Download as pdf or txt
Download as pdf or txt
You are on page 1of 84

Marketing Business Analytics Session1 Prof.

Manisha Kapoor

INTRODUCTION TO MARKETING BUSINESS ANALYTICS


In this session we will briefly cover the analytics techniques that we will study in this course. We will
also understand how business firms can benefit from the insights using these marketing analytics
techniques.

Learning Outcomes: Marketing Business Analytics


• To equip students with an understanding of the “importance and role of marketing
analytics” in modern business enterprises and how business firms can take advantage
of marketing analytics.

• Further for students who wish to specialize in analytics, the course provides a strong
foundation in the application of marketing analytics with analytical platforms.

1. Customer Analytics: Understanding Customers


Understanding customer behaviour is a part of customer analytics. We can understand
consumers through 3 ways
• Modelling perceptions: How consumer perceive my brand?
• Modelling choices: How consumer decide which brand to purchase or not?
• Segmenting consumers: How consumers differ in their wants/needs and preferences?

1.1 Modelling Consumer perceptions: Perceptual Mapping using PCA


A perceptual map enables the marketing analyst to determine how a product perceived
compared to the competition so that they can (re) How I am positioned vis a vis competition?
position them appropriately.
What are my strengths and weaknesses?

When a perceptual map of products or brands is


Who all are my close competitors?
available, the marketing analyst can use an
individual’s ranking of products to find the
Am I differentiated ?
individual’s ideal point or most-wanted location
on perceptual map Is there any current gap in the market ? – new product &
existing product

Benefits of Perceptual Mapping: What is my ideal point?

Perceptual Mapping can help us to answer If I want to shift myself to my ideal point/market leader
what strategy /changes in positioning messages I need
following questions to make ?

1.2 Modelling Choices: using Logistic Regression


Modelling consumer choices can help us to predict whether a customer will buy my brand or
not. Using the past data of consumer behavior and with the help of analytics we can model
choices and predict whether a customer will buy a brand or not.

• How do consumers choose which brand to buy from a choice set?


• What determines whether a consumer renews a contract or churns?
Benefits of Modelling Choices:
• Purchase (Buy/not buy) → Who will buy and not buy?
• Response to marketing efforts → buy after being sent a coupon or an email ad?
• Churn → Predict churn before it happens
Marketing Business Analytics Session1 Prof. Manisha Kapoor

1.3 Segmenting consumers: using Cluster Analysis


Market segmentation is breaking the market into distinct subsets of customers

• What is the need of segmentation? What if we do not segment? One size fit all? Benefits
of segmentation
• Different ways of segmentation?
Benefits of Segmentation:
• Identify right customers, whom to target?
• Segment profiles (segmenting and targeting variables) – how to target?

2. Customer Relationship Management: CLV and RFM

2.1 Customer Lifetime Value


CLV is a part of customer relationship management. Basically, the concept is to Manage
customers as investments and not looking at a customer as one-time transaction. CLV is the net
present value of all future streams of profits that is generated by a customer over the duration
of the relationship with the firm.
What is the customer lifetime value to the firm? How much firm should spend to acquire a new
customer? If the CLV of customer is x, the firm would not want to spend x+1 money.
How much firm should spend to increase retention rate to 70%?
Customer Expansion – cross selling, bundling, recommendation

Benefits of Calculating Customer Lifetime Value:


• Optimize acquisition cost (more customers and reduce acquisition cost)
• Increase profitability of customer (increase share of wallet and buys more)
• Increase retention rate

2.2 Recency, Frequency and Monetary Value


RFM analysis starts by basically asking three questions about each customers.
• How recently they have shopped with you?
• How often do they make a purchase?
• How much have they spent?
Through RFM analysis, you can rank your customers into categories to find your most loyal
customers, those who are more at risk, or, for example, those who haven’t shopped recently.

Benefits of Calculating RFM Value:


RFM analysis is used to organize customer databases into specific segments. Basing
these segments off customer’s purchase behaviour gives crucial insight into the type of
marketing you should use to target them and to move them between segments.

3. Retail Analytics: Market Basket Analysis


Market Basket analysis is an important concept of Retail Analytics. Imagine you go to super
market to buy Bread. What are the other items that are likely to land in your basket other than
bread? Milk, eggs, Jam, butter. What if we could mine this huge data to understand what are
the products that goes together.
Marketing Business Analytics Session1 Prof. Manisha Kapoor

Benefits of Market Basket Analysis:


• Market Basket Analysis is one of the key techniques used by large retailers to uncover
associations between items
• Market Basket can be used in deciding the location and promotion of goods inside a
store

4. Advertising Analytics: Using Regression


Companies spend money on advertising to impact sales. By modelling the sales response
function, we can compute the causal impact of advertising.

Benefits of Advertising Analytics:


• Decide the impact of advertisement on sales

5. Sales Promotion: Using Regression


Sales promotions are short term inducements designed to have a direct impact on the buying
behaviour of end users and trade. We can model incremental sales to assess it’s impact.

Benefits of Sales Promotions:


• Assess the profitability of sales promotions

6. Pricing Analytics
In Pricing we start by understanding people’s willingness to pay, then we translate it into
demand curves and then finally set optimal price from the demand curve. So we will fit
distributions to willingness to pay data and we will look at how to obtain a demand curve
from it and then ultimately we will look at some examples of optimization of prices.

Benefits of Sales Promotions:


• Modelling Willingness to Pay
• Modelling Demand Curves
• Optimal Pricing

7. Social-Media Analytics
Social media is gradually being recognized as a significant resource for understanding customer
sentiments, the reputation of products, willingness to buy, and customer satisfaction with
products and/or services.
Social media analytics in hospitality and tourism/film industry/mobile companies
Social Media Analytics
• Text mining & Sentiment Analysis
Marketing Business Analytics Session2 Prof. Manisha Kapoor

R TUTORIAL
In this session we will cover the basics of working in R with emphasis on working with data frames. We
will also learn the basics of R Markdown and how to create HTML reports. Please note in this course
we will do all the analysis in RMD file.

# **1. Getting started: R Markdown essentials**


## **1.1 Create R Markdown File** - New file → RMD file
### **1.1.1 Code Blocks** - We use 3 back texts ``` to create code blocks.
### **1.1.2 Knit** - To knit in HTML/PDF format
### **1.1.3 Page Break** - Three or more asterisks or dashes: ****** or ------
This is a block -

```{r}

```

# **2. Basics of working with R Commands**


##**2.1 Object and assignment operator**
##**2.2 Running the code in R**
##**2.3 R is case sensitive**
##**2.4 Use of # to write comments**
##**2.5 For help put '?' before function**
##**2.6 Variable and data in environment**
##**2.7 Understand the layout**
##**2.8 In R spaces doesn't matter. It is just for aesthetics**

Here x is object to which we have assigned value 2+4. The symbol '<-' is used to assign value to an
object. Run this block using Ctrl+Enter in windows and command + Enter in Mac. In order to print this
value type x. Check the environment when you run this code. Also, try typing X.
```{r}
x<- 2+4 #sum
x
```

2.7 Understand the layout


Top left corner - text editor - you type all your code here and can then run selected code using Cmd +
Enter (Mac) or Ctrl + Enter (Win).
Bottom left corner - console - this is where the code run and where output (including error messages)
is shown.
Top right corner - list of defined variables and loaded data.
Marketing Business Analytics Session2 Prof. Manisha Kapoor

Bottom right corner - variety of windows. Plots will appear here under “Plots” tab. You can install
packages and view already installed ones under the “Packages” tab. And you can type in names of
functions under “Help” tab to find explanation of their operation*

# **3. How to read csv file?**


Here we are reading it directly from the website. In case you have csv file, you can select it by setting
working directory folder. After reading the data set into R, we store it in a data frame named seg.df.
We assign object using assignment operator "<-"

```{r}
read.csv("http://goo.gl/qw303p")
seg.df <- read.csv("http://goo.gl/qw303p")
seg.df
```

1. Setting working directory setwd()


2. Getting working directory getwd()
2. Use file choose to pick file manually from folder and read

# **4. Data Exploration**


##**4.1 Data Exploration Functions**
### **4.1.1 Head Function: head()**
### **4.1.2 Column Names: colnames()**
### **4.1.3 Dimension: dim()**
### **4.1.4 Structure: str()**
### **4.1.5 Tail Function: tail()**

```{r}
head(seg.df) #Returns first 6 rows of your data
colnames(seg.df) #Returns column names
dim(seg.df) #Number of rows and column i.e. no of observations & variables
str(seg.df) #Structure of different variables
tail(seg.df) ##Returns last 6 rows of your data
```

seg.df data has 300 rows and 7 columns. Gender, own home, subscribe and segment are categorical
variables.

### **4.1.6 Summary: summary()**

```{r}
summary(seg.df)
```
Marketing Business Analytics Session2 Prof. Manisha Kapoor

The summary function generates a full summary of the data set. For categorical variable it tells the
frequency. The seg data set has 7 variables 143 Males and 157 Females.

**Metric variables**
Age, income, kids and segment are metric variables.
The summary function gives 5 value summary, mean and missing values if any. Missing values are
coded as NAs in R. There are no missing values.

*As can be seen from this summary, the segmentation data set contains answers from 157 women and
143 men. The age of the respondents is a metric variable summarised by the minimum value (Min.),
the first quartile (1st Qu.),the median, the mean, the third quartile (3rd Qu.), and the maximum (Max.).
The youngest respondent is 19, and the oldest 80 years old. Half of the respondents are between 33 and
47 years old.

**Categorical variables**
The summary function shows the frequency of categorical variables i.e.how frequently they occur.
There are three categorical variables i.e. Gender, own home, subscribe and segments. These categorical
variables are saved as factors. (Check in environment)

# **4.2 Select columns from Data frame**


## **4.2.1 Use $ to pull specific column**
## **4.2.2 Use : operator**
## **4.2.3 Use [] indexation**
```{r}
summary(seg.df)
summary(seg.df$gender) #Use $ to pull specific column
seg.df[,1:4] #Indexation show first 4 columns
seg.df[1:3,1:3]#Show three rows and three columns
```

**Exercise 1 : Data Exploration**


A. [Store_Dataset] (http://goo.gl/QPDdMl)
1. Read this data in CSV format
2. Store this in a data frame named store.df
3. Run data exploration functions
4. How many rows and columns are in store.df ? Which function will you use for this?
5. Tell names of all the columns ? Which function will you use for this?
6. Can you tell the names of categorical variables? What is the other name of categorical variables ?
Which function will you use for this?
7. Can you tell the names of continuous variables ? Which function will you use for this?
8. What is the maximum value of week ?
9. The summary function gives 5 value summary for ........ variable
10. The summary function gives count/frequency of ........ variable
11. Use summary function to tell summary of p2sales. Which symbol will you use to pull p2sales
column
12. Print second, third and fourth column of store.df
13. Print three rows and three columns of store.df
Marketing Business Analytics Session2 Prof. Manisha Kapoor

# **4.3 Summarising Dataframe**


## **4.3.1 Aggregate Function using tilde (~)operator**

R provides a standard way to describe relationships among variables through formula specification. A
formula uses the tilde (∼) operator to separate response variables on the left from explanatory variables
on the right. The basic form is: y∼x

This is used in many contexts in R, where the meaning of response and explanatory depend on the
situation. For example, in linear regression, the simple formula above would model y as a linear function
of x. In the case of the aggregate() command, the effect is to aggregate y according to the levels of x.

Let’s see that in practice.

```{r}
seg.df <- read.csv("http://goo.gl/qw303p")
seg.df
aggregate(income ∼ Segment, data=seg.df, mean)
```

The general form is aggregate(formula, data, FUN).In our example,we tell R to “take income by
Segment within the data set seg.df, and apply mean to each group.”

## **4.3.2 Descriptives for two way group**


A common task in marketing is cross-tabulating, separating customers into groups according to
two(or more)factors.Formula syntax makes it easy to compute across tab just by specifying multiple
explanatory variables:
y∼x1+x2+···
Using this format with aggregate(), we write:

```{r}
aggregate(income ∼ Segment + ownHome + subscribe, data=seg.df, mean)
```

**Exercise 2 : Data Summarisation**


A. [Store_Dataset] (http://goo.gl/QPDdMl)
14. Summarize p1sales by country using aggregate function of store.df (by mean and bysum)

B. [ausvacmot_Dataset](https://homepage.boku.ac.at/leisch/MSA/datasets/vacation.csv)
15. Read data & name it ausvacmot.df.
16. Summarize Education by Gender and Occupation. Which cohort scores highest on education? and
which one scores lowest?
17. Summarize Age by Gender and State. Which cohort is the oldest ? What is the average age of this
cohort?
Marketing Business Analytics Session2 Prof. Manisha Kapoor

# **5. Creating Crosstabs for categorical variables**

We can check associations between two categorical variables and later check Chi Square test for
significance.

```{r}
table(Gender = seg.df$gender, seg.df$ownHome)
table_A <-table(Gender = seg.df$gender, seg.df$ownHome)
chisq.test (table_A)
```
No significant difference between men and women in terms of home ownership as p value is greater
than 0.05.

**Exercise 3 : Cross Tabulation**


B. [ausvacmot_Dataset](https://homepage.boku.ac.at/leisch/MSA/datasets/vacation.csv)
18. Read data in csv and name it as ausvacmot_data
19. Run crosstab between Gender and Occupation. Is there significant difference between men and
women in terms of Occupation ?
20. Run crosstab between Gender and Income2. Is there significant difference between men and women
in terms of Occupation ?
21. Run crosstab between Occupation and Income2. Is there significant difference between income in
terms of Occupation ?
22. Does travel behaviour of rest and relax differ by Gender?
23. Does travel behaviour of rest and relax differ by Occupation?

#**6. Visualization methods of different types of variables**

In R, we obtain a numeric summary of the data with command summary(). This command returns the
range, the quartiles, and the mean for numeric variables.

For categorical variables, the command returns frequency counts. The command also returns the
number of missing values for each variable.

Helpful graphical methods for **numeric data** are histograms, boxplots and scatter plots.

Bar plots of frequency counts are useful for the visualisation of **categorical variables**.

## **6.1 Histograms : Create Visualization for single variables**

Histograms visualise the distribution of numeric variables. They show how often observations within a
certain value range occur. Histograms reveal if the distribution of a variable is unimodal and symmetric
or skewed. To obtain a histogram, we first need to create categories of values. We call this **binning**.
The bins must cover the entire range of observations, and must be adjacent to one another. Usually, they
are of equal length. Once we have created the bins, we plot how many of the observations fall into each
bin using one bar for each bin.
We plot the bin range on the x-axis, and the frequency of observations in each bin on the y-axis.A
number of R packages can construct histograms. We use package lattice because it enables us to create
histograms by segments.
Marketing Business Analytics Session2 Prof. Manisha Kapoor

We can construct a histogram for variable AGE using "lattice":


First install packages and then call library.

```{r}
install.packages("lattice")
library("lattice")
```

```{r}
histogram(~ age, data = seg.df)
```

By default, this command automatically creates bins. We can gain a deeper understanding of the data
by inspecting histograms for different bin widths by specifying the number of bins using the argument
breaks:

```{r}
histogram(~ age, data = seg.df, breaks = 30,type = "density")
```

This command leads to finer bins, as shown in the plot above. The finer bins are more informative,
revealing that the distribution is bi-modal with many respondents aged around 35–50 and around 20-30
years.

Argument type = "density" rescales the y-axis to display density estimates. The sum of the areas of all
bars in this plot ads up to 1. Plotting density estimates allows us to superimpose probability density
functions of parametric distributions.

This scaling is in general viewed as the default representation for a histogram.

## **6.2 Barplots : Create Visualization for categorical variables**

```{r}
barplot(table(seg.df$ownHome))
barplot(table(seg.df$ownHome), ylab ='Frequency', main = 'Barplot of Home ownership') #with labeled axis
fem_ownHome <- table (seg.df$ownHome[seg.df$gender=='Female']) #filtered by gender
fem_ownHome
barplot(fem_ownHome,ylab ='Frequency', main = 'Barplot of Female Home ownership')
```

**Exercise 4 : Visualisation**
24. Create Histograms of p1sales and p2sales using 20 number of bins of store.df dataset. Which product
has better sales? Why ?
25. Create barplot of State by Females
26. Create barplot of State by Income2
Marketing Business Analytics Session3 Prof. Manisha Kapoor

MARKET SEGMENTATION BASICS


In this session we will cover the basics of Market Segmentation. Why we need to segment and what if
we do not segment.

What is market segmentation?


• Breaking your market down into different subgroups
• Market segmentation - subdividing market into distinct subsets of customers.
• Segment members are different between segments but similar within.
• Segmentation is a method of trying to understand your customer by dividing into
categories or homogeneous segments
• Here homogeneous segments means that they have common sets of needs, behavior and
attitude
Successful companies treat different customers differently and in business what we call this is
Segmentation.
So, the underlying assumption is that these homogeneous segments will have -

• Comparable purchase behaviour


• Similar response to different marketing activities

What is a market?
• Market is the people that you realistically think will be attracted in your product.
• Market is your prospective customer base
What is the market for?

Automobile Kids wear Tea brand Pet supplies ex. Brand that sells
brand brand Dog Food Meat

People over 18 Mothers of People who Dog owners Non-


yrs. as they will children aged drink Tea Vegetarians
have driving 6 months to 12
license years

Need for segmentation?


Do you think we can sell same product to all the people in a market? Can we sell one type of tea to all
the people who drink tea? Can we sell one type of car to all the people who want to buy a car?

No, we can’t because people in a market do not share universal wants and needs. THEY HAVE
DIFFERENT WANTS, NEEDS, TASTES AND PREFERENCES
Marketing Business Analytics Session3 Prof. Manisha Kapoor

Risk of going with one size fits all approach


What if I say I do not want to go with segmentation, and I decide to go with one generic product? What
is the problem with that?
Example 1 – Lukewarm Tea
If you do one size fits all – it’s dangerous i.e. if you don’t segment and try to sell one product to different
customers, it’s dangerous.
Let me illustrate this – I am selling tea. Some of you love hot tea and some of you love cold tea. So I
just want to come out with one tea and I take the average and what I end up coming out with? Lukewarm
tea.
And who wants to buy my lukewarm tea? Nobody right? That’s dangerous when you have one offer
and you are providing it to two very different groups. you end up providing nothing.
Example 2 – Laptop company
Suppose there is one laptop company which has 4 options of screen size, colour, battery life, weight
and RAM.

Option1 Option 2 Option 3 Option 4


Screen Size 11 inches 12 inches 13 inches 14 inches
Colour Black Silver Burgundy Blue
Battery Life 10 hours 12 hours 15 hours 20 hours
Weight 0.9 kg 1 kg 1.4 kg 2 kg
RAM 2GB 4GB 8GB 16GB

Whether they should offer one product or all the options? Which option they should select?

Go with Go with only


1024 options one laptop

The mid-way!
The compromise or mid-way would be to collect data from customer about their preferences and try to
cluster that data or those preferences and try to group them into homogeneous segment. As that
people within a certain segment would want more or less the
same product features and that product would be different from 2 SEGMENT

the preferences of another segment. In this case the company SEGMENT 1


would not produce all products/laptops but will produce a
product line of Laptops/devices that’s appealing to the
different market segment.

Different product lines


SEGMENT 3

• Notebooks
• Tablets
• Laptops
• Desktops
Marketing Business Analytics Session3 Prof. Manisha Kapoor

Basis of segmentation
What is the approach of creating these segments? On what basis can we segment the market?

The key question to ask while deciding the basis of segmentation


Why is this service being bought?”
or
to put this in a marketing context,
“What benefits are sought?”
Marketing Business Analytics Session3 Prof. Manisha Kapoor

“Customers buy Benefits not Products”


Quote from Theodore Levitt ; “If you’re not thinking segments, you are not thinking marketing”
• Companies that concentrate on products rather than benefits can be defined as suffering from
“Marketing Myopia”. In his definitive work Theodore Levitt summarized the need for a
benefit led approach.
• Benefits based segmentation approaches the question from the perspective; If we find what
benefits people seek, we can then group them accordingly.
• Then a method must be found to identify the external characteristics that can be used to define
the members of the group.

Benefits of segmentation
By understanding people’s preferences, by segmenting them into homogeneous segments, you are
customizing the product to their taste, you are personalizing experiences of the customer if it is a service.
If the product or the service comes close to what people want, then they would be more satisfied.
Satisfied customers are loyal and they come back, and they can also refer your products to others. So,
what you gain out of this is that when you appeal to people preferences, you get customer loyalty and
retention. From the firm the fact that you're segmenting the market, some markets are appealing, some
segments may not be appealing to you, it is your decision to decide on which segment to target. So, in
that case, it's an exercise that can help you identify valuable customers. And now that because you know
what your people want and need, then you can have more targeted communication. You have more
targeted promotions product people. At the end, what you get is what we call higher customer life-time
value,
Benefits of segmentation -
Marketing Business Analytics Session4 Prof. Manisha Kapoor

MARKET SEGMENTATION: SEGMENTATION METHODS


In this session we will cover the two methods of clustering i.e. Hierarchical and K-means
We know that market segmentation is

What is market segmentation?


• Breaking your market down into different subgroups
• Market segmentation - subdividing market into distinct subsets of customers.
• Segment members are different between segments but similar within.
• Segmentation is a method of trying to understand your customer by dividing into
categories or homogeneous segments
• Here homogeneous segments means that they have common sets of needs, behavior and
attitude
So, if you translate it to the goal of clustering then we can say that our goal of clustering is to club
similar respondent together. Now, the question arises how system knows which respondents are
similar or dissimilar. It does this based on distance.
So, let us check this by example
Here we have 6 respondents and their Income and education is mentioned against them.

ID Income Education Euclidean Distance:


(in lakhs) (in yrs)

1 20 18
2 20 16 Concept of clustering is related to distance. The Euclidean
distance between person 1 and 2 is 2.
3 30 20
Similarly, you can calculate the distance between all the
4 25 15
respondents as given below.
5 25 12
The lesser is the distance between two cases/people, the
6 30 22 similar they are.

The Distance Matrix D


We can compute 15 pairwise distances. The
ID 1 2 3 4 5 6
distance is symmetric. Like distance between
Delhi and Gurgaon is same as the distance
1 0 2.0 10.2 5.8 7.8 10.8 between Gurgaon and Delhi.
2 0 10.8 5.1 6.4 11.6
3 0 7.1 9.4 2.0
4 0 3.0 8.6
5 0 11.2
6 0
Marketing Business Analytics Session4 Prof. Manisha Kapoor

Hierarchial clustering works on Euclidean distances or any form of distances. And we have to supply
it in form of matrix. It then identifies the closest people. So, in this case person 1 and 2 have shortest
distance. So, it will group one and two into one cluster and 3 and 6 in another cluster at a distance of
two. How many clusters do we have now – 4.
Okay now we have to update this distance matrix. And we have to select the linkage criteria
Step 2 : Update D using minimum distance rule and Minimum linkage criteria
Okay now we have to update this distance matrix. And we have to select the linkage criteria. So, we
have 1 and 2 as cluster and 3&6 as cluster. What is the distance between (1 and 2 ) and 3? It’s like if I
have to calculate distance between Delhi and Gurgaon
- whether I take distance from South Delhi (1) to new Gurgaon (3).
- Or take distance from North Delhi(2) to new Gurgaon (3).

We check this through minimum linkage criteria which we have selected in this example

Example :
D[1,2]3 = Min [10.2,10.8] = 10.2
D[1,2]6 = Min [10.8,11.6] = 10.8
D[1,2][3,6] = Min [10.2,10.8] = 10.2
D [1,2]4 = Min [5.8,5.1] = 5.1
D [1,2]5 = Min [7.8,6.4] = 6.4
D [1,2][4,5] = Min [5.1,6.4] = 5.1

Dendrogram
Marketing Business Analytics Session4 Prof. Manisha Kapoor

The hierarchical clustering method begins with each observation in its own cluster. It then successively
joins neighboring observations or clusters one at a time according to their distances from one another
and continues this until all observations are linked. This process of repeatedly joining observations and
groups is known as an agglomerative method.
From this dendrogram you can see which
cluster of respondents joined first and were
similar. Remember Our goal is to join similar
cluster together and try not to join dissimilar
cluster. We can do this by looking at the
joining distance so the cluster which join early
or smaller joining distance are more similar
but the clusters which joined at larger joining
distance are most dissimilar.
From this dendrogram can you tell how many
clusters we will get if we cut the tree at joining
distance of 15? You need to count the number
of branches it cuts by drawing a hypothetical
parallel line starting from joining distance of 15. You can see the dendrogram suggests 3 cluster
solution.
Marketing Business Analytics Session4 Prof. Manisha Kapoor

K- Means clustering
- K--means requires the specification of the number of clusters in advance, say S=3.
- The method aims to group the observations based on their similarity using an optimization
procedure. Indeed, the aim is to minimize the within-cluster variation which is defined as the
sum of square of the euclidean distance between each data point to the centroid of its cluster.
More precisely, the algorithm works as follow:
- K--Means algorithm:
- Start by randomly assigning each subject to a cluster, s=1,…,S
- Compute the centroid of each cluster and the distance of each subject to each of the clusters
centroids
- Reassign each subject to the cluster with closest centroid
- Repeat steps 2 and 3 until no further reassignment is possible (i.e., when the within--cluster
variance is minimized)
K-means is a way of taking a dataset and finding groups (or clusters) of points that have similar
properties. K-means works by grouping the points together in such a way that the distance between all
the points and the midpoint of the cluster they belong to is minimized.
Midpoint - Think of the simplest possible example. If you are asked to create 3 groups for the points
below and draw a star where the middle of each group would be, what would you do?

You will do something like --


Marketing Business Analytics Session4 Prof. Manisha Kapoor

Steps in K-Means clustering

Step 1 Randomly select 3 data points. These are initial clusters

Step 2 Measure the distance between the 1st point and 3 initial clusters

Step 3 Assign the 1st point to the nearest cluster

Step 4 Do the same thing for the next point

Step 4 Once all the points are in clusters we go to step 5

Step 5 Calculate the mean of each cluster

Measure and cluster using the mean values – Measure the distance of a
Step 6 point with the mean of each cluster

Calculate the center of each cluster and recluster and keep doing this
Step 7 iterations till the cluster assignment doesn’t change
Marketing Business Analytics Session5 Prof. Manisha Kapoor

MARKET SEGMENTATION: CASE STUDY SHOPPING


ATTITUDE
In this session we will cover segmentation of mall customers based on shopping attitude

Segmentation of Mall Customers


The “Business Decision”
• The management team of a large shopping mall would like to understand the types of people
who are, or could be, visiting their mall. They have good reasons to believe that there are a few
different market segments, and they are considering designing and positioning the shopping
mall services better in order to attract mainly a few profitable market segments, or to
differentiate their services (e.g. invitations to events, discounts, etc) across market segments.

Dataset : Consumers’ Attitude Towards Shopping


Based on past research, six attitudinal variables were identified. 160 consumers were asked to express
their degree of agreement with the following statements on a 7-point scale (1=disagree, 7= agree)
• V1 : Shopping is fun
• V2 : Shopping is bad for your budget
• V3 : I combine shopping with eating out
• V4 : I try to get the best buys when shopping
• V5 : I don’t care about shopping
• V6 : You can save a lot of money by comparing prices
• Age, Gender, Frequency of mall visits per month (Profiling variables)

A Marketing Application
The objective of segmentation is to find -

ARE THERE HOW MANY HOW ARE WHICH IS WHOM TO


DIFFERENT SEGMENTS? THEY THE MOST TARGET?
BENEFIT DIFFERENT? PROFITABLE?
SEGMENTS?

MARKETING
STRATEGY
Marketing Business Analytics Session5 Prof. Manisha Kapoor

Excerpt from the Dataset

Steps in Clustering

Data pre-processing

1. Missing values
2. Standardizing/Scaling the data

is.na(seg_data) #check whether there is any missing values


seg_data <- na.omit(seg_data) #replace missing values with na
Marketing Business Analytics Session5 Prof. Manisha Kapoor

Euclidean distance

hclust() output

Three clusters
Marketing Business Analytics Session5 Prof. Manisha Kapoor

The Three cluster Solution


Number of individuals in each cluster

Cluster renaming based on importance/mean value


Marketing Business Analytics Session5 Prof. Manisha Kapoor
Marketing Business Analytics Session5 Prof. Manisha Kapoor
Marketing Business Analytics Session5 Prof. Manisha Kapoor
Marketing Business Analytics Session5 Prof. Manisha Kapoor

Customer Profiling – Cluster 1


Marketing Business Analytics Session5 Prof. Manisha Kapoor

Customer Profiling – Cluster 2

Customer Profiling – Cluster 3


Marketing Business Analytics Session5 Prof. Manisha Kapoor
Marketing Business Analytics Session6 Prof. Manisha Kapoor

Perceptual Mapping through Principal Component Analysis


(Dimensionality Reduction)
In this session we will Model Consumer perceptions to determine how a business/brand position itself
in the market.

Perceptual Mapping
• A perceptual map enables the marketing analyst to determine how a product perceived
compared to the competition so that they can (re) position them appropriately.
• When a perceptual map of products or brands is available, the marketing analyst can use an
individual’s ranking of products to find the individual’s ideal point or most-wanted location
on perceptual map
• This provides a useful graph for marketers to understand the competition structure among
brands, products or industries.
• So for the brand rating data like the one we will use here, performing a perceptual map can let
us understand how each brand is comparatively positioned in peoples’ minds. It is also helpful
for developing a marketing strategy for differentiating from other competing brands.

How firms can benefit by reading perceptual maps?


These maps will help in answering questions like

How I am positioned vis a vis competition?

What are my strengths and weaknesses?

Who all are my close competitors?

Am I differentiated ?

Is there any current gap in the market ? – new product & existing product

What is my ideal point?

If I want to shift myself to my ideal point/market leader what strategy /changes in


positioning messages I need to make ?
Marketing Business Analytics Session6 Prof. Manisha Kapoor

Questions & Answers


Q1 What does the perceptual map display?
• Perceptual mapping / Market mapping is a diagrammatic technique used by marketers that
attempts to visually display the perceptions of customers or potential customers.
Q2 Why it’s important to understand customer perceptions? Can there be difference in positioning of a
brand by business and positioning in customer mind?
• The positioning of a brand is influenced by customer perceptions rather than by those of
businesses. For example, a business may feel it sells upmarket products of high quality, but if
customers view the products as low quality, it is their views which will influence sales.
Typically the position of a company's product, product line, or brand is displayed relative to
their competition.
Q3 What is the other name of perceptual map? How many dimensions does a perceptual map has?
• Perceptual maps, also known as market maps, usually have two dimensions but can be multi-
dimensional; they can be used to identify gaps in the market and potential partners or merger
targets as well as to clarify perceptual problems with a company's product. So, if a business
wants to find out where its brand is positioned in the market, it might carry out market research.
This will help them to find out how the customers sees their brand in relation to others in the
market

Dimensions & Benefit of dimensionality reduction


• Marketing data sets often have many variables – many dimensions – and it is advantageous
to reduce these to smaller sets of variables to consider.
• For example – we might have many items on a consumer survey that reflect a smaller number
of underlying concepts such as
➢ Customer satisfaction with a service
➢ Category leadership for a brand or
➢ Luxury for a product

• If we can reduce the data to it’s underlying dimensions , we can more clearly identify the
relationships among concepts
Marketing Business Analytics Session6 Prof. Manisha Kapoor

Reducing data complexity


3 common methods to reduce complexity by reducing the number of dimensions

PCA EFA MDS


attempts to find also attempts to capture Maps similarities among
uncorrelated linear variance with a small number observations in terms of a
dimensions that capture of dimensions while seeking low- dimension space such as
maximal variance in the data to make the dimensions a two-dimensional plot.
In marketing, PCA is often interpretable in terms of the MDS can work with metric
associated with perceptual original variables data and with non-metric
maps, which are data such as categorical or
visualizations of respondent ordinal data
associations among brands or
products

What is PCA?

An unsupervised learning PCA calculates a new projection PCA creates plots based
mathematical procedure that of the data set and the new axis on correlations among samples
uses an orthogonal based on the standard deviation or in other words a PCA plot
transformation to convert a set of variables converts the correlation among
of observations of possibly all of the sample into a 2 D (2
correlated variables into a set of dimensional) graph
linearly uncorrelated principal
components.
Marketing Business Analytics Session6 Prof. Manisha Kapoor

What is dimension?
• A dimension is underlying pattern or common variation across variables
• Gentle and Good for children tended to receive similar responses from respondents which
means they share common variance.
• There shared variance is 1 Dimension
• These dimensions are the backbone of perceptual mapping because they allow us to chart the
variables and brands relative to the dimensions and one another in a 2 Dimension map.
Marketing Business Analytics Session7 Prof. Manisha Kapoor

Perceptual Mapping through Principal Component Analysis


(Dimensionality Reduction)
In this session we will Model Consumer perceptions to determine how a business/brand position itself
in the market.
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session7 Prof. Manisha Kapoor
Marketing Business Analytics Session8 Prof. Manisha Kapoor

Perceptual Mapping through Principal Component Analysis


(Dimensionality Reduction)
In this session we will Model Consumer perceptions to determine how a business/brand position itself
in the market.

Case Study: Positioning strategy of Brand e using Perceptual Maps


Suppose you are a manager of brand e. What does the map tell you? How your brand is perceived?
Whether the brand is differentiated? What is the ideal point? What should you do about the position of
your brand e? what messages the brands need to emphasize?

3 things that brand e can do!


1. Do Nothing – do not differentiate
• For one thing, your brand is in the center and thus appears not to be well-differentiated on any
of the dimensions. That could be good or bad, depending on your strategic goals. If your goal
is to be a safe brand that appeals to many consumers, then a relatively undifferentiated position
like e could be desirable.
• On the other hand, if you wish your brand to have a strong, differentiated perception, this
finding would be unwanted (but important to know). What should you do about the position of
your brand e? Again, it depends on the strategic goals.
Marketing Business Analytics Session8 Prof. Manisha Kapoor

2. Increase differentiation, one possibility would be to take action to shift your brand in some
direction on the map.

Suppose you want to move in the direction of brand c. You could look
at the specific differences from c in the data:
```{r}

brand.mean["c", ] - brand.mean["e", ]

```

3. Not to follow another brand but aim for differentiated space where no brand is positioned
• Gap between f,g and b,c --→ value leader area
• How do we find out how to position there?

How do we find out how to position at value leader area?


• Let’s assume that the gap reflects approximately the average of those four brands (some risks
with this assumption). We can find that average using colMeans() on the brands’ rows, and
then take the difference of e from that average:
Marketing Business Analytics Session8 Prof. Manisha Kapoor

Summary
• To summarize, when you wish to compare several brands across many dimensions, it can be
helpful to focus on just the first two or three principal components that explain variation
in the data.
• You can select how many components to focus on using a scree plot, which shows how much
variation in the data is explained by each principal component.
• A perceptual map plots the brands on the first two principal components, revealing how the
observations relate to the underlying dimensions (the components).
• PCA may be performed using survey ratings of the brands (as we have done here) or with
objective data such as price and physical measurements, or with a combination of the two. In
any case, when you are confronted with multidimensional data on brands or products, PCA
visualization is a useful tool for understanding differences in the market.
• Product ratings, position of consumer segments, ratings of political candidates,
evaluations of advertisements, or any other area where you have metric data on
multiple dimensions that is aggregated for a modest number of discrete entities of
interest

Cautions with Perceptual Maps


• Choose level and type of aggregation carefully
• The relationships are strictly relative to product category and the brands and adjectives that
are tested
• Strength of a brand on a single adjective cannot be read directly from the chart – “Compared
to its position on other attributes, brand X is relatively differentiated by perceptions of
strength (or weakness) on such and such attribute”
Marketing Business Analytics Session9 Prof. Manisha Kapoor

REGRESSION BASICS
In this session we will cover the basics of Regression.

The concept – Supervised and unsupervised learning

To give an overview, ML
Linear Regression is not just a
models can be classified on the
machine learning algorithm, it
basis of the task performed
plays a huge role in statistics.
and the nature of the output:

Regression: Output is a
Regression & Classification fall
continuous variable.
under supervised
Classification: Output is a
learning while Clustering fall
categorical variable.
under unsupervised learning
Clustering: No notion of output.

Regression is a form of
predictive modeling technique
where we try to find a
significant relationship between
a dependent variable and one
or more independent variables.
There are various types of
regression techniques: Linear,
Logistic, Polynomial, Ridge,
Lasso, Softmax.

• In supervised learning you train the machine using data which is well labelled.
Unsupervised learning is a machine learning technique, where you do not need to
supervise the model. Supervised learning allows you to collect data or produce a data
output from the previous experience

What is Regression?
• When you think of regression, think prediction. A regression uses the historical
relationship between an independent and a dependent variable to predict the future
values of the dependent variable.
• Businesses use regression to predict such things as future sales, stock prices, currency
exchange rates

Q1What is independent and what is dependent variable?


Q2 What is historical relationship?
Q3 What do we predict – independent variable or dependent variable?
Marketing Business Analytics Session9 Prof. Manisha Kapoor

• A regression models the past relationship between variables to predict their future
behavior.
• As an example, imagine that your company wants to understand how past
advertising expenditures have related to sales to make future decisions about
advertising.
Note : Dependent variable is what you have to calculate/predict. Here we are trying to predict future
sales. The dependent variable in this instance is sales and the independent variable is advertising
expenditures.

• Q How much sales I can generate next year if I spend 30 lakhs rupees on
advertisement?
o You can find this by creating a regression equation of past year sales and
advertisement data.
o (i.e. historical relationship)
o You already have advertisement numbers (that the company is thinking of
spending i.e. 30 lakhs) which you can put in place of X and can
calculate/predict Y i.e. future sales….

How does regression work to enable prediction?


One of the first steps in regression is to plot your data on a
scatter plot. The following table lists the monthly sales and
advertising expenditures of last year by a digital electronics
company.

In this case, you would plot last year's data for monthly sales and advertising expenditures as shown on
the scatter plot below. (Data for independent and dependent variables must be from the same period of
time.)

1
Marketing Business Analytics Session9 Prof. Manisha Kapoor

Scatter plots are effective in visually identifying relationships between variables. These relationships
can be expressed mathematically in terms of a correlation coefficient, which is commonly referred to
as a correlation. Correlations are indices of the strength of the relationship between two variables. They
can be any value from –1 to +1.
When you use regression to predict future values of the dependent variable, the ideal correlation
between the independent and dependent variable is high—in absolute value terms, somewhere in the
range between .5 to .99. Viewing the scatter plot in the right, you can see that there appears to be some
degree of correlation between the level of advertising expenditure and product awareness. When
calculated, this correlation equals .89.
This historical data will enable you to predict the relationship between the two variables in the future,
before any further expense is incurred.
In order to make these predictions, a regression line must be drawn from the information appearing in
the scatter plot.

Regression Line
The figure below is the same as the scatter plot above, with the addition of a regression line fitted to the
historical data.

The regression line is the line with the smallest possible set of distances between itself and each data
point. As you can see, the regression line touches some data points, but not others. The distances of the
data points from the regression line are called error terms.
Marketing Business Analytics Session9 Prof. Manisha Kapoor

Regression Line: The best fit line

A regression line will always contain error terms because, in reality, independent variables are never
perfect predictors of the dependent variables. There are many uncontrollable factors in the business
world. The error term exists because a regression model can never include all possible variables; some
predictive capacity will always be absent, particularly in simple regression.
The typical procedure for finding the line of best fit is called the least-squares method. This calculation
is usually performed using computer software. In this calculation, the best fit is found by taking the
difference between each data point and the line, squaring each difference, and adding the values
together.
The least-squares method is based upon the principle that the sum of the squared errors should be made
as small as possible, so the regression line has the least error.

Hope you are able to understand now the error is the difference between the data points on predicted
line of regression and actual past data points of advertisements and sales.
So, how we arrive at future values…..
Regression Line : The best fit line and future predictions
Once this line is determined, it can be extended beyond the historical data to predict future levels of
product awareness, given a particular level of advertising expenditure.

The extension of the line of regression requires the assumption that the underlying process causing the
relationship between the two variables is valid beyond the range of the sample data. Regression is a
powerful business tool due to its ability to predict future relationships between variables such as these.
When you run a regression in R or SPSS, the program will provide you with a report.
Marketing Business Analytics Session9 Prof. Manisha Kapoor

We will now see the details of these reports, and the definition of all the terms included in the
report.
But, before that let’s check what is Sum of Square Residuals, Sum of Square Regression ….

What is SSR, SSE, SST ?


Let’s suppose we just have value of sales and with only one variable and no other information, if we
have to predict sales for next month, the best prediction for the next measurement would have been
the mean ?

Scenario 1 – when we have one variable i.e sales Scenario 2 – when we have two variables sales
and we are predicting sales of next month by and advertisement we can do much better
calculating mean prediction by using regression model.
Is it so?
So we compare the two scenarios to find
whether regression is doing good job or our old
method of taking mean of one variable is better.
How we do that?
We check SST, SSR AND SSE
Note : This comes in your ANOVA table. Try to
understand the relationship of all this with
Rsquare.
Marketing Business Analytics Session9 Prof. Manisha Kapoor

Coefficient of Determination (R Square)


Does the regression line fit the data better than the mean?
R Square → Goodness of fit
The total sum of square is partitioned or allocated to SSE and SSR
SST = SSE + SSR
R Square = SSR/SST
= Sum of square Regression/ Sum of Square Total
= Explained variance/Total variance
The coefficient of determination quantifies this ratio as percentage. The strength of the relationship is
measured by coefficient of determination.
So, R Square is the percentage of variance in the dependent variable explained by the independent
variable.

Regression and ANOVA


• What is important is knowing the degree to which our model is successful in explaining the
trend (variance) in dependent variable.
• With the help of ANOVA techniques, we can analyse a model performance
• Through our ANOVA test we would like to know the amount of variance explained by the
Independent variables in Dependent Variable VS the amount of variance that was left
unexplained.
• It is intuitive to see that larger the unexplained variance(trend) of the dependent variable smaller
will be the ratio and less effective is our regression model. On the other hand, if we have a
Marketing Business Analytics Session9 Prof. Manisha Kapoor

large explained variance then it is easy to see that our regression model was successful in
explaining the variance in the dependent variable and more effective is our model.
• The ratio of Explained Variance and Unexplained Variance is called F-Ratio

Adjusted R Square
• Adjusted R square value only accounts for significant predictors.
• The R square value will increase if we increase more number of independent variables even if
it doesn’t improve the model. R square cannot decrease as more independent variables are
added to the regression equation. Yet diminishing returns set in, so that after the first few
variables, the additional independent variables do not make much of a contribution
• Whereas the Adjusted R Square value will increase if we add significant predictors.
• So the Adjusted R Square value is adjusted for the sample size and independent variables (i.e.
number of predictors in the model)

N is number of samples
p is number of independent variable
Marketing Business Analytics Session10 Prof. Manisha Kapoor

REGRESSION CASE STUDY: SEC H ROAD TRIP


In this session we will analyse and interpret the outputs of Regression through one case study.

Setting the context of road trip


Imagine Sec H students are planning a road trip to Spiti. While everyone is in charge of the party
operations, Prateek is in charge of all the logistics involved. He has to plan every detail : the schedule,
when to stop and where, make sure that we all get there on time.
So, what’s the first thing that Prateek will do?
A. He will sneakily disappear from the face of the Earth and stop answering his friends calls,
because they’ll have fun while he’ll be the party police.
B. No, he will get himself a blank sheet of paper and start planning!

Budget !
• First item on his checklist ? Budget ! It’s a 18h – approximately 800 kms – fun ride, so a total
of 24h on the road.
• The follow up question : How much money should Prateek allocate for petrol?
• He approaches this problem with a science-oriented mindset, thinking that there must be a
way to estimate the amount of money needed, based on the distance he will be travelling.
• First, he looks at some data.

Scenario 1
Prateek was laboriously tracking his bus’s efficiency for the last year – so somewhere in his computer
he had this spreadsheet
Training Data

He ends up with a mathematical model that describes the relationship b/w miles driven and money
spent to fill the tank
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Scenario 2
Mitali and Kartikeya feels that Prateek needs to be more rigorous in his calculations and he should
also add oil price as an independent variable in this model
Training Data

He ends up with a mathematical model that describes the relationship b/w miles driven, oil price and
money spent to fill the tank
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Scenario 3
Now, Kunal suggest that Prateek should also add number of passengers as an independent variable in
this model
Training data

He ends up with a mathematical model that describes the relationship b/w miles driven, oil price,
number of passengers and money spent to fill the tank

Scenario 4
Someone now suggest that Prateek should also add Driver’s Mood as an independent variable in this
model
Marketing Business Analytics Session10 Prof. Manisha Kapoor

He ends up with a mathematical model that describes the relationship b/w miles driven, oil price,
number of passengers, Driver’s mood and money spent to fill the tank

Change of mind !
At this, point some of the students starts reconsidering the current destination option and comes up
with 8 options.

Prateek doesn’t want to say no to his friends but in a class of 88 students it’s now quite difficult for
Prateek to calculate the amount for each destination option manually.

So, he creates one training dataset…


Let the machine learn and work !
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Test data of Destination

Destination Kms_Driven Oil_Price


Shimla 340 75
Spiti 800 77
Rishikesh 240 78
Neemrana 125 76
Agra 217 75
Jim Corbett 225 77
Lansdowne 250 78
Italy 5918 77
How much amount Prateek needs to budget for each of these destinations ?
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Sessions 11 to 33
Dr Shirish Jeble,
ICFAI Business School, Pune

1 5. MARKET BASKET ANALYSIS (SESSION 11-13)

1.1 CONCEPT OF LIFT, MARKET BASKET ANALYSIS


Market basket is a concept that is applicable in retail industry. At a retail store, customers pickup
several items one by one and put it in their basket. After they are done shopping, they checkout at the
counter. When number of such transactions at point of sales (POS) are analysed, a common pattern
of items being bought together can be observed. A clear association or link between 2, 3 or more
products can be seen. For example, bread and butter, milk and toast, toothpaste and toothbrush, milk
and bananas, nail polish and lipstick etc. Many such combinations of 2, 3 or more products being
bought together can be observed. Most of these combos can be expected but some unusual
combinations emerge from this analysis. For example, beer and diapers, magazine and book mark
(without purchasing a book) etc.

The analytics technique used for identifying such affinity of products being bought together and using
it to increase sale of products is called as market basket analysis.

In this technique, first we collect POS transactions data from the retail stores. This data is used to
calculate a variable called as “Lift” which is an indicator of higher frequency of purchasing in the
transactions. Higher lift indicates, these 2 or 3 products are frequently purchased together.

Once we know the “Lift” of products, we can use this insight in various ways:

A. Placing of products next to each other so that a greater number of customers who purchase
one product are likely to pick up the second one
B. Though there is association between 2 products, one of them is may have lower sale. In such
cases, retailer can package or bundle them together to increase sale of one of the products
which has lower frequency of sale (Cross selling)
C. Special deals can be offered to enhance product sales
D. Optimizing store layout
Well known ecommerce website, Amazon employs innovative way of data mining with market basket
analysis. It displays on its home page, items “Frequently bought together” and “Customers who
bought this item also bought.”. This recommendation engine accounts for more than one third of its
sale.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

1.2 ASSOCIATION RULES IN MARKET BASKET:


In the market basket analysis, purchase transactions from one or more retail stores are downloaded
and analysed for identifying products being purchased together. The association rules are defined to
find out frequency of items that occur together.

1.3 LIFT CALCULATION


Lift can be calculated as two-way (for identifying association of 2 products), three-way (for identifying
association of 3 products).

1.4 TWO-WAY LIFT


Two way Lift = Actual number of transactions where products A and B were purchased /

[ Total no of transactions * Fraction of time A was purchased * Fraction of times B was purchased]

= P( A & B) /[ N * A * B]

P(A & B): number of transactions where products A and B were purchased together

N: Total no of transactions considered for this analysis

A: Fraction of time A was purchased

B: Fraction of times B was purchased

Suppose N = 1000

P(A&B) = 200

A = 0.4 (400 transactions where A was purchased 400/1000)

B = 0.3 (300 transactions where A was purchased 300/1000)

Then Lift = 200 / [1000 * 0.4 * 0.3] = 1.67

vegetables baby fruit milk DVDs meat

vegetables 1.00 0.96 0.99 1.00 0.96 1.01

baby 0.96 1.00 1.05 1.00 1.40 1.01

fruit 0.99 1.05 1.00 1.00 1.03 0.90

milk 1.00 1.00 1.00 1.00 0.98 1.06

DVDs 0.96 1.40 1.03 0.98 1.00 0.96

meat 1.01 1.01 0.90 1.06 0.96 1.00


Marketing Business Analytics Session10 Prof. Manisha Kapoor

Above table shows that baby product and DVDs have highest lift.

1.5 REFERENCES:
• Winston, W. L. (2014). Marketing analytics: Data-driven techniques with Microsoft Excel. John
Wiley & Sons.
• https://searchcustomerexperience.techtarget.com/definition/market-basket-analysis
Marketing Business Analytics Session10 Prof. Manisha Kapoor

2 6. LOGISTIC REGRESSION (SESSION 14-16)

2.1 LOGISTIC REGRESSION (USING R SOFTWARE)


In simple and multiple regression, the dependent variable (explained variable) Y is continuous or
sometimes it is discrete. Therefore, relationship between dependent and independent variables can
be easily established. However, there are situations when the dependent variable assumes few finite
values such as very satisfied, satisfied, not satisfied etc. Sometimes it can take binary values such as
1/0 or Yes or No. In such cases, simple or multiple regression may not be feasible. However, in such
situations with multiple response variable Logistic regression or logit can be used. When the response
variable takes only 2 values 1/0 or Yes/No is it known as binomial logistic regression. When it takes
more than 2 values, it is called as multinomial logistic regression.

Many marketing situations require probability associated with certain events, outcomes which can be
established on past data. For example:

• Predicting whether a customer is likely to buy a product based on available information such
as age, income group, lifestyle, family size etc.
• whether a loan seeker from a bank is likely to default on the loan
• whether a customer of a telecom company is likely to churn
• whether an employee of a company is likely to quit
Binomial logistic regression works on probability associated with the outcome. Probability of any
possible outcome can go from 0 to 1. Higher the probability, higher the possibility that the event may
happen. For example, probability of a prospective customers buying a car is 0.92 then there is higher
likelihood that he/she will buy the car. If we can get list of such prospective customers with higher
probability then marketing efforts can be directed towards them with better outcomes and higher
profits.

Like multiple regression, logistic regression takes more than one independent variable that include
categorical, continuous and discrete variables. We can check whether independent variables are
statistically significant or not. We need to retain only those variables which are statistically significant
and drop others. As we go through multiple iterations, final regression equation can be formed.

2.2 LOGISTIC REGRESSION USING R


R commands for logistic regression are as follows:

Table testdata.csv has following attributes:

Attributes: y, x1, x2, x3, x4


Marketing Business Analytics Session10 Prof. Manisha Kapoor

y is binary variable with values 1 and 0. 1 means Yes, 0 means No.

x1: continuous

x2, x3: categorical

x4: discrete

# read the flat file (CSV) into dataframe variable f1

f1 <- read.csv(file.choose(), header = T)

glm1 <- glm(y ~ x1 + x2 + x3 + x4, family = binomial, data = f1)

# Check the regression model

summary(glm1)

# this will display the significance of variables

# if any variable is not significant, then drop the variable and create 2nd model glm2

glm2 <- glm(y ~ x1 + x2 + x3, family = binomial, data = f1)

# Again Check the second regression model

summary(glm2)

2.3 MODEL FIT VERIFICATION


# for this purpose, we will create pseudo R-Squared

Pseudo_R_Squared <- 1 – glm2$deviance / glm2$null_deviance

Pseudo_R_Squared

0.72

# This shows that above model glm2 explains 72% of deviance which means model is a good fit.

2.4 DATA ISSUES IN LOGISTIC REGRESSION


Like multiple regression models, logistic regression may suffer from problem of Multicolleanirity. For
this purpose, we need to check the variance inflation factor of the model.

vif(glm2)

If VIF for all variables is less than 5 then there multicollinearity is not an issue in the model.

2.5 CREATE PREDICTION USING NEW DATA:


For prediction using above model glm2, we need to create data structure similar to earlier file with
the new data for which we want to predict.

newdata1 <- read.csv(file.choose())

predictYesNo <- predict(glm2, newdata = newdata1, type = 'response')


predictYesNo
Marketing Business Analytics Session10 Prof. Manisha Kapoor

1
0.94
Above value of 0.94 shows high probability of customer response as Yes.

2.6 COMPARE BETWEEN PREDICTED AND ACTUAL VALUES IN DATA


> predict <- predict(ml2, type = 'response')

2.7 CREATE CONFUSION TABLE TO REVIEW THE RESULTS FROM PREDICTION


> t1<- table(f1$y , predict > 0.5)

# here y is dependent variable with binary values

FALSE TRUE
0 132 168
1 71 629

N = 1000 Predicted No Predicted Yes


Actual No 132 168
Actual Yes 71 629
65% 78% 76% Accuracy

A confusion matrix is a table that is often used to describe the performance of a classification model
(or "classifier") on a set of test data for which the true values are known. It is another way to test the
accuracy of the model using the same data that was used to build the model.

2.8 REFERENCES:
• Hodeghatta, U. R., & Nayak, U. (2016). Business analytics using R-a practical approach. Apress.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

3 7. PRICING (SESSION 17-20)

3.1 ESTIMATING DEMAND CURVE AND OPTIMIZING PRICE


Understanding how pricing impacts revenue and profitability of firm is important aspect to be studied.
Firms follow variety of pricing strategies. Sam Walton of Wal-Mart followed low prices as a theme of
his business strategy and achieved commendable success by offering grocery and other household
goods at lowest prices. His focus was on attracting customers by offering low prices and thereby
increasing customer base and volumes. By achieving higher volumes, margins are lower but overall
profits increase. Pricing in luxury goods work differently. There are different pricing strategies and
consumers react differently to each one of them. In this module of analytics we will study different
pricing approaches to maximize profits.

3.1.1 Forms of demand curve:


There are multiple forms of demand curves that you can use to analyse marketing data.

3.1.2 Linear Demand curve:


q = a – bp.

q = quantity demanded, p = unit price, a and b are constants to be determined using linear demand
curve method.

Example: q = 20 – 5p

3.1.3 Power Demand curve


q = a * pb (q = quantity demanded, p = unit price, a and b are constants to be determined.

Example: q = 20 * p -2

3.2 PRICE BUNDLING


Recently there was a news that Reliance is going to offer Direct TV including Netflix, Amazon Prime,
YouTube, Disney-Hotstar, Internet service, landline phone as a single offering. Similar examples can
be quoted in several other industries. Automobile companies offer car, insurance, audio system,
security system, touch screen navigation system, keyless entry at a single attractive price. Computer
companies create a package of computer, printer, mouse, monitor and some other accessories at a
single price. Software companies such as Microsoft package Excel, Word, Visio, Outlook, Access and
sell them successfully to millions of customers. Retail companies package toothpaste with
toothbrushes and sell them together as an offer. These are all examples of bundling. Companies often
bundle products in an attempt to get customers to purchase more products than they would have
without bundling. When products A, B and C are bundled together they will be sold as a package at
Marketing Business Analytics Session10 Prof. Manisha Kapoor

total price which is lower than when they are sold individually. Therefore, it requires analytic model
for pricing decisions.

3.3 NONLINEAR PRICING


There are business scenarios when each unit sold will not be priced at the same rate. As we have seen
in the previous section, price bundling is a special case of nonlinear pricing because bundle price of A,
B and C is lesser than the sum of their individual prices. There are more examples of nonlinear prices:

Quantity discounts: There are several versions of quantity discounts. For quantity 100 or less price is
$10, for quantity 101 to 200 price is $8. Another version of quantity discount is for first 100 units price
is $10 and for any additional units above 100 price is $8.

Two-part tariffs: Some of the sports club have annual club membership fee of Rs 15,000 and then for
availing any facilities members have to pay a small fee of Rs 100 per use.

For maximising profit with nonlinear pricing, excel solver can be used. For a linear demand curve,
nonlinear pricing earns twice as much profit as linear pricing.

3.4 PRICE SKIMMING AND SALES


Pricing strategy for high-tech products is different than many other type of products. Product lifecycle
of such products is shorter. Therefore business needs to have a clear strategy to maximize their profits.

Prices of high-tech or fashion products generally drop over period of time. There are several reasons
for the drop in prices for such products.

Learning Curve: cost of producing products follow learning curve or experience curve (T.P. Wright,
1936). As per data prices drop between 10 to 30% when the product volumes double. Firms pass on
this cost saving to consumers resulting in price drops and it also help them in gaining more market
share. Thus company is able to expand capacity and produce more.

Competition: As market for new product such as smart phones expand, new firms enter that market
and launch their products. Due to competitive situation prices are dropped by firms.

Price Skimming: When a new product is launched, different potential buyers perceive different value
for the new product. Some buyers are willing to pay $100 while others will buy it at $50. If the company
launches product at lower price, it loses an opportunity to tap high valuation customers.

Therefore, prices are usually kept high initially as companies want to charge high prices to customers
who value the product highly and want to have something new, unique. After demand from such
Marketing Business Analytics Session10 Prof. Manisha Kapoor

customers slows down, price must be lowered down to be able to sell to those customers who may
be willing to buy it at lower prices. This is called as price skimming.

Sale: Companies announce “Sale” from time to time especially during holiday season by offering
discounts to attract those customers who perceive low-value for product or they are willing to buy at
lower prices.

3.5 REFERENCES:
• Marketing Analytics by Wayne Winston, Wiley Publication
Marketing Business Analytics Session10 Prof. Manisha Kapoor

4 8. CUSTOMER CENTRIC MARKETING MODELS (SESSION 21-24)

5 8.1 RFM ANALYSIS AND CASE STUDIES

5.1 APPLICATIONS OF RFM IN RETAIL AND OTHER INDUSTRIES


RFM Analysis

RFM is another approach to get insights into customer transactions in retail industry. Marketers
analyse the customer transactions to segment customers into different groups with the goal of
increasing customer retention, expanding customer base and focus marketing efforts on right set of
customers.

From the retail transactions, its possible to classify customers by time (when was the most recent
visit), frequency of their transactions and the amount they have spent.

Recency: when was the most recent visit by each customer, days passed their visit is an indicator
whether customer is still active or not. Active customers are more likely to return to store respond to
promotions.

Frequency: How frequently customer has visited the retail store or website. Customers which visit
frequently are important for the retailer as they keep the inventory moving.

Monetary value: How much customer spends on an average in the store or website? Some customer
visit frequently but they spend less on an average, whereas some visit once in a while and the value
of their bought items is quite high. Customers that spend more are more valuable for the retailer as
that results in higher revenue and profits for the company.

Each of the above parameters are important in their own sense. But any single parameter fails to get
a complete picture. A customer who visited recently may not be a regular customer, another customer
coming frequently buys low value goods, customer has purchased few high-priced items turns out to
be a one-time customer. Therefore, its in RFM above 3 are combined into single RFM score and that
is used as a basis for grouping them and analyse relative importance of each customer.

Customers are important for the business. It does no good to retail business by giving equal
importance to all customers. More efforts need to be focused on retaining a regular customer who
spends $1000 per week in a store than another regular customer spending $100 per week. Therefore,
there needs to be a way to understand relative importance of each customer. RFM addresses this
issue.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

RFM helps to answer some of the following questions:

• Who are my best customers which contribute higher revenues and profits?
• Which customers are likely to stop giving business?
• Considering their importance, which customers need to be retained?

5.2 COMPUTING RFM SCORE


Recency – R-Score

• calculate number of days since most recent visit of customer


• top 20% most recent visiting customers are assigned R score of 5, next 20% as 4
• continue assigning 20% customers each to 3, 2 and 1 R Score
Frequency – F Score

• Assign the highest 20% frequency customers F Score of 5


• Assign next 20% frequency customers F score of 4…continue till score 1
• Score 1 will be lowest 20% customers
Monetary Value – M Score

• Assign highest 20% monetary value customers as M Score of 5


• Next 20% assign M Score 4
• continue till M score 1
RFM Score

• At this step, each customer has been given R, F and M Score


• Average of R, F, M becomes RFM Score for each customer
• Sort all customers in decreasing order of RFM Score
We will get 125 different possible RFM Scores – 5 (5, 5, 5) to 1 (1, 1, 1).

Now analysing the RFM score, retailer needs to interpret the data on these lines and plan marketing
strategies accordingly:

• Customers with highest RFM Scores – most loyal and profitable customers, retain them by
giving them special privileges or offers.
• Customers with High Frequency and High Monetary score but low recency score – retailer
needs to reach out to such customers to bring them back to store
Marketing Business Analytics Session10 Prof. Manisha Kapoor

5.3 REFERENCES:
• Blattberg, R. Database Marketing: Analyzing and Managing Customers [Electronic
Resource]/Robert C. Blattberg, Byung-Do Kim, Scott A. Neslin New York, NY.
• https://clevertap.com/blog/rfm-analysis/
Marketing Business Analytics Session10 Prof. Manisha Kapoor

5.4 8.2 CUSTOMER LIFETIME VALUE AND CASE STUDIES


One of the main functions of Marketing is to acquire new customers. It takes lot of time for sales
people to engage and acquire new customers. Thus, customer acquisition requires investment of time
and money for the firm. Some of the customers are likely to move on to competitors services after
certain period, which is called as customer attrition rate. Therefore, a firm needs some analytical tool
to assess how much was invested in customer acquisition, how much was revenue generated, what is
attrition rate and whether customer is profitable or not over their association with the firm. This is
called as customer lifetime value calculation (CLV).

By using this tool marketing also moves from focusing on short term transactional relationship with
customers to long term association with customers (Rust et al., 2004)

This concept is quite useful in industries such as telecom, retailing where customer association with
the firm is for long term. After computing CLV, it can be determined whether its profitable to retain
the customer.

• CLV Determines present value of customers brought for an organization during customer’s life
cycle.
• CLV is used to classify its clients based on CLV rankings so that various marketing strategies can be
defined for each segment.

5.4.1 Metrics of Marketing:


• Sales / Revenue
• Market share
• LTV (Customer centric)

5.4.2 Segmentation Methods using CLV


The most common method of analysing customers based on profitability is to rank customers in terms
of profit as follows (Zeithaml et al. 2001). The top 5 % customers account for 40% profits and the next
15 % account for another 40%. These can be classified as Platinum and Gold customers respectively.
This, the top 20% of the customers account for 80% of the profits.

• Platinum: 5% of customers, 40% of Profits


• Gold: 15% of customers, 40% of Profits
• Iron: 30% of customers, 30% of Profits
• Lead: 50% of customers, -10% of Profits
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Firm can design strategies around each of these segments on investing into relationship with them.
For example, firm may want to develop Gold customers by investing into marketing efforts so that
move into platinum category.

5.4.3 Example 1
customer retention 70%
revenue per customer $ 300.00
discount rate 10%
profit% 50%
marketing cost $ 40.00
Net Profit/customer =
Net Profit per customer $ 110.00 revenue per customer - Marketing cost

Survival discount Discounted


Year retention rate Profit rate Profit
1 70% 1.00 $ 110.00 1.00 $ 110.00
2 70% 0.70 $ 77.00 0.91 $ 70.00
3 70% 0.49 $ 53.90 0.83 $ 44.55
4 70% 0.34 $ 37.73 0.75 $ 28.35
5 70% 0.24 $ 26.41 0.68 $ 18.04
6 70% 0.17 $ 18.49 0.62 $ 11.48
7 70% 0.12 $ 12.94 0.56 $ 7.31
8 70% 0.08 $ 9.06 0.51 $ 4.65
9 70% 0.06 $ 6.34 0.47 $ 2.96
10 70% 0.04 $ 4.44 0.42 $ 1.88
$ 299.21

in the above example, customer retention rate is 70%, revenue per customer is $300 with profitability
at 50%. Marketing cost is $40.

Which means profit per customer is $300 *50 – $40 = $150 - $40 = $110

The next table, see the calculations for 1 customers lifetime value.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Discount rate take the time value of money into account.

5.5 REFERENCES:
• Marketing Analytics using data driven techniques with Microsoft Excel – Wayne Winston
• Database Marketing – Analyzing and Managing Customers – Blattberg, Kim, Neslin, Springer
• Rust, R. T., Lemon, K. N., & Zeithaml, V. A. (2004). Return on marketing: Using customer equity to
focus marketing strategy. Journal of marketing, 68(1), 109-127.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

6 9. MARKET SEGMENTATION: CLUSTER ANALYSIS (SESSION 28-29)

6.1 CLUSTER ANALYSIS AND CASE STUDIES


Cluster analysis is a technique of forming groups of customers based on number of variables. This
grouping helps in finding out common characteristics of those customers and put them into segments.
Marketing students are aware of segments and how we use them for positioning our products.
Therefore, to develop understanding into customer groups is important. This is facilitated by cluster
analysis.

Therefore, clustering can be summarised as:

• Unsupervised Machine Learning Technique


• Exploratory data analysis used to understand data and its properties
• helps in identifying hidden groups using some common characteristics
• Used quite often in retail industry to find interesting buying patterns
• Helps in cross-selling, product promotions and other marketing business decisions
Cluster analysis can help in finding out previously unknown relationships within data to be able to
make marketing decisions.

6.2 CLUSTERING AND CLASSIFICATION:


Clustering and classification appear to be the same. But they are different in their goals.

Clustering is an unsupervised learning technique used to separate the instances into different clusters
based on common characteristics such as age, income, lifestyle, family size, education etc. This
grouping happens through a iterative process using computer algorithm.

Cluster analysis results in customer segments such that customers within a segment are similar to
each other, and collectively different than other segments. The clustering variables are demographic,
psychographic, transactions, online data or even call centre data based on customer interactions.

Once cluster are formed, those can be used for targeting with particular marketing offer.

Classification is supervised learning technique; wherein predefined categories are assigned to each
instance based on available features. e.g. of classification is logistic regression. For example, when
classification model is applied to the customers banking data, customer is classified into defaulter or
no-defaulter based on his transactions, demographic and psychographic variables as independent
variables. Thus in classification models, 2 groups (defaulter, no-defaulter) are predefined. Whereas in
cluster analysis such groups are not predefined. These get created. Once system creates clusters
Marketing Business Analytics Session10 Prof. Manisha Kapoor

through iterative process, we can visually represent clusters on scatter plots and then decide either
to increase or decrease the number of clusters.

6.3 CLUSTERING STEPS:


Clustering consists of several steps:

• Select variables on which to cluster


• Select similarity measure and scale the variables
• select clustering method
• determine number of clusters
• conduct cluster analysis
• interpret results and apply them

6.4 CLUSTERING METHODS

6.4.1 Hierarchical Clustering


Hierarchical develops tree-like structure either by serially merging clusters (agglomerative method)
or by successively dividing clusters (divisive method).
Marketing Business Analytics Session10 Prof. Manisha Kapoor

6.4.1.1 Agglomerative Method


Given n customers, this method starts with n clusters. Each
customer is considered as a cluster. it then combines 2 most
similar customers into a single cluster and produces n-1
clusters. This step is reduced to produce n-2, n-3 …. clusters.
This merging process is continued till we reach single cluster at the
top.

We have 5 customers; each one is considered as a cluster.

Iteration 1: customers 1, 2, 3, 4, 5…5 cluster solution

In above example, we have customers 1 and 2 are closest to each other (or similar). Hence we create
one cluster of 1-2 in iteration 1

Iteration 2: 1-2, 3, 4, 5…4 cluster solution

Distance between 3 and 4 is lowest (3). Therefore, we will combine these 2 into a cluster 3-4.

Iteration 3: 1-2, 3-4, 5…3 cluster solution

Iteration 4: 1-2, 3, 4, 5…4 cluster solution


Marketing Business Analytics Session10 Prof. Manisha Kapoor

Iteration 5: 1-2, 3, 4, 5…4 cluster solution

6.4.1.2 Divisive Method


Divisive method is in opposite direction. We start with a Single cluster consisting of all n customers.
Then divide customers into 2 customers so that customers in 1 cluster are different than the other.

6.4.2 Non-Hierarchical Clustering

6.4.2.1 K-Means Clustering


K-Means clustering is one of the most popular methods of clustering.

In this method, user can decide to create k clusters from the given data.

Let us say, we have 100 records for customers with their age, income. Now we want to create 2
clusters from this data.

Randomly select any 2 points, i.e. customers A and B as centroids of the 2 clusters. Their coordinates
Age, income become the values which are considered for computing shortest distance with other 98
customers.

Iteration 1: Then we assign rest of the 98 customers to these 2 clusters, based on shortest distance
with the centroids A and B. 2 clusters are formed with customers assigned to each cluster.

Iteration 2: Calculate new centroids of the 2 clusters by taking means for all the points assigned to
them. Find out shortest distance for each of the 100 customers with the 2 centroids (say C1 and C2)
and assign them. Now customers will move from existing cluster (with centroids A,B) to newly
formed clusters with centroids (C1 and C2).

Iteration 3: Again, calculate another revised 2 centroids of the 2 clusters from iteration 2 above by
taking means for all the points assigned to them (let’s call them C3 and C4). These will be different
than from iteration 2. Find out shortest distance for each of the 100 customers with centroid C3 and
Marketing Business Analytics Session10 Prof. Manisha Kapoor

C4 and reassign them. Now customers will move from existing clusters (C1,C2) to newly formed
clusters (C3,C4).

This process repeats till centroids do not change and customers do not get reassigned.

6.5 APPLICATIONS OF CLUSTER ANALYSIS


Cluster analysis is used for segmenting customers based on common characteristics. Once these
segments are identified, marketer can target one of the segments for promoting products.

6.6 REFERENCES:
• Hodeghatta, U. R., & Nayak, U. (2016). Business analytics using R-a practical approach. Apress.
• Blattberg, R. Database Marketing: Analyzing and Managing Customers [Electronic
Resource]/Robert C. Blattberg, Byung-Do Kim, Scott A. Neslin New York, NY.
• https://www.differencebetween.com/difference-between-clustering-and-vs-
classification/#:~:text=Clustering%20is%20unsupervised%20learning%20while%20Classification
%20is%20a,subsets%20to%20group%20the%20instances%20with%20similar%20features.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

7 10. SOCIAL MEDIA ANALYTICS (SMA) (SESSION 30)

7.1 SMA DEFINITION


Social media analytics is the process of gathering and analysing data from social networks such as
Facebook, Instagram, LinkedIn and Twitter.

It can be defined as the practice of gathering data from social media websites and analyzing that data
using social media analytics tools to make business decisions. The most common use of social media
analytics is to mine customer sentiment to support marketing and customer service activities.

It is commonly used by marketers to track online conversations, comments about products and
companies. One author defined it as "the art and science of extracting valuable hidden insights from
vast amounts of semi-structured and unstructured social media data to enable informed and insightful
decision making.

7.2 SOCIAL MEDIA ANALYTICS


Social media analytics is an emerging concept which is becoming part of mainstream marketing
strategy. It is based on social media data created on sites such as Twitter, Facebook or WhatsApp. It
is concerned with developing and evaluating informatics tools and frameworks to collect, monitor,
analyse, summarize and visualize social media data (Zeng et al., 2010).

Big Data created on the social media consist of text messages, songs, pictures, videos etc. People share
information through text messages, videos, pictures, songs etc. They often express their intention to
purchase a product, request for feedback, share their service experience or product reviews on the
social media. This data from social media has valuable information for firms which can analyse and
mine this data. Figure below summarizes a generic process consumers follow for purchasing products.
This process illustrates how consumers research products using search engines and social media. They
also contribute their views, opinions on social media before and after purchasing the product.
Technology enables firms to use this medium to support business strategies and tactical decision
making. We summarize some of the ways in which social media analytics can be helpful for the firms
in the next sub-section.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

Figure - Process of Product Purchase and content generation on social media

7.2.1.1 Sentiment Analysis:


Some companies collect and analyse consumer sentiments expressed on social media about their
products or services. There are several ways of analysis using natural language processing, sentiment
lexicons or machine learning algorithms in wide variety of industries. Sites such as sentiment140.com
provide insights on customer sentiment on products or services. Google Analytics provide various
services – smarter advertising (data driven relevant and effective advertising), deep customer insights
(captures customer behaviour across CRM, point of sale, call centre) etc.

7.2.1.2 Competitive Intelligence:


In today’s highly competitive markets, business decision makers seek consumer feedback about their
own products as well as those of their rivals. Kim et al. (2016) suggest use of social media analytics for
gathering competitive intelligence about firm’s product and products offered by competitors in the
same market segment. Their comparative analysis of Twitter data for 2 competing smart phones
(Apple iPhone 6 and Samsung Galaxy S5) over a period of time revealed correlation between
sentiments expressed on the social media and difference in market demand for the 2 products. They
investigated 3 metrics a) social media volume, b) purchase intention and c) consumer sentiment for
further analysis. This approach can assist firms to predict market sales performance and estimate the
gap between competing products. As a result decision makers can adjust market strategy rapidly and
compensate weakness contrasting with the rivals as well. There are various possible ways in which the
intelligence from social media can be tapped by firms to develop competitive intelligence by helping
organization understand their suppliers, competitors, environments and overall business trends.
Business intelligence obtained from social media can enable business analysts and decision makers to
develop market insights into consumer behavior, discover new marketing ideas, improve customer
satisfaction and finally improve ROI (Kim et al., 2016).
Marketing Business Analytics Session10 Prof. Manisha Kapoor

7.2.1.3 Marketing and Brand Promotion Strategy:


More and more people are getting online due to popularity of smart phones, tablet computers, leading
to generation of higher levels of large datasets and more particularly geo-location data (Bell, 2012).
Online presence of consumers for reading news, searches, ecommerce sites and social media sites
provides a huge window of opportunity for firms to connect with them. Consumers typically research
products and services over internet and social media before arriving at purchasing decisions. The
extent of research depends on price and risk of purchasing that particular product or service. For
example, for high price and high risk purchase such as home or airline ticket consumers tend to put in
extensive research, whereas for commodity items their research may be limited to few searches on
web. Firms can understand the nature of this activity, analyse the data associated with it and plan
their marketing and branding strategy accordingly. For high price products wherein consumer
purchase involvement is higher, firms are better off providing information about products rather than
using entertaining content for advertisements on the social media (Lally, 2007). For low cost
commodity items, firms can use entertaining methods to attract consumer attention (Coursaris et al,
2016).

7.2.2 Social Media data:


Social media data has numerous data types originating from variety of data sources. It includes social
network media (Facebook, Twitter, WhatsApp, Instagram etc.), wikis, blogs, RSS feeds and news),
financial data, customer transaction data, telecom data, spatial data.

7.3 THE 3 STAGES OF SOCIAL MEDIA ANALYTICS ARE AS FOLLOWS:


• Data Identification

• Data Analysis

• Information interpretation

7.4 BENEFITS OF SOCIAL MEDIA ANALYTICS


• Improved marketing strategies (75% of the cases)

• Better customer engagement (65%)

• Better customer service (35%)

• Better reputation management and brand awareness (30%)

• Product innovation (30%)

• Business process improvement (25%)


Marketing Business Analytics Session10 Prof. Manisha Kapoor

• Discerning new business opportunities (20%)

7.5 REFERENCES:
Zeng, H. Chen, R. Lusch and S. Li, "Social Media Analytics and Intelligence," in IEEE Intelligent
Systems, vol. 25, no. 6, pp. 13-16, Nov.-Dec. 2010, doi: 10.1109/MIS.2010.151.

Batrinca, B., & Treleaven, P. C. (2015). Social media analytics: a survey of techniques, tools and
platforms. Ai & Society, 30(1), 89-116.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

8 11. STRATEGY (SESSION 31-33)

8.1 MARKETING PLANNING AND STRATEGY DECISIONS:


At an organizational level, the top management decides on firm’s vision, mission and values. The
decide growth strategy, product portfolios, markets and geographies to be tapped. This is the context
for divisions to formulate their own business strategy. Then the managers in division decide their
marketing strategy. The strategic marketing process refers to the steps taken at product and market
level to promote products and gain market share.

8.1.1 Wind and Roberson Framework for marketing concepts and models:
Section I – traditional assessment of market opportunities and business strengths
a) Analysis of opportunities and threats
b) Analysis of business strengths and weaknesses
Section II – The marketing strategy core
a) Segmentation by positioning analysis
b) Opportunity analysis
c) Synergy analysis (in advertising, distribution, manufacturing etc. among products,
segments, marketing mix components)
d) Functional requirements (specifications)
e) Portfolio analysis, the analytical core of the process
Section III – The generation and evaluation of objectives and strategies
a) Generation of objectives and strategies
b) Evaluation of objectives and strategies
c) Implementation, monitoring and control of the program
Marketing Business Analytics Session10 Prof. Manisha Kapoor

8.1.2 The Product Life Cycle (PLC)


An important consideration for business
planning is product life cycle. In competitive
situation the product’s stage in its lifecycle
needs to be clear to decide product strategy.

PLC can be divided into 4 stages – inception,


growth, maturity and decline. Most of the
profits come during growth and early stage
of maturity.

Firms have to use right marketing strategy


depending on where product is within PLC. It can be used in several marketing decision making –
predictions, pricing strategies, promotions and production planning (Smallwood, 1978).

There are various factors that contribute to shorter product life cycles – number of options for
consumer, various price points at which products are available, introduction of newer products with
better features. Firms which ignore the PLC will not be able to sustain in the long run. PLC time varies
as per the price, technology evolution in the field and the duration for going “from product concept
to market”. Technology driven products’ such as smart phones, laptops, tablets have few months PLC
whereas R&D process for developing car can go up to few years. Therefore, car models have longer
life cycles.

Thus PLC is an important concept in designing marketing strategy. It can be used for marketing
planning. There are several strategic alternatives in the context of PLC such as developing newer
model of the same product, advertising campaigns, pricing strategies. These will extend the life of
existing product. Alternate strategy which is commonly used by mobile and electronic companies
(such as Samsung, Sony) is to keep introducing new products with better features with altogether new
life cycle for the new product.

8.1.3 Cost Dynamics: Scale and Experience Effects:


It has been established that with the higher market share for products, firm’s profitability also gets
better. With higher market share, volume of products sold is higher and per unit fixed cost gets
allocated to higher number of units manufactured. Thus overall cost reduction is possible for higher
volumes. A difference in market share between competitors of 10% translate into 5% difference in
pre-tax ROI (Lilien et al., 1995). Secondly, the duration required for manufacturing and assembly also
falls over long term, resulting in lower labour costs per unit. This is the effect of experience or learning
curve.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

This learning or experience effect can be mathematically represented as follows:

Cq = Cn (q/n)-b

Cq = marginal cost of qth unit

Cn = marginal cost at nth unit

q = cumulative product to date

n = cumulative production at particular earlier time

b = learning constant

If experience for the employees doubles, let us say cost per unit drops to 80% of original levels. In this
case the learning rate is 80%. Learning rate is related to learning constant as:

r = 2-b X 100

b = ln 100 – ln r / ln 2

There are several other models that demonstrate the common characteristic that with experience
labour costs decline over time.

But not all researchers agree that it is experience that leads to entire cost reduction. Alberts (1989)
argues that causes of cost reduction can be partly because of economies of scale and partly because
of innovations. These innovations could be operators (better sources of input raw materials,
manufacturing process, alignment of resources etc.), improvement in technology, process
improvements.

Scale based cost reduction could be due to reduction in excess capacity, cost-effective substitutes,
better deals on sourcing due to higher volumes.

Above observations outline the fact that experience by itself does not lead to cost reduction, it
provides a framework for cost reduction projects.

8.2 MARKETING MIX ANALYSIS AND MODELS


The term “Marketing Mix” has been in use since several decades. McCarthy was the first person to
suggest four components of Marketing Mix as Product, Price, Promotion and Place. Marketers use
these 4 variables to design marketing strategies. In the long term, business firms can all 4 variables
can be changed to suit the strategy but in the short term only few of them can be tweaked. As service
sector is growing, another set of variables got added to marketing mix – People, Physical evidence and
Process. For service sector such as travel, tourism, hotel industries role of human resources i.e. people
Marketing Business Analytics Session10 Prof. Manisha Kapoor

is important. The quality of service depends on people aspects of service. Process is another important
component of the mix. Process can be made more efficient by improvements.

Marketing mix modelling is an analytical approach that uses data from various internal and external
sources. The data sources can be transactional data for the past periods, incremental data in response
to promotional and advertising efforts, customer data and financial data. Statistical techniques such
as linear or non-linear regression methods are applied with dependent variable as sales. The models
are used to determine statistically significant marketing mix variables and then use the knowledge to
design marketing strategies or optimise the marketing budgets. It is not uncommon to use sales data
for forecasting or simulations purpose for upcoming periods.

For example, price is an important element in the marketing mix which impacts sales volumes. Slight
increase in price adversely affects sales volumes. A model can estimate % change in sales in response
to each percent change in price. In a competitive situation, this provides an important guideline to
decision makers to decide right price for the products. Similarly, promotions have favourable impact
on sales. A model can be designed to include promotion as input variable and estimate the volumes
generated by each promotional events.

8.3 DESIGNING DATABASE MARKETING COMMUNICATION:


Blattberg et al. (2008) define) database marketing is a process of using customer database to enhance
marketing productivity through more effective acquisition, retention and development of customers
(Blattberg et al. 2008). In marketing analytics, we develop analytical models for calculating customer
lifetime value, regression and logit models for predictions, customer segmentations, consumer
behaviour models etc. Once we have these models ready we need to engage customers through
proper communication. Therefore, communication needs to be planned, designed and executed to
achieve marketing objectives. It includes selection of media, personalization of communication and
engagement. The stages of this communication can be narrated as follows:

• Planning Process – This stage is for setting objectives, strategy for achieving objectives and
budgeting. Objectives are quantitative in nature though there could be some qualitative
elements in it. Strategy consists of creative, offer, product and media plans. For budgeting
baseline can be previous year’s budget and based on objectives and strategy it can be
modified.
• Developing copy Content – This stage includes creative, offer and product.
Marketing Business Analytics Session10 Prof. Manisha Kapoor

• Select Media – Database marketer has several options for communicating with customers –
emails, social media, catalogs etc. Media selection depends on expected exposure for the
contents to the potential customers with the intention of maximizing response rate.
• Evaluate Results – to measure effectiveness of database marketing communication, 4
methods are adopted – calculation of profits, statistical analysis, embedded testing, surveys.

8.4 REFERENCES:
• Lilien, G. L., Kotler, P., & Moorthy, K. S. (1995). Marketing models. Prentice Hall.
• https://en.wikipedia.org/wiki/Marketing_mix_modeling
• Blattberg, R. Database Marketing: Analyzing and Managing Customers [Electronic
Resource]/Robert C. Blattberg, Byung-Do Kim, Scott A. Neslin New York, NY.

You might also like