceed this number of 317. The
answer is 317
55. Fraction of cookie sales on a Monday in january
• SUMIFS(sum_range, criteria_range1, criteria1, [criteria_range2,
criteria2], ...)
=SUMIFS(Cookies,Namemonth,$Q10,daywk,R$9)
Actual Cookie Sales Monday Tuesday
Wednesd
ay Thursday Friday Saturday Sunday
January 5604 5649 6921 6928 8749 8081 6138
February 6184 5641 5887 5732 7527 7710 6205
March 6573 6039 5709 5674 8173 9480 9116
April 6402 6647 6475 6092 6597 7434 7093
May 5295 5231 7034 7254 8695 9038 7318
June 6187 6270 5542 5519 7131 8588 8061
July 6327 6704 6960 6498 7931 7525 8283
August 6342 5853 5821 6334 7898 10079 8347
September 7058 7149 6418 5835 7097 7980 8068
October 6568 6672 6451 8051 8011 8389 7385
November 5805 6300 6334 6056 7930 10529 8663
December 6862 7303 6886 6749 7446 7303 7971
Fraction Monday Tuesday
Wednesd
ay Thursday Friday Saturday Sunday
January 0.95% 0.95% 1.17% 1.17% 1.48% 1.37% 1.04%
February 1.04% 0.95% 0.99% 0.97% 1.27% 1.30% 1.05%
March 1.11% 1.02% 0.96% 0.96% 1.38% 1.60% 1.54%
April 1.08% 1.12% 1.09% 1.03% 1.11% 1.26% 1.20%
May 0.89% 0.88% 1.19% 1.23% 1.47% 1.53% 1.24%
June 1.05% 1.06% 0.94% 0.93% 1.20% 1.45% 1.36%
July 1.07% 1.13% 1.18% 1.10% 1.34% 1.27% 1.40%
August 1.07% 0.99% 0.98% 1.07% 1.33% 1.70% 1.41%
September 1.19% 1.21% 1.08% 0.99% 1.20% 1.35% 1.36%
October 1.11% 1.13% 1.09% 1.36% 1.35% 1.42% 1.25%
November 0.98% 1.06% 1.07% 1.02% 1.34% 1.78% 1.46%
December 1.16% 1.23% 1.16% 1.14% 1.26% 1.23% 1.35%
56. Find the median sale of coffee on days in which at least 75
cakes were sold.
• =MEDIAN(IF(Cakes>=75,Coffee," "))
• When you are entering this formula instead of pressing enter press
“Control+Shift+Enter”. This function creates a new array of coffee
sales, wherein cakes>=75. then it will find a median of this new array
of coffee sales.
The answer is 404
61. Pred Y =2x+1,
constant=1, slope=2, x = independent variable
Observed Y Predicted Y =2x+1 Error = Obs Y – pred Y SSE = SUM(….)
4 3 1; 1*1
6 5 1; 1*1
5 7 -2; -2*-2 =4
0; 6
62.
63. Identifying Interaction and non linear effects
in regression
• Open file price and ads
• Obj: investigate interaction
and non linear effects of
price and ad on sales
• Create three new variables
• (1) A*P
• (2)A square
• (3)P square
64. Identifying Interaction and non linear effects in regression (contd)
• Check F value
• Check P values
• If interaction and
non linear effect
significant
66. Uses of Modelling Trend and Seasonality
• Using a 12-month or 4-quarter period enables marketing managers to
easily see the trend in a product’s sales and forecast forward.
• Separate the monthly de-seasonalised sales trend from the influence
of seasonality
• Quantify the seasonal surge or dip of each month/season on sales.
• Assist functions of demand and supply management
• Plan promotions, distribution, and pricing strategies on the basis on
anticipated surge due to seasonality
67. Modelling Trend and Seasonality
An Additive Model
Objective: Identify the seasonal influence of each month on sales.
Equation: Predicted Period t Sales = Base + Trend*Period Number + Seasonal
Index for Month t
■ Base: The base is the best estimate of the level (without seasonality) of
monthly airline miles at the beginning of the observed time period.
■ Trend: The trend is the best estimate of the monthly rate of increase in
airline miles traveled.
■ Seasonal Index: Each month of the year has a seasonal index to reflect
if travel during the month tends to be higher or lower than average.
68. Estimating trend and seasonality – additive model
1. Enter trial values of the base and trend in cells B2 and B3. Name cell B2
baseadd and cell B3 trend.
2. Enter trial seasonal indices in the range B5:B16.
3. In cell B18, average the seasonal indices with the formula =AVERAGE(B5:B16).
The Solver model can set this average to 0 to ensure the seasonal indices average
to 0.
4. Copy the formula =baseadd+trend*D9+VLOOKUP(F9,$A$5:$B$16,2) from H9
to H10:H42 to compute the forecast for each month.
5. Copy the formula =G9-H9 from I9 to I10:I42 to compute each month’s forecast
error.
6. Copy the formula =(I9^2) from J9 to J10:J42 to compute each month’s squared
error.
7. In cell K6, compute the Sum of Squared Errors (SSE) using the formula
=SUM(J9:J42).
69. Modelling trend and seasonality (contd) through solver
8. Now set up the Solver model, as shown in Figure
12-5. Change the parameters to minimize SSE and
constrain the average of the seasonal indices to 0.
Do not check the non-negative box because some
seasonal indices must be negative.
The forecasting model of Equation 1 is a linear
forecasting model because each unknown parameter
is multiplied by a constant.
When the forecasts are created by adding together
terms that multiply changing cells by constants, the
GRG Solver Engine always finds a unique solution to
the least square minimizing parameter estimates for a
forecasting model.
70. A Multiplicative Model with Trend and
Seasonality
Predicted Period t Sales = Base * (Trend^t) * (Seasonal Index for Month t)
■ Trend: The trend now represents the percentage monthly increase in the level
of airline miles.
1.03 = means monthly air travel is increasing 3 percent per month,
0.95 = means monthly air travel is decreasing at a rate of 5 percent per month.
■ Seasonal Index: The seasonal index for a month now represents the percentage
by which airline travel for the month is above or below an average month.
1.16 = means July has 16 percent more air travel than an average month,
0.83 = means February has 17 percent less air travel than an average month.
Note: multiplicative seasonal indices must average to 1.
71. A Multiplicative Model with Trend and
Seasonality (contd)
• Change the forecast formula
in Column H9 =
base*(trend^D9)*VLOOKUP(F9
, $A$5:$B$16,2)
This is a nonlinear model and
hence from the GRG tab, select
multistart engine.
For efficiency put upper
bounds on all changing cells.
72. Configuring solver in GRG multistart for airline
miles file
• for the estimated parameters. For
example, a seasonal index will
probably be
• you can choose an upper bound of 3
for each seasonal index
• and an upper bound of 2 for the
trend.
• For this example, choose an upper
bound of 100 for the base.
• $B$18 =1 to ensure that the seasonal
indices average to 1.
73. • ■ The estimated base level of airline miles is 37.4 billion.
• ■ You can estimate airline miles increase at a rate of 0.15 percent per
month or 1.0014912 – 1 = 1.8 percent per year.
• ■ The busiest month for the airlines is July, when miles traveled are
16 percent above average, and the least busy month
74. Additive vs multiplicative model – select the
one with lower sum of squared errors.
additive
• . If per period growth is
independent of the current
sales value, the additive
trend model will probably
outperform the
multiplicative trend model
multiplicative
• . On the other hand, if period
growth is an increasing function
of current sales, the
multiplicative trend model will
probably outperform the
additive trend model.
75. Understanding the concept of exponential
smoothing
Simple moving average - the past observations are weighted equally,
Exponential functions are used to assign exponentially decreasing weights over
time.
The weights assigned to recent past may be different from far past
Current forecast est = alpha*current obs forecast(t) + (1-alpha)*(Smoothened estimate(t-1)
76. Winters method of forecasting
• The base, trend and seasonal
indices are not constant but
continuously evolving. They
are exponentially smoothened
versions of themselves.
• The Base is called level of
Series and after observing data
through the end of the month
t you can estimate the level,
trend and seasonal index.
78. Winters method (contd)
Ft,k as your forecast (F) after period t for the period t + k. The
equation to compute estimated base k periods forward.
79. Winters method(contd)
• 1. In H11:J11, enter trial values (between 0 and 1) for the smoothing
constants.
• 2. In C26:C113, compute the updated series level with Equation 1 by
copying the formula =alp*(B26/H14)+(1–alp)*(C25*D25) from cell
C26 to C27:C113.
• 3. In D26:D113, use Equation 2 to update the series trend. Copy the
formula =bet*(C26/C25)+(1-bet)*D25 cell from D26 to D27:D113.
80. Winters method(contd)
• 4. In H26:H113, use Equation 3 to update the seasonal indices. Copy
the formula =gama*(B26/C26)+(1-gama)*H14 from cell H26 to
H27:H113.
• 5. In E26:E113, use Equation 4 to compute the forecast for the current
month by copying the formula =(C25*D25)*H14 from cell E26 to
E27:E113.
• 6. In F26:F113 compute each month’s error by copying the formula
=(B26-E26) from cell E26 to E27:E113.
81. Winters
method(contd)
7. In G26:G113, compute the
squared error for each month by
copying the formula =F26^2 from
cell F26 to F27:F113. In cell G21
compute the Sum of Squared Errors
(SSE) using the formula
=SUM(G26:G113).
8. Now use the Solver to determine
smoothing parameter values that
minimize SSE.
The Solver finds that alp = 0.55,
bet = 0.05, and gamma = 0.59.
82. Forecasting future months May 2012- with
winters smoothing constants
• alp = 0.55, beta = 0.05,
and gamma = 0.59.
• Copying the formula
=($C$113*$D$113^B116)*
H102 from cell D116 to
D117:D123.
• G22=STDEV(F26:F113)
provides std. dev. of errors
= 0.9369. Forecast would
be true within 1.88 Bn
83. Another method to minimise error.
• Calculate absolute percentage error in J26 = =ABS(B26-E26)/B26
• Calculate Mean percentage error in J24 = =AVERAGE(J26:J113)
84. Method to develop forecasts using Ratio to
Moving average Method
■ Estimate the de-seasonalized level of the series during each period
(using centered moving averages).
■ Fit a trend line to your de-seasonalized estimates .
■ Determine the seasonal index for each quarter and estimate the
future level of the series by extrapolating the trend line.
■ Predict future sales by re-seasonalizing the trend line estimate.
86. Terms in a Conjoint Analysis
Attributes – characteristics
consumer consider when
evaluating products
Attribute levels – degree to which
attribute is present, or attribute
types
Bundles – set of attributes in the
product or service
Conjoint Analysis – examine
tradeoffs consumers make to
determine marketable
combinations( or bundles) of
attributes at different levels
Part-worths – Conjoint analysis
decomposes overall preference
into values that show
utilities(usefulness) of particular
level of attribute. These utilities
are called part-worth.
Profile – specific bundles
preferred by customer segment
88. Prepare for Conjoint – Know the imp
attributes
• Interview customers, conduct an importance survey, read customer reviews,
trade information >> to sense key attributes that customers value
• Avoid the unimportant or hygiene attributes and for the motivators, identify
levels of attributes
• The total number of cards is level^attributes. So if there are 2 attributes with
3 levels you need 3^2 or 9 cards.. Exponentially increase!!!
• Apply fractional factorial techniques or orthogonal arrays
Not at all
important
Less imp neutral Somewhat
important
Screen Size X
Battery Life
89. Collect Preference Data
• Pairwise Comparison – easier than ranking, too many comparisons
required for relative comparison.
• Rank Ordering – see all bundles at a time. Easy to rank the best and
worst, the middle may get muddled
• Rating Scale – each bundle gets rated on a scale (1-100, 1-10, 1-9,1-
7). Easy to rate each alternative independently, easy to compute.
Some respondents are unable to distinguish for fine rating scales.
• Also explore self segmentation by the consumer on a behavioural or
important segment descriptor. Collect demographic profiling data
91. Calibrating regular vs logistic regression model
• The probability of all the observed phenomena being repeated
• Maximizing the Likelihood of
P(a) Ո P(b) Ո P(c) Ո P(d) = P(a) x P(b) x P( c) x P(d)
Solver will maximise
The interpretation is for any age, 1 year increase in age increases the odds
ratio by e0.1281 = 13.7 %
P(a) 20 0
P(b) 23 0
P( c) 24 0
P(d) 25 1
92. Interpretation of Logit regression
Suppose the slope =0.1281
and intercept=-5.661.
Hence for a 44 year old ,
prediction of subscription is
The interpretation is for any
age, 1 year increase in age
increases the odds
ratio by e0.1281 = 13.7 %
93. Quotes by researchers -When conjoint and
when discrete choice
"The difference between discrete choice models and conjoint models is that discrete choice models
present experimental replications of the market with the focus on making accurate predictions regarding
the market, while conjoint models do not, using product profiles to estimate underlying utilities (or
partworths)
“Conjoint analysis makes the most sense in categories where the decision to buy is defined as rational,
and you can categorize the features,”
“It doesn’t make sense for image products, like Coke vs. Pepsi, where the buying decision is most likely
emotional.
But for products and services such as medical devices, high-tech products like cell phones, automobiles,
insurance and financial services – things where the features can be clarified – conjoint analysis done right
is uncanny in its ability to predict consumer behavior.
94. Why or When discrete choice analysis
• A discrete choice analysis helps the marketing analyst
determine what attributes matter most to decision makers and
how levels of each attribute are ranked by decision makers.
• For calibrating model, finding brand weights and price sensitivity of
brand
• For finding brand equity
• For making a market simulator to find profit maximising price
• For finding Nash equilibrium
95. Process
• To begin a discrete choice analysis, decision makers are shown a set of alternatives and asked to choose their most
preferred alternative.
• The analyst must determine a model that is used to “score” each alternative based on the level of its attributes.
• The fraction of decision makers that choose alternative j is assumed to follow the multinomial logit model:
• eVi , where Vj = score for alternative j
• Maximum Likelihood is used to estimate the parameters (such as brand equity and price sensitivity) in the scoring
equation.
• In a discrete analysis, a Chi Square Test based on the change in the Log Likelihood Ratio can be used to assess the
significance of a changing cell.
• The Independence of Irrelevant Alternatives (IIA) that follows from Equation 2 implies that the ratio of the probability to
choose alternative i to the probability to choose alternative j is independent of the other available choices.
• The multinomial logit version of discrete choice enables you to easily compute price elasticities using the following
equations:
• (8) Price Elasticity for Product j = (1 – Prob(j)) * Price(j) * β(j)
• (9) E(k,j) = – Prob(j) * Price(j) * β(j)
96. Multinomial Logit Model –
Read about the analytic technique behind Discrete Choice Theory
Random Utility Theory
The concept of random utility theory provides the theoretical basis for discrete choice analysis. Suppose a
decision maker must choose among n alternatives. You can observe certain attributes and levels for each
alternative.
• The decision maker associates a utility Uj with the jth alternative. Although the decision maker knows these
utilities, the marketing analyst does not. The random utility model assumes the following:
• Here Vj is a deterministic “score” based on the levels of the attributes that define the alternative, and the εj ’s
are random unobservable error terms. The decision maker is assumed to choose the alternative j (j = 1, 2, …,
n) having the largest value of Uj. The εj ’s are independent Gumbel (also known as extreme value) random
variables, each having the following distribution function: F(x)=Probability εj <= x)= , then the probability
that the decision maker chooses alternative j (that is, Uj = max k=1,2,,....,nUk) is given by the following:
• The above equation is analogous to the logit model and is often called the multinomial logit model. It is of
crucial importance because it provides a reasonable method for transforming a customer’s score for each
product into a reasonable estimate of the probability that the person will choose each product. In the rest of
this chapter you will be given the alternative chosen by each individual in a set of decision makers. Then you
use this equation and the method of maximum likelihood (introduced in the slide on calibrating logistic
regression) to estimate the importance of each attribute and the ranking of the levels within each attribute.
97. Steps of discrete choice analysis
• For calibrating model, finding brand weights and price sensitivity of brand
• Score Exp probabilities log of probabilities log of probabilities*frequency
of occurence maximum likelihood solver brand/attribute/price weights
marketing analysis
• For finding brand equity
• Score Exp probabilities- log of probabilities log of probabilities*frequency
of occurence maximum likelihood solver control price weights by using similar
price weights brandweights marketing analysis
• For making a market simulator to find profit maximising price
• Probabilities Prob*100 =demand Profit = R – C max price in solver by
changing price
99. Why S curves
• To forecast new product sales cycle from early sales data
• To schedule the R&D investments
• Strategic planning of new product innovation cycles
100. Two methods of technology forecasting using
s curves to forecast sales of a new product
• Pearl curve
• Inflection point
• Gompertz Curve
• Inflection point
101.
102. Bass Model Forecasting – Why
• forecast product sales before the product comes to market.
• an explanation of how knowledge of new products spreads throughout the market place.
• determine the relative importance of innovators and imitators in driving the spread of the product.
103. Bass Model - Assumptions
• n(t) = Product sales during period t.
• N(t) = Cumulative product sales through period t.
• N = Total number of customers in market; assume that all of them eventually adopt the product.
• P = Coefficient of innovation or external influence.
• Q = Coefficient of imitation or internal influence.
104. The Bass Model Equation Component
number of people (N − N(t − 1)) who have
not yet adopted the product.
P is called the coefficient of innovation or
external influence.
the number of interactions between
previous adopters (N(t − 1)) and people
who have yet to adopt (N − N(t − 1))
This imitation or internal influence
component reflects that previous adopters
tell nonadopters about the product and
thereby generate new adoptions.
Q*
107. Segmentation
Identify segments
Targeting
Select segments
Positioning
Create competitive
advantage
Marketing resources are focused to better meet customers
needs and deliver more value to them
Customers develop preference for brands that better meet
their needs and deliver more value
Customers become brand/supplier loyal, repeat purchase,
communicate favorable experiences
Brand/supplier loyalty leads to increased market share and
creates a barrier to competition
Fewer marketing resources needed over time to maintain
share due to brand or supplier loyalty
Profitability (value to the firm) increases
How STP Creates Value
108. 109
Market Segmentation
• Market segmentation is the subdividing
of a market into distinct subsets of
customers.
Segments
• Members are different between
segments but similar within.
109. 110
Markets are Dynamic
Segments may be unstable over time
• Buyer behavior changes
• Competitors change
• The business environment changes
This means that it is important to view segmentation as . . .
• A process to support business decisions
• Not a static classification of the market
110. A Four-Phase Process for Successful
Segmentation Analysis Project
Internal Assessment
& Planning
Objective(s) of
segmentation
Resources
Constraints
Database
Review
Primary data
already available
Secondary data
…
Prototype
Implementation
Exercises
What ifs?
Relevant groups
involved?
…..
Qualitative
Research
Interview Materials
Development
Qualitative Data
Collection
“Deep needs”
Identification
Decision-Making
Process
Assessment
Quantitative
Survey
Sample Design
Questionnaire
Development
Data Collection
Segmentation
Analysis
Cluster Analysis
Portfolio Analysis
Positioning Analysis
Implementation
Through
Database Tools
Call Center
Web
Sales call
patterns
Promotion
….
Phase I
Planning and Design
Phase II
Qualitative Assessment
Phase III
Quantitative Measurement
Phase IV
Analysis and Implementation
112. 113
Primary Characteristics
of Segments
• Bases—characteristics that tell us why segments
differ (e.g., needs, preferences, decision
processes).
• Descriptors—characteristics that help us find and
reach segments.
(Business markets) (Consumer markets)
Industry Age/Income
Size Education
Location Profession
Organizational Life styles
structure Media habits
113. Descriptor or Identifier variables for
consumer and business markets. (alt
view)
114
How to serve
How to communicate
115. 116
1. Define Segmentation Problem
• View market segmentation problem as a
series of hierarchical stages -- for example
• Identify broad strategic “macro-segments” that
effectively define market structure
• Industry groups
• Product usage (rate of usage, application, etc.)
• Geographic location, etc.
• Within macro-segments, conduct research to
find “micro-segments” for competitive
advantage
• Segmentation on buyer needs and value
• Segmentation on product benefits
116. 117
2. Conduct Market Research
Market research supports the segmentation process
Market
Segmentation
Study
Phase 1 Phase 2 Phase 3
Segment
Response
Analysis &
Planning
Exploratory
Study
Implement
Marketing
Program
117. 118
4. Define Market Segments
• Assume the following data matrix includes the responses
from six customers on four key components of value (10
point rating scale of importance)
• How would you segment this “market”?
Customer
Importance
of Durability
Importance
of Service
Importance
of Ease of
Use
Importance
of Price
Cluster
Assignment
1 9 4 5 6
2 3 5 10 5
3 4 5 5 10
4 10 6 7 6
5 5 5 7 10
6 6 3 9 5
“Market” 6.2 4.7 6.3 7.0 X
118. 119
5. Describe Market Segments
• Suppose the market for 300 users of industrial adhesives
were represented by the table below
• What can we conclude from this table?
Organization
size
Price
sensitive
Durability
sensitive
Service
sensitive
Large 30 20 50
Medium 50 20 30
Small 20 60 20
Segments based
on needs
Organization size
as a descriptor
122. 123
Cluster Analysis Issues
• Defining a measure of similarity (or distance) between
segments.
• Identifying “outliers.”
• Selecting a clustering procedure
• Hierarchical clustering (e.g., Single linkage, average
linkage, and minimum variance methods)
• Partitioning methods (e.g., K-Means)
• Cluster profiling
• Univariate analysis
• Multiple discriminant analysis
123. 124
Interpreting Cluster Analysis
Results
• Select the appropriate number of clusters:
• Are the bases variables highly correlated? (Should we
reduce the data through factor analysis before
clustering?)
• Are the clusters separated well from each other?
• Should we combine or separate the clusters?
• Can you come up with descriptive names for each
cluster (eg, professionals, techno-savvy, etc.)?
• Segment the market independently of your ability to
reach the segments (i.e., separately evaluate
segmentation and discriminant analysis results).
124. Use of Dendograms in segmentation
• Dendograms provide graphical representations of the loss of information generated by grouping
different clusters (or customers) together. The dendogram is generated only if you choose the
Hierarchical Clustering
• At one extreme (upper part of the dendogram), all customers group into one cluster, and the loss
of information is maximum, because they all receive undifferentiated treatment, regardless of
their characteristics.
• At the other extreme (lower part of the dendogram), customers appear in separate, small
clusters, and only those customers very similar to one another group together (“similar” or
“close” in this context refers to the distance between two customers in terms of the segmentation
variables).
• When reviewing a dendogram, look for significant distances or “jumps” in the distances. Grouping
these three clusters into two generates a significant loss of information; in other words, it results
in grouping within the same cluster customers who are very dissimilar.
125
125. 126
Profiling Clusters
Two Cluster Solution for PC Data: Need-Based Variables
size power office
use
LAN storage
needs
color periph.
wide
connect.
budget
–1
1
0
Business
Design
Means of
Variables
126. 127
Managerial Uses of
Segmentation Analysis
• Select attractive segments for focused effort
• Develop a marketing plan (4P’s and positioning)
to target selected segments.
• In consumer markets, we typically rely on
advertising and channel members to selectively
reach targeted segments.
• In business markets, we use sales force and direct
marketing. You can use the results from the
discriminant analysis to assign new customers to
one of the segments.
127. 128
Checklist for Segmentation
1. Is it values, needs, or choice-based? Whose values and needs?
2. Is it a projectable sample?
3. Is the study valid? (Does it use multiple methods and multiple
measures)
4. Are the segments stable?
5. Does the study answer important marketing questions (product
design, positioning, channel selection, sales force strategy, sales
forecasting)
6. Are segmentation results linked to databases?
7. Is this a one-time study or is it a part of a long-term program?
128. 129
Segmentation Summary
In summary,
• Use needs variables to segment markets.
• Select segments taking into account both the
attractiveness of segments and the strengths of the
firm.
• Use descriptor variables to develop a marketing plan
to reach and serve chosen segments.
• Develop mechanisms to implement the segmentation
strategy on a routine basis (one way is through
information technology).
129. 130
Which Segments to Serve?
—Segment Attractiveness
Criteria
Criterion Examples of Considerations
I. Size and Growth
1. Size • Market potential, current market penetration
2. Growth • Past growth forecasts of technology change
II. Structural Characteristics
3. Competition • Barriers to entry, barriers to exit, position of
competitors, ability to retaliate
4. Segment saturation • Gaps in the market
5. Protectability • Patentability of products, barriers to entry
6. Environmental risk • Economic, political, and technological change
III. Product-Market Fit
7. Fit • Coherence with company’s strengths and image
8. Relationships with • Synergy, cost interactions, image transfers,
segments cannibalization
9. Profitability • Entry costs, margin levels, return on investment
130. Positioning recognises that
• Attention span of the customer is small
• Customers way or highway
• you cannot be everything to everyone
Perceptual Maps for Positioning
• The need for positioning maps – the value of visualisation
• Understanding customers perceptions
• Building and understanding positioning maps
human brain is not very good at visualising numbers, hence
it helps to have a method of graphically represent complex
patterns
131. Positioning: Some Key Concepts
• Differentiation: Creation of differences on key dimensions between a
product and its main competitors.
• Positioning: Strategies to ensure that the main differences between
the focal product and its competitors occupy a distinct position in the
minds of customers.
• Mapping: Techniques (using customer-data) that enable managers to
develop differentiation and positioning strategies by helping them to
visualize the competitive structure of their markets as perceived by
their customers.
133. Perceptual Map of Beer Market
(This slide shows only the attributes)
Popular
with Men
Heavy
Special
Occasions
Dining Out Premium
Popular
with
Women
Light
Pale Color
On a
Budget
Good Value
Blue Collar
Full Bodied
Premium
Budget
Light
Heavy
Less Filling
134. Perceptual Map of Beer Market
(This slide includes both products and attributes)
Popular
with Men
Heavy
Special
Occasions
Dining Out Premium
Popular
with
Women
Light
Pale Color
On a
Budget
Good Value
Blue Collar
Full Bodied
Premium
Budget
Light
Heavy
•
Meister Brau
Stroh’s
•
•
•
Beck’s
• Heineken
Old Milwaukee
•
Miller •
Coors
•
Michelob
•
Miller
Lite
• Coors
Light
•
Old
Milwaukee Light
•
Budweiser
Less Filling
135. C1
C2
• Generate a matrix of inputs for the analysis consisting of each customer’s (C1,
C2,...) ratings of each brand on each of the attributes (A1, A2, A3,....)
• Compute average ratings of each car on each attribute. Submit data to a
suitable perceptual mapping technique (e.g., Factor Analysis).
• Interpret the underlying key dimensions of the map using the directions of the
individual attributes.
• Articulate the implications of how customers’ view the competing products and
concepts.
A1 A2 A3 A4............... A15
Audi 90 6 3 7 2 2
Toyota Supra 4 3 4 1 5
New G20 3 6 2 7 7
..
Audi 90
Toyota Supra
New G20
Perceptual Maps Using
Attribute Ratings
136. Perceptual Maps Using
Attribute Ratings
Select a set of cars which are of interest to the target group of customers (including the new
product/concept of interest).
Identify a set of key attributes on which these cars are evaluated by the target group (e.g.,
through focus groups).
Ensure that customers are familiar with all the products of interest (e.g., through video
presentation).
Have customers evaluate each car on the chosen set of attributes.
Example: Positioning of a new car concept
Unattractive ...........................… Attractive (A1)
Quiet ..............…............. Noisy (A2)
Unreliable …........................... Very reliable (A3)
Uninteresting …………………... Interesting (A4)
Low prestige ...........…............… High prestige (A5)
.
..
Definitely would
not buy ......................…..... Definitely would buy (Preference)
137. Mapping Techniques
• Mapping perceptions
• Attribute-ratings methods (particularly useful for
functional products)
• Overall-similarity methods (particularly useful for
image-oriented products)
• Mapping preferences
• Include an overall preference vector in a perceptual
map
• “External” analysis to fit preferences of individuals on
a common perceptual map
138. Mapping Methods in Marketing
Perceptual Maps Preference Maps
Joint Space Maps
(includes both
perception &
preference)
Similarity-based
methods
Attribute-based
methods
Ideal-point model
(unfolding model)
Vector model
External analysis
using PREFMAP-3
Simple “joint space
maps” using
modified perceptual
mapping methods
Italicized items are included in the text/software (Coming soon: Ideal Point Model)
139. Guidelines for Interpreting Perceptual Maps
• The arrow indicates the direction in which that attribute is increasing
(The attribute is decreasing in the direction opposite to the arrow).
Thus, cars positioned farther in the West direction offer more
prestigious and those positioned in the East direction are less
prestigious.
• The length of the line from the origin to the arrow indicates the
variance of that attribute explained by the 2D map. The longer this
line, the greater is the importance of that attribute in helping you to
interpret the map.
• Attributes that are both relatively important and close to the
horizontal (vertical) axis help determine the meaning of the axis.
• To position a car on each attribute, draw an imaginary perpendicular
line from the location of that car onto that attribute. (These are
shown by dashed lines on the map).
140. Mapping Preferences
Objective—Introduce customer preferences into perceptual maps:
• A simple ideal point method: Introduce an “ideal” brand as an
additional stimulus evaluated by customers.
• A simple vector method: Introduce “preferences” as an
additional variable in the attribute ratings data
141. Two Preference Models
Attribute Attribute
Preference Preference
Ideal-Point Preference Model Vector Preference Model
Ideal Point
Increasing
Preference
Decreasing
Preference
(eg, sweetness) (eg, service speed)
142. Interpreting Preference Maps
(a)
A is preferred twice as much as B.
(dIB = 2dIA)
Ideal-Point Map
Ideal
Point (I)
A
B
dIB
dIA
Vector Map
(b)
A is preferred to B and B is preferred to C.
With reference to A, C is preferred half as
much as B.
(dAC = 2dAB)
Preference
Vector
A
C
dAB
dAC
B
143. Uses of Ideal Point Maps
• The ideal point map is useful to understand gaps in the market for
future launches or brand extension exercises.
• It also explains the relative position of current brands in the market
from an ideal brand from the consumers perspective.
• It provides instructions to managers for inclusion of attributes and
association in future brand equity building exercises.
144
144. Limitations
• Provides a static model - ignores dynamics of
customer perceptions.
• Interpretation is sometimes difficult.
• Does not incorporate cost or likelihood of
being able to achieve a desired positioning.
• Does not incorporate a “probability model” to
indicate goodness of a map.
• Generally, need about 6 to 8 products in a
category to make the technique useful.
145. Some Uses of Mapping Techniques
• Check how customer perceptions of client products
compare to perceptions of competitors.
• Identify product strengths and weaknesses.
• Select competitors to compete against.
• Determine how much change is needed on key
product attributes to move products to more
favorable positions.
• Visually determine impact of communications
programs on market perceptions.
146. Strengths and Weakness of Perceptual Maps
Strengths
1. The ability to get a visual snap-shot of brand competition as
perceived by customers.
2. The ability to “name” dimensions of the map based on the
relevant attributes that “load” on the dimensions.
3. The ability to identify how each brand is perceived on each
attribute, and on each dimension.
4. The ability to identify what re-positioning strategies are
practical, and what are not.
5. The ability to engage multiple decision makers/stakeholders in
a common view of the marketplace as seen from the
customers’ perspective
Weaknesses
1. The need to identify all relevant attributes, and all relevant
competitors.
2. The need for data collection from relevant customers. In other
words, the customers must belong to the same segment; or
else we could have the possibility of producing an “average”
ideal positioning, when there are really multiple ideal
positionings.
3. The need for appropriate sample size.
4. Maps are a static representation of the marketplace.
5. Maps do not explain why customers perceive brands the way
they do.
6. Maps address perceptions (and preferences as well here, in
terms of the ideal brand) but no not directly reflect likely
customer choice