Quantitative Marketing Research

COURSE LAYOUT
Week 01: Value and Willingness to pay

• Setting the Framework

• Elements of Value
• Value and Willingness to Pay (WTP)
• An Example: WTP
• From Problem Definition to Decision Making

Week 02: Indirect Measurements of Value

• Exaggeration Bias
• Second Price Auctions
• The Van Westendorp method
• Conjoint Analysis

Week 03: Survey Design, Constructs and Scales

• Defining Survey
• Asking the Right Questions
• Common Pitfalls in Survey Design
• Construct Validation
• Likert Scale
• Cronbach’s Alpha

Week 04: Segmentation, Targeting and Positioning

• Defining STP
• An Example: STP
• k means clustering algorithm
• Application of k means clustering to market segmentation

Week 05: Ethics of Marketing Research

• Ethical Dilemmas
• Price Discrimination
• Privacy and Targeted Advertising
• Influencer Marketing
 Week 1: Basics of Quantitative Marketing Research

By the end of this week, you will be able to:

 Define and explain Willingness to Pay (WTP)

 Summarize the concept of value using a value diagram
 Illustrate how to go from defining a problem to coming to a solution
or decision

VALUE

Could there be different kinds of value? Yes, they are. For example: Functional, Economic,
Experiential and Social. And what do they do?

Do they define consumer's choices? Does it help a marketer to understand the consumers'
preferences? Does it help you to sell your goods or services? And most importantly, knowing and
understanding the concept of value, does it guide you (or, a marketer) to take more profitable
decisions?

Title: Value
I have in my hand, a bottle of Bisleri mineral water. Many of you would have bought such a bottle
too. Recall one such incident and why did you buy it? Maybe pause the video for a minute and think
about it. You were probably thirsty, and the water quench your thirst, or you may have bought it for
the future anticipating thirst, you also probably bought this bottle instead of drinking out of the tap,
believing it will be safer, right? And these are in marketing parlance called Functional Value. But
then why did you buy a 15-rupee bottle of bisleri and not a much more expensive bottle of Evian
bottled somewhere in beautiful Switzerland or some other similarly expensive brand. To the
marketer, you saw that the less expensive bisleri was sufficient, and there was no need to buy
something more expensive and this we will be called Economic Value. And then there are features of
this bottle or brand that have nothing to do with quenching your thirst, maybe you like the design of
the bottle, maybe you just derive pleasure from the experience of buying it, etcetera, etcetera,
completely unrelated to any functional our economic benefits. Maybe this may not make much
sense in the case of a mineral water bottle, right? But think of when you buy clothes or jewellery or your
last visit to a spa, these elements relating to pleasure are called Experiential Value. And finally, there
may be an interpersonal reason for buying this bottle. Maybe your friends like this brand . Maybe you
can share it with a loved one and things like that, we will term this a Social Value. Now, think of any
product or service, and you will realize that all of them have different elements of these four
dimensions of values and complex combinations of these, along with different contexts and what we
call your personal value systems will drive your the consumers' choices. On the other hand, if you're
a marketer, it may be important for you to understand a large number of consumers preferences
and make strategic decisions that will help you sell you of goods or services and thus make profits.
Let's try out the following exercise. You see a table on the screen, which has four columns or take
any for goods or services and try to evaluate your own perceptions of value for them. One way to do
this is to assign a score of 1 to 10 for each of the following four dimensions of value. Try doing this
on your own and you will understand or appreciate the concept that we spoke about a little more
Value Quantified

Think of any product. Or, think of any service. Let's make this easy for
you.

You are taking this course in some device. Laptop, desktop, mobile phone,
tablets etc.

Would you value a device more which has more storage capacity than the
rest? Probably yes! And why?

If you think through now, you must be thinking in terms of the four
dimensions of value. Which dimension of value is most important here?
Functional? Economic?

In the following lecture video, examples are discussed so as to make you

understand value can be quantified and this helps in decision making.

o Title: Value
Continued In the previous video, we spoke about the four dimensions of value Functional, Economic,
Experiential and Social, but more from a qualitative angle. In this session, we will try to come up with
a quantitative or monetary number to this value. How do we reconcile to the possibility that some
aspects or attributes may add more value than others? For example, we call the bisleri bottle that I
showed in the previous video, put a value of 525 ml bottle more than a 500 ml bottle. If yes, how
much would you be willing to pay for it? Which dimension of value would it effect, and how? So,
most likely the experiential and social aspects will be unaffected by the quantity of water that you
give while the economic value may increase, you are getting more water for the same price.
However, functional value may increase or decrease. Given that the bottle will now be a little
heavier. What if I were to change the color of the label? Would it cause you to value the bottle less
or more? What about aesthetics you have designed? Could I get Shahrukh Khan to endorse Bisleri
and get you to value it more or what if I got Serena Williams instead? These are some of the
decisions that marketing managers routinely face. If you have some exposure to marketing before,
you may have heard that the duty of the marketing manager is to create and communicate value to
the consumer or end-user. But how do they do it? By tinkering with aspects of product, place,
promotion, and place. Famously, called the 4 P's of Marketing. These are decision variables in your
control, you can tinker and experiment with them and the market will respond to your actions by
buying or not buying your products. But now here's a hypothetical question. Imagine you're a
marketing manager at Lenovo. They manufacture laptops. Market research has shown that the ideal
computer should have one terabyte hard disc space, 16 gigabytes of RAM where less than a kilo be
priced under $100 and have at least five years of warranty and be advertised during prime time and
during major events like the Soccer World Cup and the Olympics, your boss tells you to do all of this,
should you? Pause the video and think about this for a while. Now, you would like to do all of these
and satisfy every consumer, right? But all of these activities have a cost associated with them. There
are costs associated with improving product specifications, there are costs associated with reducing the
price, with offering a longer warranty, and, of course, costs associated with advertising during the World
Cup, etcetera, etcetera. What should guide you is the careful evaluation of each alternative and
careful cost-benefit analysis. This course will give you an idea about such analysis. We will discuss a
few methods of collecting data along with their analysis to guide your decision making
Willingness to Pay (WTP)
What is WTP? Well, it is quite intuitive.

It is the maximum amount of money that a consumer is ready to pay for a certain good or
service. We do that all the time when we look at or intend to buy any product or service.

As a marketer, it is of utmost importance to figure out the WTP for the product or service that
he/ she is offering. Based on that, pricing decisions are taken.

Title: WTP and Value Diagram

Let's move to another related concept that of Willingness to Pay or WTP. This is an important
definition. It's defined as the maximum amount of money a consumer is ready to pay for a certain
good or service. Now, keep this definition in mind. It is the maximum amount of money that a
consumer is ready to pay for a certain good or service. Now, remember our bottle of bisleri water,
you perceived some value for it and are ready to spend money to buy it. 15 rupees seems a fair
price, right? Maybe even 20. But the value you get in the transaction now is lesser because you have
to give up more money for it. Would you play 21 rupees? Maybe, but grudgingly, 22 rupees Yes, 23/-
no. What about 22 rupees and 50 paise? No. Here I'm just outlining my own value system. Right
now, I may be willing to pay at most 22 rupees for this bottle of water, which I will term as my
willingness to pay. It may increase if I'm in a desert with no other options or it may reduce in my
office where the water cooler next door gives me chilled and purified water for free. Though this
willingness to pay is just contextual, it is still a useful theoretical concept to build on, especially to
guide our decisions while launching new products into the market. Consider this diagram. So, now
let us draw what we will call a value diagram, and this will illustrate the concept of value better. The
value diagram is nothing but a straight line that starts at zero and at the other end we will represent
a single customer’s willingness to pay and imagine you're selling some product and there is a cost
and there is a price. Now, what you will see here is that since the price is lower than the customer's
willingness to pay, customer persists some sort of value and since the customer perceives a positive
value, there will be a transaction. On the other hand, what you will see is that since your cost is
lower than in the price you sell it at, you as a firm will observe a profit, right? And in other words,
you can say that this profit is a firm value. So now you see that both you as a seller and the customer
as a buyer are perceiving positive value and this is good for both of you and that's why a transaction
happens and a little bit after this, we will show you some alternate formulations of the same value
curve as I call it or a value diagram, and you will see that not every situation can be as ideal as this.
But for now, let's just see where a marketer can come in and increase this profit. So again, a firm’s
role is usually to try and increase the profit. And as a marketer working for a firm, this is what you
want to do. So, do you see the price? So, one thing you could do is increase the price. And if you
increase of price, what you see is this profit term, cost minus price naturally increases. Now,
assuming that this is a single consumer, this is true. This would also be true if all consumers across
the market had the same willingness to pay then you could go up to the willingness to pay and get a
max profit per consumer that is, of course, of willingness to pay minus cost. So, you set the price as
willingness to pay and this is very good, except that if you go even slightly above willingness to pay,
then... So, if price is even slightly greater than willingness to pay, then no transaction happens. Of
course, in real life, you don't have the same willingness to pay for everybody. What I will pay for
something can be very different from what somebody else pays for something. So, one thing as a
marketer's you will absolutely have to do is find out what this willingness to pay is. Finding out
willingness to pay is a very, very complicated exercise and there is no way he will ever get it
perfectly. And we will see why over the next two weeks, but finding out willingness to pay, finding
out or in marketing research terminology, Estimation of Willingness to Pay is thus a very, very
important part of marketing research and of marketing in general
The 3rd thing you could do is reduce cost, right? So, if you reduce cost, you will see that
automatically cost minus pricing, price minus cost increases, and profit can increase. Now, there are
many nonmarketing ways of reducing cost. If you're an industrial engineer, if you're a production
manager, etcetera, etcetera. You could reduce cost by making more efficient systems in place,
supply chains, etcetera, etcetera. If you're a marketer, on the other hand, what you would do is
incorporate cost-saving into the product design itself, and marketing research becomes very
important there. One of the topics that we will be talking about is Conjoint Analysis, which helps you
determine what feature a consumer is willing to pay for and what features they're not that much
willing to pay for. And you know that every feature you add will have a cost associated with it. So,
one of the very important things that a marketer does is to find out what these features are and
where people are willing to pay more or where people are willing to pay less. Now, I'm going to draw
that figure again, this WTP is price, there is cost, and there is zero. Let's now, make a slightly
different version of the same thing, cost what I'm doing is interchanging, willingness to pay, and
price. What do you think happens here? Just pause for a couple of seconds and try to think, what
you will notice is... minus price is less than zero in this case, and your product will never sell. So, if
you have made wrong estimates of willingness to pay and end up over-pricing your product, you are
killed instantly in the market. Many products actually failed in the market simply because they have
been overpriced. And one of the things as a marketer you want to do is find out this willingness to
pay well so that you could price your product accordingly. Now, there are many products for which
you cannot charge a price simply because consumer willingness to pay is necessarily low. Think
about it. Would you ever pay to use Facebook? Would ever pay to use Gmail, etcetera, etcetera?
Probably not. And these companies know that very well, which is why, instead of charging you a
price, they know that they will still have a huge market base. And instead of charging you a price,
they get advertisers to pay for the same thing. Even the TV channels you watch actually operate on
the same model. Another case here and this is actually fairly common what you will see. Price is very
low, and cost is higher than price, and both are lower than willingness to pay. So, what do you think
will happen here? You see that the customer is very, very happy here, willingness to pay minus price
is much greater than zero now and the customer is getting huge amount of value. On the other
hand, you as the firm getting negative profit, or you are a loss-making entity now. Can you think of
companies that operate on loss and not without ads? Think of some e-commerce portals or you may
have some taxi-sharing services, etcetera. If you read the newspaper, you will see that they are
burning a lot of cash, and they may make some profits sometime in the future. But at the present,
what they're doing is operating it costs, which are higher than the prices that they charge. This is
fairly common despite defying your common sense that they may be loss-making and why would
you do that? This is fairly common because what these companies are trying to do is to gain you a
loyalty and a large user base by giving you a very high consumer value, hoping that over-time you
will be loyal to them and over time, what they would like to do is flip the cost and price so slowly
they will start increasing price. And over time, as they get more and more people, they will realize
some sort of economies of scale, which is not necessarily something to do with marketing, but if
you're manufacturing more product, the per capita cost or the per-unit cost of the product becomes
less, and they reduce their cost somehow and increase their price because now they have more
market power and hopefully over time they would like to transit to the ideal scenario

So just to recap, this value diagram is a very, very useful diagnostic tool for you as a marketer to find
out where you stand. Do you know something about the willingness to pay? Are you overpriced or
under-priced? You do not want to be under-priced either because then you're leaving a lot of money
on the table. You want the price to go as close to the willingness to pay as possible, assuming that
you know what the willingness to pay is and you want your cost to remain under the price at least in
the long run. So that's one of the goals of any marketing manager.

WTP - The Car Example

So, let's do an example to understand WTP. So, here is the situation.

Imagine that you are selling made to order cars to a market of 10 people. Each person
may or may not buy a car based on their WTP. If they buy a car, they buy at most one
car.

It costs you 500,000 rupees to manufacture one car. Given below are the WTPs of 10
people.

At what price should you sell your cars?

Also, assume that you are bound to sell every car at the same price.

WTP – An Example So, here's a very stylized example of a small willingness to pay exercise. So,
imagine you are selling cars and some fancy cars and through some secret method, you know that
your market consists of exactly 10 people and these are the 10 people and, of course, this is again
stylized and so I have sorted them by the willingness to pay, A, B, C, D, E, F, G, H, I, J, right? And you
have a cost per car. So, you, make cars only to order at cost of 500,000 rupees per car. Now, just
take a look at this table and tell me what would your pricing strategy be. Imagine that you have to
charge the same price because of legal restrictions, you can't change the price for different people.
Would you charge the mean? Would you charge the median? Would you charge the mode? Maybe
you subtract the cost from everyone and then charge the mean, median, mode and a lot of other
things are possible, right? And when I ask this in my class, typically at the outset, that's the kind of
answer that I get from people. What we're going to do now is do a little bit of a what-if analysis. So
again, we know that cost is 500,000 rupees and clearly charging anything less than 500,000 rupees is
the loss-making proposition and you don't want to make loss. So, you know that price has to be
greater than cost. So, straight away what we can tell is that there is no point selling to Amir. There's
no point selling to Bill. There is no point selling to Chandrika. Suppose you sold at 5lakh rupees or
500,000 rupees, or you don't really care if David buys your product or not, you make no profit, no
loss. The lowest you could sell it is at 500,000 rupees and these are the people who your target
market would be, right there you can eliminate that. So, now what we're going to do in the nextstep
istry out different prices and I'm just going to increase them by 50,000 and you will realize why? So,
suppose I want to sell at 500,000 rupees here, David would buy it, Ehsan, Farah, Ganesh, all of them
would buy it and so you would have seven units sold, but now price and cost are the same 50,000
rupees. So, irrespective of how many you sell he will still make a net profit of 0 rupees. That's not a
very desirable situation. Clearly, you can price a little higher than that. So, now let's try the next
iteration, 550,000 rupees. At this point, you know, it's higher than David’s' willingness to pay, he is
not going to buy it. So, Ehsan up to Juliet will buy it, these people will buy it and thus you will have 6
units sold at a profit of 50,000 and if you multiply these two you get 300,000- rupee profit. Great. So,
can you do better now? Can you increase the price a little more? Let's try a slightly higher price,
600,000 rupees and now you know even Ehsan will not buy it, Farah to Juliet will buy it. So now the
market becomes 1, 2, 3, 4, 5,so 5-people buy it and 600,000 and your costis 500,000 so 100,000-
rupee profit and this will give you a profit of 500,000 rupees. You see it is a little more 300,000 -
500,000. In the same logic, what we can do is we will try different prices for each of them find out
how many units are going to be sold and multiply with the profit margin. If you have a spreadsheet
program like Excel, this is done instantaneously. Now you see something very interesting here. The
price goes up. So, if I were to draw a curve of profit versus price, this is a very crude diagram. It could
be something like this. Below 500,000 you have a negative profit. At 500,000 you have a zero profit.
As you can see, it goes up, then you see a maximum being reached at two points 150,000 and
200,000. So, these are points to a maximum is reached. So, if you look at these two prices 650,000
and 700,000 you will see that profit is maximized. These other two prices at which profit is
maximized. So now this gives you a natural question, right? You have two points of maxima clearly
somewhere in between... So, whatis price were in between 650 and 700,000. Surely, you should be
able to get some more right at 675,000 - 680,000 whatever. Copyright © IIM Bangalore. All rights
reserved Let's just investigate. So, what I'm going to do is very simple, let's assume price. Let's
increase by one rupee. So, 650,000, what happens now? At 650,001 rupees who buys your product?
Ganesh does not buy it anymore; you have exceeded his willingness to pay. So only three people buy
your product and profit you get equals a 150,001 which is your margin into 3 and if you do the
calculation, you will see that this is much less than 600,000. Same thing, so, actually you will see that
this is for 450,003 is that right? So, this is much less than 600,000. Let's try the other upper limit.
now, coming to the upper limit suppose you take 6 a price of 699,999 atthe same price again you
willsee only three products are being sold. So now profit would be 199,999 times the three that you
have sold, which would be 599,997 so that is still 3 rupees less than what you could have got with a
price of 700,000 or 650,000. Take a few minutes and let this somewhat counterintuitive resultsink in.
Before we generalize this for a large number of customers with a statistical distribution function
which approximates a willingness to pay distribution.

WTP - The Car Example - Generalized Case

So, in the previous example, we took a sample of 10 people. Now, we will do the
extend the same example in a more generalised case.

To understand this you need to brush up on your basic mathematics and

statistics.

Defining the symbols used in the lecture:

N = Market size C = Constant cost

P = Price WTP ~ f(x)

L = Lower bound U = Upper bound

: WTP – An Example – Extended Edition

So, the previous exercise we did was a very small but illustrative exercise and what we will try to do
right now is generalize this to a more a generalizable case using a few mathematical derivations. If
you are uncomfortable with any of these please take a look at the notes at the link below. So, now
assume that you have done your sampling correctly over large market size and assume that your
market has a size of N. Now, market sizing itself is problem. But at the moment, let's assume that we
know what the market size, we roughly know that, and we also know that we have a constant cost.
So N is the market size, and we know that we have a constant cost C of producing any good that you
have. In real life, this cost C could be variable, but we are just taking a very simple case again and
now willingness to pay, if you've done your sampling correctly comes out, let us assume as some
continuous probability distribution function. Now what does this mean? Let's assume that there is a
lower bound called L and an upper bound called U. So now let's just draw a representative figure on
the Y-axis we will plot f(x), which is the distribution function and X here, let's mark out an L and a U.
So, the reported willingness to pay is between L and U even theoretically and some sort of curve. So
now, if you were to set a price let's call it P then area under the curve on the left would be the
cumulative distribution function which is capital F(P) and recall that this is the area under the curve
or the integral L to P, F(x) dx and what does this mean in our respect? It means the fraction of people
whose willingness to pay is less than P. Very simply put area under this curve here F(P) is a fraction
of people who will never buy, and this portion on the right between P to U, which is 1 - F(p), and you
can take a look at the properties of cumulative distribution function. You will see that this total area
is one, which means this is 1 - F(P) equals integral from P up to U, F(x) dx is the fraction of people
that you're interested in they will buy. So, if you set the P, 1 - F(p) is the fraction of people whose
WTP is greater than or equal to P and they buy. So now the number of buyers is the marketsize
times fraction of the market that will buy at your price P and if you assume that every buyer buys at
most one, now thisis a reasonable assumption for certain kinds of products, for example, durable
goods, TVs, cars, apartments, etcetera, this is a reasonable assumption for something smaller, like
chocolate or pens or something like that. This is not a reasonable assumption anymore, but we will
see under the assumption that each buyer if they buy, buys only one. And using that, what we can
now do is come up with the expression of profit Pie (π), the capital Greek letter π is usually used to
denote profit is the number of buyers which is market size times the fraction of the market that will
buy times the profit margin. So, if you sell it a price P and it cost you C then P minus C is the unit
profit and you multiply it by the number sold, number sold and unit. So usually, the main aim of any
company is to maximize this profit. And you want to manipulate your price to maximize his profit
which is given by the simple condition dπ/dp at some price P* which you have to solve for equals
zero which implies that and if you do the differentiation N is a constant forget about it d/dp (1 - F) is
– f and we will use the product rule here plus 1 - F(p)* and d/dp of P - C is 1. So, this is the condition
to be satisfied or to come across the profit-maximizing price. Now, one interesting thing you will
notice here is that the profit-maximizing price is completely independent of the market size. Now,
this is good news for us because market sizing itself is a complex exercise and assuming that you
have sampled the market well enough to have a representative sample. The profitmaximizing price is
no longer dependent upon the market size and that's a very important result. So, P* is independent
of market size N, remember this, this is a very important result. So, now, does this correspond to any
summary statistic like mean, median, mode, standard deviation, whatever you can think of? Most
likely not. And for most theoretical distributions, you will find that Copyright © IIM Bangalore. All
rights reserved there is no real closed-form solution either. One of the cases in which a closed-form
solution exists is for the uniform distribution, and what we will do is try to solve the following for P*
for a uniform distribution. 1 - F(p*) - (p* - c) *f(p*) = 0. The solution of this usually has to be
numerically approximated, but let's go ahead and see how this expression can be solved for the very
specific case of a uniform distribution. So now we have the profit-maximizing condition is given by
the equation here. Let's assume that willingness to pay for your samples is distributed as a uniform
distribution with upper bound U and lower bound L and now f(x) is a constant that is the property of
the uniform distribution and it is equal to 1 over U - L. And if you set a price P then capital, F(p) is the
area of this rectangle and you could either integrate or very simply it is the area of this rectangle
which is P minus L over U minus L. If you did the integration like we did in the previous step, you will
get exactly the same result. Now, the condition is that you have a cost C. Let's assume your C is
somewhere here and remember, this is very normal. When you survey your participants, often they
will actually tell you willingness to pay, which is lower than your cost. Now, of course, you don't
want to sell to those people, and they will never buy if you price higher than your cost, but don't be
alarmed if this happens. Data a data. C could also be below L, and your cost could also be above U,
there is no problem where C lies in your data or your analysis will go just in the same way. So now if
you were to take this equation and substitute the values of capital F and small f of p*, we are going
to get the following 1 - p* - L over U - L - p* - C times 1 over U - L equals zero, and you need to solve
this equation for p*. So if you were to solve for p* and you could do this yourself, it is a simple
algebra from this point, you will see the p* is the average of the upper bound of the uniform
distribution and cost are simply U + C over 2. It has nothing to do with the lower bound which is kind
of intuitive because people below the cost are not worth selling to anyway. And irrespective of
where your Cost is, the average of your upper bound and cost is your profit-maximizing price here.
Now, this is a somewhat counterintuitive result, especially to those of us who tend to rely on a
wellknown summary statistics like mean, median, and mode etcetera and first intuition to a lot of
people, including most students I teach is to use one of these parameters, but actually in such a
case, it is the solution of this equation that we have here and for the special case of a uniform
distribution, we can see that it has a closed-form solution. As an exercise, do try out the same
exercise for things like the exponential distribution and normal distribution. Most likely, you will not
get a closed-form and easy solution like this. Most likely, what you will have to do is simulate some
numbers and find out where this expression crosses zero and through a method of trial and error or
some sort of numerical solution algorithm is how you will have to arrive at the solution. Real markets
can have millions of people and it is not possible to survey all their willingness’s to pay. These market
research agencies may have to do this exercise with a sample of a few 1000 people. Now, if they
have done the sampling appropriately, the willingness to pay distribution of the sample can be
extrapolated to the entire market. Sampling itself is a vast science with techniques spanning vast
reams of academic and practitioner literature. Think about market research agencies like AC Nielsen
or opinion polling agencies like Gallop, or even the National Sample Survey Organisation. They are
specialized in setting up large surveys where they sample a few 1000 people, assuming that the
sample is somewhat representative of the population. While we will not discuss the nitigrities of
sampling here, it may be good for you to remember as a manager that sample selection matters

o Title: From Problem Definition to Decision Making

Okay, so in the last couple of videos what we did was take this concept of willingness to pay and
then did those exercises about how willingness to pay distributions can be used to price your
product to maximize profit. The question is why did we do that? And is there a structure or a
framework that marketing researcher in general or any researcher can follow to proceed efficiently?
So, most marketing research problems and maybe not most but all, I'll say all marketing research
problems and in fact, I will go ahead and say all research problems would actually start with a
problem definition or a goal in your mind. And whatever you do after that will actually be dictated by
the problem that you have defined or the goal. This seems to be a very trivial point but is actually
probably the most important point or the most important step of marketing research. So, you define
a goal, in the previous case, what we did in the example that we did, the goal was to maximize profit,
and often you will see in business problems, including marketing, profit maximization is usually the
most common goal. It need not always be the only goal though. For example, let's assume that
you're an NGO who wants people to use toilets instead of using open-air defecation or you may
want more kids to join school and solve the problem of literacy. These are not necessarily profit
maximization problems, but still you want to maximize the usage of toilets or you want to maximize
the enrolment in school. So, that could be another kind of problem. To understand these problems,
what you will need to know is who your customers are? And who the people in question are the
people you're dealing with, what do they want, and how are they going to react based on your
decisions? Essentially, this is, what you are looking for in a marketing research problem? There could
be other problems in question as well. For example, you could be a political candidate, and if you
have followed recent political campaigns both in India, US, UK, etcetera, you will see that they have
large research teams that tried to figure out what the electorate wants, that also is a kind of
marketing research problem to give you another example. So, now, once you have defined this
problem, the next thing you will do is look for theory and theory could come either from the
marketing literature itself or you could borrow from Economics, Psychology, Sociology,
Anthropology, etcetera, anything that deals with human behaviour and tried to come out with
relevant theories. So, let's just go back to the example that we spoke about that of maximizing profit
for your so-called made to order cars. In that case, the theory I have used was the theory of
willingness to pay that comesfrom economics and there is a little bit of psychology also involved in it.
Now, I know that I have a cost and I could simply and a mark-up, but the customer has to buy my
product, right? And because I know that I decided to use the theory of willingness to pay. Now, once
you have the relevant theory and this... choosing what the relevant theory is could be a tricky
exercise in itself and might take a lot of time based on what your problem at hand is. But now, once I
have this theory in place, I have to design a study, and thatstudy is going to tell me about what the
customers on the field actually think. Now, I could do this informally or I could do this in a qualitative
method where I could have focused group, I could have interviews, and I could just observe people.
This course itself will not deal with those qualitative methods in detail. We may sometimes talk a
little bit about interviewing focused groups, but what we are going to deal with or what are known
as quantitative techniques. What I mean by quantitative techniques is that we are going to use
numbers in our analysis

So, now when you have to design a study, remember, it's a resource allocation problem, why? Data
collection involves time. It involves using manpower. People have to actually go out there and collect
data and all of this, of course, involves money, and you need to know whether it makes sense for
you to actually do this data collection exercise. If you're a big company, this is a no brainer, you need
to continuously do this. If you're a smaller company you may need to optimize your resources here,
especially if you are a small start-up or a small business that is just started out. Market research or
marketing research is still extremely important to you, but you may need to optimize on the
resources that you can allocate, right? You won't have that many people and you're probably
stopped the funds. So, think about what you could actually do. Just go out and see if somebody has
collected data that suits your purpose, you could have census data, you could have government
agencies collecting data, and giving them out for free if that works for your that is excellent, you
don't need to go any further, you have actually saved a lot of money. On the other hand, you will
find that a lot of data collection hasto be very specific in nature and very specialized in nature to
understand your particular problem. Now, there are companies out there. Market research
companies say a company like AC Nielsen, which is a very big market research company which sells
reports to do exactly this. So, they may have realized that there is a need for a certain kind of data in
the market companies in a certain sector may want a certain kind of data. They may have gone
ahead and collected that data and then they will put these datasets on sale. Now, of course, these
companies are doing it for their own profit, so you will find that there's a price associated with it.
And if this data actually suits your purpose and the price is also within your price range and go ahead
and buy this and analyze that, that way what you would have done is saved some time and some
resources on your data collection itself, you don't need to hire people, you don't need to spend extra
money or time because that data would have come to you ready-made. Now, you'll find that in most
cases or in many cases that even these datasets may not be sufficient. What do you do then? You
could either commission your own study in-house, which means that you hire people who are
qualified. You make themspend time, make them go outinto the market, floatservice orsome other
data collection methods, and they collect your data for you, or you go back to those big companies.
A company like AC Nielsen or there are other companies out there. There's TNS. There is Gartner.
There is Gallup. There's a lot of market research companies and you ask them to collect the data for
you and you pay them. Now, of course, there is a cost-benefit analysis there and you need to make a
prudent decision, whether you need to collect this data yourself or you're okay both from a price
point of view and any other point of view of allowing a third-party to collect your data. Great. So,
once you collect this data, you have to analyze it. So, if you look back at the example of the so-called
Made to Order cars that we discussed, we did not actually discuss this data collection method. We
assume that we had this distribution at hand. Now what we did is we exploited the theory of
willingness to pay, which we know is the maximum price that somebody is going to pay and the
other thing we did is we exploited the property, the statistical distributions. So, you know that the
area under a curve between A and B is the probability that is a fraction of people rather than whose
willingness is to pay are between A and B. We use that to find out the fraction of people whose
willingness to pay will be more than any given price. So, once you do that, you know what fraction of
the population is going to buy your product. And using some basic calculus we found out a profit
maximization or optimizing price. Copyright © IIM Bangalore. All rights reserved So, what we did was
we had a theory, we had a problem in mind, we had theory, we used those and collected data. Of
course, we did not dwell on the data collection, but we still assume that we collected that data and
now we again used a very consistent and theoretically relevant method to analyze that data to come
up with a price. So, go back to your problem. Our problem was profit, maximization and if you see
every step of the data collection was actually guided by this very problem of profit maximization.
Now, are you always going to get the right answer? Is your data collection always going to work?
Maybe not. You may end up getting messy data. You may end up getting results that are not very
useful to you. For example, if you had found a price where the profit-maximizing price was lower
than our cost then we would be sure that we are in a situation that doesn't make any sense. We
might have had to go back to either a product designed, reduce our costs in somewhere or trying to
find out if you have made any errors, maybe there may have been sampling errors, they may have
errorsin designing the surveys. They may have been errorsin the questions asked. They may have
been errorsin also analysing the data. Maybe we did not use the right theoretical framework. We do
not know. What we would do is go back to every step and see if we could redesign any of them so
that we can finally come up with a decision that makes sense. Finally, after all of this, if you see that
your profit-maximization price is still lower than your cost then actually your decision becomes do
not launch this product into the market. At least cut your losses and that itself could be a very
important decision that a company may need to make. Now, the question is how do allocate these
resources if you are a large company. Let'ssay you're a big company, Fortune 500 company, etc.
Typically, they are going to have a large marketing research team that may or may not interferes
with the large market research companies outside like AC Nilsen. Some of the data may be collected
in-house, some of the data collection may be outsourced. If you are a smaller company, then you
need to pinch your penny's a little more and find out what makes sense. Again, in a large company,
market research becomes a continuous exercise. You need to continuously find out how the market
is evolving, how your customers are evolving, maybe tastes are changing over time, etcetera,
etcetera. And typically, you have a large marketing research team, which either supports the core
marketing team or is a part of the core marketing team itself. And this problem definition to decision
making, this entire loop is repeated time and again over and over
Week 02: Indirect Measurements of
Value
Learning Outcome Week 02
By the end of this week, you will be able to:

 Define exaggeration bias

 Make use of auction theory to derive WTP of consumers
 Estimate WTP by using theories like Value in Pricing, Van Westendrop Pricing
Method and Conjoint Analysis

 EXAGGERATION BIAS

 Do you know that human beings are not good at sticking to words?

 What happens is that, generally speaking, revealed preferences (what you actually
did) do not match with stated preference (what you said you would do)!

 Research shows that estimates of the prices customers are willing to pay for new
products or services using responses from survey questionnaires are notoriously
biased on the high side ( Source - Estimating Willingness to Pay with Exaggeration
Bias-Corrected Contingent Valuation Method by Joo Heon Park, Douglas L.
MacLachlan).

 This is exaggeration bias and it becomes difficult to correctly estimate WTP in this
situation.

o Title: Exaggeration Bias

Hello and welcome back. Last week, we talked about the concept of value and how the marketer can
tinker around with the 4 P's to deliver and communicate value to the consumer and hopefully make
a profit in the process. We spoke about willingness to pay or the maximum amount a consumer
would be willing to pay for a certain good or service and how knowledge of this can help you make
better pricing decisions. This week we will talk about a trickier problem. How, if at all, can we
measure willingness to pay? Imagine the following scenario. You're a product manager at Amazon
and have developed a new Ebook reader, The Kindle Voyage. It weighs 188 grams, which is about
13.4% lighter than the previous model, the Kindle Paperwhite. It also allows you to turn pages using
the new page press buttons apart from the touch screen that both the paperwhite and your model
have. The Paperwhite retails at rupees 10,499. How much would you charge for the Kindle Voyage?
Marketing managers and product managers routinely deal with such questions. You can, of course,
look at the cost of introducing the new features, and add a markup to the price of the Voyage. But
how much can you add? And will consumers be willing to pay for these new features? At this point,
you have only gut feeling to guide you. The next thing you could do is find a representative sample of
consumers and ask them the following question. Here is my new product. How much will you be
willing to pay for it? And each person tells a number, it could be 17,000, 21,000, 14,000, 12,000 etc.
What if you were to use these numbers approximate a distribution and do a calculation like we did
last week? Chances are that you would end up massively overpriced in your product, leading to a
failure in the market, why? This is because of a phenomenon called Exaggeration Bias. A long line of
research says that human beings are not good at sticking to the words. What you say is not always
what you do. For example, you may have convinced yourself that you will exercise every day or cut
outsmoking, butin reality, you may shirk your activity orsneak out for an occasionalsmoke.
Remember, an election, where people surveys indicated that nobody would vote for a certain
candidate. But then, you saw him come to power or the time you promised someone you would see
their amateur standup comedy show but ended up skipping it. What happened was that peoples'
revealed preferencesthatis what they actually did, did not match the stated preferences. That is
what they said they would do. In our pricing case our simple question may have caused you a
respondent to simply overestimate her own willingness to pay. She was probably not lying, but
genuinely believed she would pay 17,000 for the new Kindle Voyage. Those are her stated
preferences. In real life, when she actually has to buy the 17,000 products, and it involves an actual
expenditure of money she shirks from the decision. Her reveal preference is not to buy. This is
known as the exaggeration bias and is common in willingness to pay service. So how do we
overcome this? We resort to other indirect methods which have their own shortcomings but are a
little more robust than directly asking.

Title: The Audio Podcast Professor:

Okay, so now let's come to this concept of second price sealed bid auction and this seems a little
counterintuitive to me. So, what I understand is that the highest bidder gets the object whatever is
being auctioned but pays the price that the second-highest bidder bids. Right? This is very counter-
intuitive to me. Can you explain this, why this kind of auction is used?
Mayurakshi: Can I borrow your pen to explain this? Let us assume that this pen your true valuation
of this pen is ten rupees.
Professor: So, which means that my willingness to pay for this pen or WTP is 10 rupees.
Mayurakshi: Yes.
Professor: Which means that even if you sell it to me at 10 rupees and one paise, I should not buy it
up to 10 rupees, I should buy it below 10 rupees also I should buy it.
Mayurakshi: Yes. So, it is the maximum price that you are willing to pay to buy this pen.
Professor: Great. So, I'm willing to pay up to 10 rupees for this pen and but when you are auctioning
it, it is very obvious that I should not pay even one paisa more than 10 rupees. But what I can't
understand is I will be very happy paying 8 rupees or 9 rupees for this pen. I just have to win, right? If
I pay anything up to 10 rupees and win this pen it is kind of like savings or profit for me, why should
not I bid lower than my willingness to pay.
Mayurakshi: So, there are other bidders with you, you do not have complete information. You do
not know what they are bidding? What is their willingness to pay?
Professor: So, if there are 100 other people in the rooms who have all put in a sealed bid, I have no
idea what they are going to bid.
Mayurakshi: So, what is the maximum that you can afford?
Professor: That is 10 rupees.
Mayurakshi: That is 10 rupees. So, you're going to pay 10 rupees, right? So that no one outbids you,
right?
 Professor: I see. So, which means that is the second-highest bid is 9 rupees and I have bid 10 rupees
then I make a profit of saving of one rupee and it could go up to one paisa if somebody bids 9 rupees
99 paise.
Mayurakshi: And you are better off.
Professor: On the other hand, if I lose the bid if I get outbid, it means that somebody would have bid
more than 10 rupees which I didn't want to bid anyway. It would have been loss-making for me. So,
this is where I'm actually kind of economics is forcing me to bid my true willingness to pay, right? So,
now you see, there is actually a very complicated economic theory where you can, mathematically
sure that you are best possible bid in a second price sealed bid auction is your true valuation of any
object and in our terminology, that is your willingness to pay. Now, believe it or not, there is an
economist called Vickrey, who won the Nobel Prize for showing exactly this. And the second price
auction has many other properties that some other auctions don't have. We will not be going to
auction theory here, but the second prize auction is once such auction where the best possible move
for you is to bid your so-called true valuation without strategically increasing or decreasing. So how
do we find out willingness to pay from a second price auction? Very simple. You conduct the second
price auction with a large number of people and then after that, the auctioneer just gives your
distribution of what bids you have made. And if that sample of people participating in the auction is
statistically representative, then very much you can say that it very closely resembles the real
distribution of true willingness to pay.

Auction
AUCTIONED OFF

In the following video, we have conducted an auction. The motive was to find out the
WTP of participants for one kilo of Alphonso mangoes. Hence, we conducted a second
price sealed bid auction. There were 24 participants for this activity.

We would like to thank them all for participating in this activity.

Name Price*/kg *Price is in rupees

Shreya 950
Pooja 500
Mridul 450
Reena 400
Rusha 320
Madhu 300
Sourav 200
Dhivya 200
Moushumi 200
Pavan 180
Dinesh 130
Ranjitha 120
Natasha 120
Girish 120
Honkote 100
Soumya 100
Aishwarya 75
Namratha 60
 Smitha 50
Swetha 50
Rajesh 50
Shobita 40
Harish 30
Kaushik 21

o Title: Conducting an Auction

So, welcome everyone. What we are planning to do here in this MOOC is to figure out how people
value goods or products. And one of these measures is known as something called willingness to
pay, which is the maximum that would pay for a certain good or product. Why this is useful is that it
helps marketers figure out what an optimal price should be in the market. So, the purpose of the
study is purely that. All of you have a pen and a paper. Can I have a show of hands there to home
many of you like mangoes here? All of you do. You don't like mangoes? You do. Okay great. How
many of you like Alphonso Mangoes? Okay. Great. What we are going to do here is have a small
auction of actual mangoes. So, I have a bowl of mangoes here. This is one kilo of top-quality
Alphonso Mangoes. You can check out the quality. Just pass it on. And just take a look and think how
much you would be willing to pay. If you don't like you can... So, I'll tell you what to do. If you don't
like Mangoes also just think of how you would value it? So, it would, of course, depend on whether
you like them or not and try to have a value. This is slightly black here, but I assure you it is pretty
fresh. So, just take a look. So, essentially this entire set of mangoes and I just want to show you this
because the price is here, and it is for sale today. We are going to do what is known as a Second
Price Auction. How many of you know what a second price auction is? You probably know. Would
you quickly like to explain what a second price auction means? Male Speaker: So, everybody will
start bidding, bidding, bidding and when the maximum bid comes, so the next best bid will get the
prize. Professor: That is effectively what it is. I'll just explain to you what you need to do. There is one
small difference here. Everybody will not keep bidding; you cannot revise your bid's. What you have
to do is take that piece of paper, write your name and a rupee value so you are going to buy one kilo
of mango and we are actually going to sell it.

Mayurakshi will collect cash, okay. Put down a number that you are willing to pay for this mango,
for this one kilo of mangoes, the winner will be the person who has put the highest bid and but the
amount you pay will be the second-highest bid. So, I'm not going to bias you. I will give you a
completely different example. Suppose, I was auctioning an apartment and the highest bid is one
crore and the second-highest bid is 90 lakh rupees, the person who has bid one crore will be in the
apartment, but pay 90-lakh rupee, which is the second. Why this is actually a pretty famous theory
that got a Nobel Prize by the way to a person called Vickery. It essentially says that with this kind of
incentive structure, it makes sense for you to bid your true valuation. So that is an additional to it. If
you have value it at rupees X and bring that X that actually is your game theoretically best bid
because you end up paying slightly less than that. That's all I have to say. You have seen the
mangoes. What I would like you to do is write down your name. So, we are interested in the names
only for purpose of actually giving out the mangoes. We do not require your names in any other
way. In my... as the marketing researcher why I am interested in this is I will get a distribution of
what each of you is willing to pay essentially so I'd like to do some statistical inference from that. So,
all you have to do right now is write down your name and a bid in rupees and the highest bidder gets
these mangoes for the second-highest bid. Please do not look at each other. Please bid completely
independently. I will collect the mangoes. If you're done, you can just handover the slips. Just fold it.
So that nobody else sees it and hand it over to me, done. So, before we actually find out who has
won and what they will pay, can I have a volunteer who would like to tell me what they bid and
why? Introduce yourself first and please talk into the mic.

Professor: What is your name?

Kaushik: I'm Kaushik.

Professor: Kaushik, what do you do Kaushik?

Kaushik: I'm a marketing intern over here at IIM Bangalore.

Professor: And how much did you bid?

Kaushik: 21 rupees.

Professor: 21 rupees for a kilo of mangoes.

Kaushik: Yeah, because I don't like mangoes.

Professor: You don't like mangoes. So, but why 21 rupees why not 0 rupees.

Kaushik: I thought it is an Alphonso.

Professor: 21 for a kilo of mangoes. Great. Anyone else here? Let's go. Yeah. What is your name?
Natasha: Natasha.

Professor: Could you use the mic, please?

Natasha: I'm Natasha.

Professor: Yeah, what did you bid?

Natasha: I bid 120 rupees

Professor: Why?

Natasha: This is a local Alphonso. It is not an authentic one, which would cost a lot more and I was
recently in the markets seeking local Alphonso.

Professor: So, you have some sort of reference price that you are used to pay. How do you know it is
a local Alphonso? Are you an expert in Mangoes? Are you from Ratnagiri?

Natasha: I'm from Pune. I have spent a lot of time in Bombay and Pune. Mango is all of my life.
Professor: So, these were not the best Ratnagiri mangoes, right? Okay. What is your name?

Roosha: My name is Roosha and I think I'm bid it at an absurd amount.

Professor: How much?.

Roosha: 300 Per Kg. Because I thought mangoes were expensive. Professor: Okay. Great and do you
like Alphonso mangoes?

Roosha: I like mangoes, but the mangoes that we got was actually pretty ugly so...

Professor: Yeah there was a slight black spot on it. I know. I know. Okay. Great. So, as you see there
are different people who have different likings of mangoes. Some people have some expertise. Some
people may not. And based on your own valuations, you can come up with different prices. What
can also happen is that if you used to buy these Alphonso mangoes, you might have a reference
price, as one of our respondents there and based on that we have different bids here. Now
Mayurakshi, can I ask you to come forward? Would you like to... I don't think... At this point, I don't
think we will go through everybody's bid. I won't say X has bid Y that is unnecessary. What I would
just like you to do is look at all the bids and let's take a look at the top 3, of course, the top bid is
going to win and the second-highest amount that is going to be paid. So, there is Madhu who has bid
300 and there is Shreya has bid 950 would like to share your valuation.

Shreya: So, with Usha here, so I just thought mangoes were a whole lot more expensive than they
are and I thought that Alphonso mangoes are the most expensive one regardless of quality intended.
Professor: Fair enough. Yes.

Shreya: So, I just thought the realistic market price is somewhere in high 800's.

Professor: And you bid 950. Great. So, this is actually very nice because see you have from
extremely low prices and extremely high prices and these are actually characteristics of real
auctions. So, one of the things that happens, and this is actually something that happened in real
auctions. So, what we as marketers are interested in is his entire distribution and we can do some
statistical inference. So, there is Mridul who has bid for 450, which is the third highest. Who is
Mridul? Okay. Professor: What was your logic? Let find that as well.

Mridul: I easily don't go to the market. I just know about the prices. I thought Alphonso mangoes
like the highest price. Professor: Okay. Great. See we have different kind of heuristics, evaluations,
and et cetera which was exactly what I wanted to bring out in this class. So, congrats Shreya, you win
this kilo of mangoes and you pay Pooja's price which is 500 rupees. Can anyone tell me... so this was
bought at Big Bazaar today. Can anyone tell me what is the MRP of this box is? 70, any other
guesses? 120. Any other guesses? We are playing prices, right? So, the most correct answer is not
going to get it. She is still going to get the mangoes. 150, 200 is the right. It is 199 and I have a special
surprise for you. We are not really interested in making money. So, you will get these mangoes for
free. We just wanted you to believe for the purpose of this exercise we wanted you to believe that
you are paying, but our intention was not to sell these mangoes. So, we will just pack this up and
these are yours to do whatever you would like with them. So, if we had told you that you don't have
to pay then you can enter any amount, right? And the highest is you are going to bid. That wouldn't
have worked out. So congrats, your time has been well spent here and sometimes the winners' curse
is not really a curse and that is all we have for today. Thank you. Thank you everyone

Value in Use Pricing - The Bulb Problem

Title: Value in Use Pricing
Consider this scenario, your company manufactures advanced light bulbs that need to be replaced
once every two years, while standard bulbs are replaced once in six months. Standard bulbs
consumed 60 watts, while yours provide the same illumination by consuming 20 watts of power.
Assuming that the average household keeps the bulb on for 10 hours a day and the standard bulbs
cost 50 rupees. What is the maximum you can price your bulb, given these data? Assume that
electricity is charged at a flat rate of one paisa per watt-hour? Also, assume that the cost of
manufacturing your bulb itself is 50 rupees per bulb. How will you set the actual price of your bulb?
So, the problem that you see is something that a lot of new technology or product companies face.
So you have a superior product, and in this case, it is superior because it consumes less energy and it
has a higher life, and you hope that it has a higher life than the current light bulbs that you have in
the market, and you hope that the consumer will thus pay a price premium or in other words, what
we can say is that the willingness to pay for your so-called superior product should be higher for the
consumer just because you have these features. So, let's assume that this bulb cost equal to 50
rupees per unit. So, it's a new product and you know that the other bulbs are being sold at 50 rupees
now because you have done R&D and the features itself, the materials are more expensive. This
costs you 50 rupees. The first way you could price it is just add a mark-up something that the
industry normally uses. Let's assume a 30% mark-up. So, one way you could do it is price is 1.3, that
is the 30% mark-up times 50 which works out to 65 rupees. Simplest way, and let's call this price
one. Can we do better? Now, you may observe that this bulb has a higher life and let's just take that
into account. Can we actually show that a rational consumer someone who wants to save on costs
should be theoretically willing to pay more than 65 rupees? So, you could add a 50% mark up and
make that 75 or whatever, but essentially, what you are doing is basing it on your cost. So, this is
called cost-based pricing. So, what we'll do is if you take the ordinary bulb it has a life of about half-
a-year versus the two years of your superior bulb. So, if you take a look at over two years, one new
bulb lasts the equivalent of four old bulbs. And if you do a simple calculation there 50 rupees for the
old bulb. So, you could also have price equals 200 rupees. So, if I'm a rational consumer and we are
going to forget here things like discount rates or interest rates etcetera that would make the
calculation in little more complicated. Let's just assume that you don't get any interest in the bank.
You know that bank interest rates are falling anywhere, right? So over two years, and for such a
small amount you don't really care about the interest. One bulb is equivalent to four bulbs. One new
bulb is equivalent to four bulbs so you can charge a price of 200 rupees. You already see from 65
rupees that is a huge jump. More than a 100% jump if you take a look at it. Can you do even better?
Suppose you were to use more information. Let's say you take the high life as well as the cost of
energy consumption rate. In our problem, we have assumed that the bulbs are switch on for 10
hours a day and each watt-hour that's a unit of energy. Watt's the units of powers, Watt power times
time is energy and Watt-hours are the units of energy, where electricity is built, and we assumed
that one Watt-hour is charged at one paisa. I don't know if that is real or not, you can take a look at
your electricity bill, but let's just assume that for this particular problem. Copyright © IIM Bangalore.
All rights reserved So now what we are going to do is assume that the maximum price you can give
to your bulb is P and now we are going to try and solve a very simple equation. What we are going to
do is compare the cost of ownership. I'm going to call this, and I'm going to write down the cost of
ownership of a new bulb and try to see at what point me as a consumer would be indifferent and by
that I mean equal to the cost of ownership of an old bulb. First thing I need to do is select an
appropriate timeframe. Now, you know that the new bulb lasts for two years. The old bulb lasts for 6
months, that's half a year, so we take essentially, the shortest period for which the new bulb can be
considered. Essentially, we take two years and the price of one bulb, here what I will do is take the
price of four bulbs. So, that's is 200 rupees plus now what I'm going to do is on both sides, you know
that the old bulb consumes more energy 60 watts more power rather 60 watts while the new bulb
consumes 20 watts. So now 20 watts times 10 hours a day, and in two years we have 365 times 2,
that is 730 days times the price of each watt-hour. So, watt-hour is the unit at which current is built,
your electricity is built, so that would be one paisa or 0.01 rupees per watt-hour. The same thing
here except 10 hours, except that is 60 times 10-hour per day times 730 days. So essentially, what
I'm doing is over a period of two years I'm considering the cost of ownership which is the cost of the
bulb plus the cost of power times 0.1 rupees per watt-hour which is a unit? So, if you do the
calculation and I'll leave you to do the multiplication, you get P equals 3,120 rupees, and this is price
point three. So, you have this which you see is way, way, way higher, more than a factor of 10 higher
than what he would get from competitive pricing, which is again more than 100% higher than what
you would get from cost-based pricing. So, method one, you just put a mark-up on your costs.
Method two, you consider only the life, and method three, you actually consider the fact that it
actually costs less. So, if I were a price-conscious consumer who did the calculations in a methodical
way, I should be indifferent between paying 50 rupees for the old bulb versus 3,120 for the new
technology. Now just pause and think for a moment, if I were the company that were launching this
bulb. Should I price it at 3,120 rupees? What do you think will happen if I go to market and say that
my new technology is great? It's really great. Look, less energy, much higher life. So, I will charge you
3,120 rupees? What do you think would happen? Without doing much analysis, I can tell you that
your product will be a complete failure. Why is that? First of all, I am indifferent rate so and second
of all there are other factors in consideration. So, you need to pass on some savings to me. So, this is
theoretically the maximum you could go based on these calculations, but I, as a consumer, have a lot
of choice in the market. So, what do I do? I can still get a bulb for 50 rupees, but clearly, you can't
sell it at 50 rupees. But if I have some sort of rationality within me, I should be willing to settle for a
lower price and you pass on some of the savings to me, so let's say 2,000 rupees, so let's say 1,500
rupees et cetera. But there's a bigger problem here. I am not really sure about whether you actually
have a high life, whether you have actually have less energy. So, remember one thing as a marketer
that you need to do is convince me, you need to communicate value. What you have done so far is
actually a create value by giving me a superior product, but you haven't still communicated the
value. So just take a look at some of the ads that you get for news bulbs LED bulbs not CFL bulbs
sorry, LED bulbs or power-saving refrigerators or maybe more fuelefficient cars, solar energy panels,
etcetera. They are usually much more expensive than conventional Copyright © IIM Bangalore. All
rights reserved alternatives that you have. And if you see those ads, they tell you, it look we are
expensive, but we are passing on some savings to you. What could still happen is that even let us say
you charge a 500 for this. Let's say there is no right answer here, but you discuss with people after all
these three price points, you could go a little more than 200 let's say you want to charge 500 rupees,
otherwise, it's not just economical to you for some reason and the consumer still doesn't buy. What
do you do? A) Either you don't really put it out on the market that have been products that failed or
B) In this case, what you see is that this is eco-friendly. You consume less power, which means
there's less load on the grid and India is a power-starved country, et cetera. So you go to the
government and ask the government if they would be willing to either give a tax up or sometimes
you can actually see that these bulbs, this LED bulbs, which save a lot of energy and have a higher
life are now available at your electricity board, at least in Karnataka. They sell it for I think lower than
200 rupees now. You just show your electricity bill, and you can buy these bulbs at 200 rupees or
less. So, what's happening, basically, is that the government sees that there is some sort of public
good or some sort of public welfare that can be used, you are saving powers so that's good for the
environment as well as reduces a load on the government. So, they pump in some money. Similarly,
you can see that with electric vehicles. They are actually very expensive to build. But then you
actually end up using much lower petrol or you don't use petrol at all. So, the government has some
incentive there in subsidizing those or solar panels if you see now, the solar industry is booming. One
of the reasons is that solar panels are actually very expensive to install. But with some sort of
government subsidy or tax breaks, etcetera, they actually become economically viable. So, this
method that we have here is called Value in Use or VIU pricing. Now, if you are not a consumer if I
were not selling this to a consumer, but let's say to an airport manager, who is actually going to run
these numbers much more carefully than me the average consumer that an airport manager has to
take care of hundreds of flights, right? You know the Airport. So, it's not 10 hours anymore. The
lights are on for 24 hours. So, actually, the energy consumption is going to be a bigger factor. It is
very expensive to replace lights because it is not just buying those lights but actually getting
manpower to come and occasionally maintain them. So, if you were in a B2B context, this makes
more sense. So, things like industrial equipment, things like MRI machines in hospitals, etc., where
there are large amounts of stake and there are managers who are actually running the numbers
more carefully or places where value in use pricing is used a lot more. If you were looking at
consumer products like in this case for example, you can't really jack up the price to 3,120 rupees
and there would be other cases where people don't really care about the price. We're going to take
a look at another method, which is not based on willingness to pay, called the Van Westendorp
Method, where you also take a look at other aspects where people are not that price conscious, but
would still like something of good quality.

Van Westendorp Pricing - Setting the

Background
So far we have covered two pricing method based on the concept of WTP. They are
auction and value in use pricing.

Van Westendorp pricing is another method which is very popular among marketing
researchers. Peter Van Westendorp is a Dutch economist, who suggested this method.
This is a practical method based on a survey which asks four questions to a large
number of respondents.

The questions are:

At what price would you consider the product to be so expensive that you would not
consider buying it? (Too expensive)
At what price would you consider the product to be priced so low that you would feel
the quality couldn’t be very good? (Too cheap)
At what price would you consider the product starting to get expensive, so that it is
not out of the question, but you would have to give some thought to buying it?
(Slightly Expensive)
At what price would you consider the product to be a bargain—a great buy for the
money? (Slightly Cheap)

We have used a survey software called qualtrics. It is a paid software that we use at
Indian Institute of Management Bangalore. You can also use free software like google
forms. Essentially the idea is that you need to host a web survey to ask a few
questions and get the responses in the form of an excel sheet.
 VAN WEATENDROP PRICING – SETTING THE PRICING THE
BACKGROUND
So, imagine this following scenario and this is the question that we asked our survey respondents.
It's a very standard holiday package for two to Goa departure from Bangalore. For those who are not
from India, Goa is a very popular tourist destination about 600 kilometres from Bangalore, and it is
more convenient to fly there because road travel takes about 10 hours by bus. So, the package
includes airport transfers in Bangalore and Goa. It includes economy class return airfare and two
nights and three days, including threemeals every day by the beach at a hotel in Calangute.
Calangute is a popular beach town in Goa, and the travel agent also does sightseeing by air-
conditioned bus and a little river cruise, very stranded package and all of us who wantsuch packages
for the weekend. So now one way to price this would be if you were a travel agent to actually look at
your costs. So, I'm just putting out some hypothetical costs here. There are 4 taxi rides and
Bangalore Airport is pretty far from the city. So, let's assume that it costs you 2,000 rupees to
arrange this. You can get your air tickets in bulk because you're doing group tours and Goa is not
that far away. So, let's assume 3,000 rupees per person, return ticket. So, you pay about 6,000
rupeesfor your tickets. Another 6,000 rupees for two nights in a hotel, three-star hotels and that
includes food. These are very typical prices in India, especially Goa, and other expenses like
sightseeing and cruise, etc., are let's say about 1000 rupees and other expenses will be borne by the
guests separately. So, this comes out about 15,000 rupees. If you had a 30% mark-up on this andd if
you charge a mark-upof 50%, which is pretty good in thisindustry, of course, taxes extra et cetera. So
now what we will try to do is come up with a range of prices that are independent of costs and
completely based on consumers' preferences. So, it's essentially one way is to look at your own costs
and see if consumers are willing to pay what you were paying. So, at this point, what we will do is
look at the four questions that we ask here. The first question here is what price in rupees you think
would be too expensive for this package? And now, as a consumer, when you look at this question
what you would try to see it go to the upper limit of what you think would be an expensive package.
So, the kind of calculations that we did is also the kind of calculation that a typical consumer is going
to do. They have an idea of what hotels cost. What airfare cost, you can easily look that up on the
web. You know what taxi's charge, etcetera. And what you're asking the consumer to do here is to
go to an upper limit. Not exactly willingnessto pay, in fact, a little more than willingnessto pay. So
probably if I charge 50,000 rupees for this to some consumers, it would be too expensive. And that's
what this question is asking. Look at the second question, and this is very interesting. What price you
think would be too cheap for this? This is very interesting because in marketing we have a concept
called the so-called Price Quality Correlation. What this means is that... and you can think about this
in your own experience. A lot of people tend to think that more expensive is better and less
expenses is inferior quality. So, while this may or may not necessarily be true, human beingstend to
think this way. And if you found this package for 5,000 rupeeslet'ssay then you would tend to
getsuspicious. Either you're being ripped off or there could be hidden costs or maybe the quality of
the package is not going to be good at all. And that is the price that we are asking you to imagine.
What price would be too cheap? MK150x – Quantitative Marketing Research Professor Prithwiraj
Mukherjee VW Pricing – Setting the Background © All Rights Reserved. This document has been
authored by Professor Prithwiraj Mukherjee and is permitted for use only within the course
"Quantitative Marketing Research" delivered in the online course format by IIM Bangalore. No part
of this document, including any logo, data, illustrations, pictures, scripts, may be reproduced, or
stored in a retrieval system or transmitted in any form or by any means – electronic, mechanical,
photocopying, recording or otherwise – without the prior permission of the author. The third
question and the fourth question are what prices would be slightly expensive and slightly cheap? So,
what we're trying to do here is first we asked what would be too expensive, slightly lower price than
that or a bit lower than that would be the prices at which you are willing to go up to. So, this is not
exactly willingness to pay, but some sort of approximation. So, we asked what price would be too
expensive and what price would be too cheap where you would get really suspicious but at what
price would you get moderately suspicious. So, if you have run this calculation similar to me, you
might think that 12,000 or 13,000 rupees, it is still suspicious, but not as suspicious as, say, a 5,000-
rupee price. And this question essentially asks that. At what do you think it would be? You will be
slightly suspicious or slightly cheap. And with these four questions, we collect multiple data points
from multiple respondents. So, what I will do next is show you how to analyze this and come up with
the Van Westendorp chart, which will help you decide price ranges

Too expensive Too Slightly expensive Slightly cheap

cheap
30000 20000 28000 22000
59999 19999 49999 29999
100000 20000 75000 30000
2000 3000 5000 5000
5000 1600 3000 1900
30000 15000 20000 12000
30000 10000 20000 15000
2500 2500 3500 1000
50000 20000 45000 30000
30000 5000 25000 8000
75000 20000 120000 40000
30000 15000 35000 10000
1100 500 2500 3000
20000 5000 15000 8000
1100 1000 5000 2000
50000 15000 35000 20000
60000 15000 45000 20000
100000 30000 70000 40000
34000 4000 2495 1500
16999 9999 13999 7999
20000 6000 18000 12000
100000 30000 61000 48500
100000 40000 80000 70000
20000 7000 15000 9000
13000 7000 11000 8000
25000 10000 30000 20000
100000 10000 75000 30000
100000 45000
60000 15000 40000 18000
50000 15000 40000 20000
60000 20000 50000 30000
20000 10000 15000 12000
20000 12000 18000 14000
o Title: Van Westendorp Pricing –
Cleaning the Data So, we circulated this survey by email amongst members of the IIMB community
and we got 33 responses. Now that's a small number, but it's enough to show you how to analyze
the data, of course, there are additional issues of sampling and we are by no means a representative
sample. But for the sake of demonstration, I will just assume that this sample is representative and
work from here. So, you can see here that there are four columns to the excel sheet called too
expensive, too cheap, slightly expensive, and slightly cheap. These are the four price pointsthat have
been entered by people there. Now, let us use this opportunity to also have a discussion about how
we handle data, not just because data are there we do not necessarily use them because we also
need to validate our data. So how would you validate data here? If you take a look at this column
right here, this row right here, you see, there are two blanks. What would you do if there's a blank, is
that valid? Clearly, this person has not paid attention and has filled in only two. This is often a
problem with survey research, and we need to deal with this in some way or what we are going to
do is omit this, but is that enough? We also need to omit a few more if necessary. So, remember,
how we formalize the four questions, too expensive, slightly expensive, slightly cheap, and too
cheap. So, what should be satisfied is the price, which is too expensive, should be greater than the
price, which is slightly expensive, should be greater than the slightly expensive price should be later
than price, which is slightly cheap, which in turn should be greater than the price, which is too
cheap. You see why, right? Because most expensive or the too expensive price should be the highest
and the too cheap price should be the lowest. This is how we formulate a problem. Now does
everybody actually pay enough attention and enter numbers which satisfy this condition? This is the
first thing that we are going to check. So, now I am going to check if each line is valid or not, by using
the and function, what the and function does is take a series of logical conditions. In this case, I am
going to check if the conditions that I had written down are true or not. So, I'm going to put down
equal to and now I'm going to put down three conditions. The first one is too expensive or A greater
than slightly expensive, this should hold along with slightly expensive or C greater than slightly
cheap, along with the condition that slightly cheap should be greater than too cheap, which is B. And
you can see that in the first case, the data input is valid because those conditions are met. I want to
drag with formula down across all the 33 rows that I have and see how many are false. So, I have a
couple more here. So, you can see a lot of them are false. And let's just take a look at it though excel
is automated, just take a look at why this is false. So, let's take a look at too expensive. You can see
that too expensive is lesser than too cheap, right? This doesn't make sense. What that just means is
that this person who has filled up this survey was not paying attention. So, what I am going to do is
delete this. So, this data is not usable for me, I have deleted. Similarly, I am going to delete all the
false's and you can check another false to see here. Why is this person false? Too expensive is
30,000, but let's take a look too cheap is 15,000 that's fine, but slightly expensive is 20,000. So, what
you see is that slightly cheap... if you compare slightly cheap and too cheap, too cheap is 15,000
which is higher than slightly cheap 12,000. So, one of those conditions is not met which is why this
person also gets rejected. You can check for the others as well. Copyright © IIM Bangalore. All rights
reserved I hope you're working this out along with me. I will delete all the false's that also can be
automated. But I am not that good at excel which means I will not do it. And this is a small enough
data set for me to just manually delete. So I have deleted all the false's and now you can see I have
only 22 left out of 33 that I initially had, which means actually 11 people responses were deleted just
because I think that their response are invalid and I assume that's because they were entering
nonsensical numbers or you will find this to be a feature of most survey research that you do. So, a
good idea is always to take a look at your data and see if your data are as you expect them to be.
Now you see, I have all true here so I have no use for this column anymore and I will delete this. All
are true here. So that column was just to make sure that my data are okay

Van Westendorp Pricing - Adding the CDF

VAN WESTENDORP PRICING - ADDING THE CDF

Now, what is required is to plot prices against Cumulative Distribution Function (CDF).

A CDF shows what percentage or what fraction of people think a certain price is too
expensive or what percentage of people below a certain price think it's too expensive.
Too
Too expensive cheap Slightly expensive Slightly cheap CDF 1-CDF
0 0 0 0 0 0 0.00 1.00
5000 1600 3000 1900 1 1 0.05 0.95
13000 5000 11000 8000 1 2 0.09 0.91
20000 5000 15000 8000 1 3 0.14 0.86
20000 6000 15000 8000 1 4 0.18 0.82
20000 7000 15000 9000 1 5 0.23 0.77
20000 7000 18000 12000 1 6 0.27 0.73
20000 10000 18000 12000 1 7 0.32 0.68
30000 10000 20000 14000 1 8 0.36 0.64
30000 10000 25000 15000 1 9 0.41 0.59
30000 12000 28000 18000 1 10 0.45 0.55
50000 15000 35000 20000 1 11 0.50 0.50
50000 15000 40000 20000 1 12 0.55 0.45
50000 15000 40000 20000 1 13 0.59 0.41
59999 15000 45000 22000 1 14 0.64 0.36
60000 19999 45000 29999 1 15 0.68 0.32
60000 20000 49999 30000 1 16 0.73 0.27
60000 20000 50000 30000 1 17 0.77 0.23
100000 20000 61000 30000 1 18 0.82 0.18
100000 20000 70000 30000 1 19 0.86 0.14
100000 30000 75000 40000 1 20 0.91 0.09
100000 30000 75000 48500 1 21 0.95 0.05
100000 40000 80000 70000 1 22 1.00 0.00

Van Westendorp Pricing - Adding the CDF What the Van

Westerndorp method requires now is to plot prices versus a cumulative distribution function. So,
what we first need to do is take each column separately and sort them. Sort them from lowest to
highest. And I'm just going to do that with Excel. Pretty easy to do. There's a button here. What it will
ask you is that there is more data would like to use that, I don't. Because I want to sort them
independent of each other. Each one will be sorted, lowest to highest. So, you see now, it has gone
from lowest to highest, and I will do this first each of my four columns. So now we have four
columns, which are completely sorted out. The reason why we're doing this is remember that the
CDF actually takes numbers in increasing order and see is what proportion is lower. And to exploit
that, what we will do is we will assume that the distribution starts at zero, so prices cannot go below
zero. So, what I'm going to do is insert a row right here and put a zero. So, the assumption be that
the distribution starts at zero or the CDF of zero is zero. The next step I am and going to do and this
is again if you understand the properties of a CDF is something you'll understand better. I am going
to put a zero here and a one. So, thisisjust a temporary column that helps me calculate approximate
a CDF. So now, take a look at what I'm going to do here again the first row I'm going to put it as zero
in the second row and we are going to also upload this worksheet in its finished form. So, you can
just take a look at the formula. So, what I'm going to do is essentially cumulate over the previous
cumulative sum and the current. So, what I will do here is take the sum of the previous row and add
an extra one. And if I keep doing this, you will see that the number gradually increases, and this is
essentially cumulative frequencies. So, what I am going to do is add a column called Cumulative
Distribution Function or CDF. A CDF is essentially what I want to do is find out what percentage or
what fraction of people think a certain price is too expensive or what percentage of people below a
certain price think it's too expensive. And that's why I have added the zero there. So, what I'm just
going to do is take this number and divide it by the total. The total has to remain same. So, in excel
terminology, we will put the dollar sign across the F in the denominators. And, of course, fraction of
people think that zero is too expensive and if you see any price point, just take a look at too
expensive and slightly expensive, we are actually going to the CDF for these two, this column and
this column. Again 5% of people think that 3000 is slightly expensive. Now, further two other
columns, cheap and slightly cheap, you want to see how many people think... so cheap is the
opposite of expensive. So, you want to see how many people think it's too cheap or higher. So, what
you have to do here is actually take the complementary CDF or one minus CDF. And if we calculate
this you just do one minus CDF, of course, you see 100% of people think that zero is too cheap. 95%
of people think that 1600 is too cheap, et cetera, and you calculate the same thing here. So, CDF
goes from 0 to 1 and one minus CDF or the complimentary CDF goes from 1 to 0

Van Westendorp Pricing - Final Analysis

VAN WESTENDORP PRICING - FINAL ANALYSIS

This method involves plotting 4 scatted plots corresponding to the 4 data points (Too
Expensive, Slightly Expensive, Slightly Cheap, Too Cheap) for each respondent.

There are 4 points of intersection. They are:

 Point of marginal cheapness

 Point of marginal expensiveness
 Optimum price point
 Indifference price point

This gives us a range of acceptable prices, as well as a suggested price (optimum

price) and an indifferent price. This is a popular method as it gives a fairly narrow
range of prices.
Title: Van Westendorp Pricing –
Final Analysis So, we are going to insert 4 plots here. This is what Van Westerndorp suggests. It's not based on
any theory, but it helps in suggesting prices. So, just follow me here. What I'm going to do is go to insert, insert
a chart, and that chart will be a scatter plot with a straight line and this version of excel that I have on a Mac
Book here. Inserts some series by default. These are not what I want. I want to insert my own scatter plots,
what I'm going to go to do is select data and remove everything that has been added by default. The first thing
I am going to insert is the CDF of prices that are considered too expensive. So, I want to add the data call it too
expensive. The X values will be the sorted prices of too expensive and the Y values will be the CDF. This is my
first series that I add, and you can see a jagged line sloping upwards. The same thing I'm going to do for slightly
expensive. I will also need to add it's CDF. So, I follow the same process. I call it slightly expensive. I plot its X
values as the sorted prices of slightly expensive and it's Y values as the CDF. Now, I have two plots, as you can
see. Too expensive and slightly expensive. Now I need to plot the reverse CDF of too cheap because here what
I'm interested in the expensive and slightly expensive cases is for a certain price, what percentage below
people think that it is too expensive or less. For too cheap, I want to do exactly the opposite. So, what I'm
going to do is add a series called too cheap, its name is going to be too cheap and the series X values are still
going to be the sorted too cheap prices. The Y values, however, will be the complementary CDF and this is too
cheap. So, you see a line here, a jagged line. It is sloping downwards. The same thing what we're going to do is
for slightly cheap. So, we will call it slightly cheap. We have its X values as the slightly cheap prices sorted in
ascending order and the complementary CDF or one minus CDF is what we're going to plot here. Now you see
four lines here, and this is essentially the Van Westerndorp chart. I'm just going to increase this chart for our
benefit so that we can all see it clearly. You see four series. The X-axis has prices and the Y-axis has fractions
from 0 to 1. So, excel gives you a default of 1.2. We don't really need that because CDF, one minus CDF both
are between zero and one. So, this is just beautification, and this is a Van Westerndorp plot that you see on
your screen. So now what you're going to see are four points of intersection. There is a point here. There's a
point here, there is a point here, and there's a point here. Now, these four points are very important according
to Van Westerndorp. This point, if you were to extend it down it's about... I'm guessing it's about 25 to 27,000.
We could, of course, measure this exactly is what Van Westerndorp calls point of marginal expensiveness. And
the other side, this right here, this point is what Van Westerndorp, and this is the intersection of too cheap and
marginally expensive is what Van Westerndorp calls point of marginal cheapness. The point above that you see
is called the point of indifference and it is the intersection of slightly expensive and slightly cheap. And this is
called the indifference point, or this is Van Westerndorp says is the average willingness to pay. Now, whether
this is right or wrong, it seems to work, but again, there is no theoretical basis. And what he suggests is this
bottom point here in this so-called diamond shape is going to be your optimal price. And this optimal price, in
this case, is about 20,000 actually it is exactly 20,000 rupees, and this is what the Van Westerndorp methods
suggest. Copyright © IIM Bangalore. All rights reserved So, the point of marginal cheapness and the point of
marginal expensiveness are the two lower and upper limit that Van Westerndorp suggests. So here it is, about
18,000 to 27,000. That is an acceptable price range according to Van Westerndorp. About 22,000 rupees is the
indifference point and according to this data and the optimal price according to this is... if you extend this point
downwards and this is about 20,000 rupees and this is exactly what we calculated or other way. This price
range that you see here is the acceptable price range. So, between marginal cheapness to marginal
expensiveness is what Van Westerndorp says is the acceptable price range and what practitioners and
researchers suggest is you could use the Van Westerndorp method as one guideline and then checkout
different price range.

Conjoint Analysis
CONJOINT ANALYSIS

Conjoint stands for Considered Jointly.

Conjoint is a survey-based research method. It is one of the most widely used

technique in market research. The decision to buy a product or not is based on the
product features and price points from the perspective of a customer. This method
helps in understanding the same by breaking down products into their characteristics.

In the following video, we have taken the example of a laptop and analysed its three
characteristics such as price, RAM and hard disk to get an estimate of the number of
configurations that can be designed. This helps the product designer to design the
most optimal product that is profitable to him (her) as well as acceptable to you.
Title: Conjoint Analysis - Laptop Example Thus far we spoke about measuring willingness to pay for a
product that already exists. But how do we design a product that would appeal to customers? In this
session, we will discuss a powerful tool called Conjoint Analysisthat allows usjust to do this with a
series ofsurvey questions. So, let us begin. Imagine that you are buying a computer. How would you
make a choice? Maybe you are attracted to certain brands, maybe you are looking for a set of
features and prices and all of these simultaneously or jointly play a role in your decision-making
process. Consider the following three laptop models and which of these do you like the most. So, the
first option priced at 38,000 rupees has 4 GB Ram and 1000 GB Hard disc. The second one priced at
40,000 rupees 8 GB Ram and 500 GB Hard disk. The third one, 35,000 rupees priced and 4 GB Ram
with 1200 GB Hard disk. Give each a score between 1 and 10. How did you make a decision? Now,
some of you may give higher important to RAM where someone else might be more price sensitive.
As you can see there are trade-offs between these product attributes in these hypothetical
scenarios. Based on your so-called internal value system, you may disproportionately rate one of
these higher than another based on which attribute price, RAM, or Hard disk you value more. A
market researcher with the right questions can extract your value system out without explicitly
asking you which attribute you value more. This is not magic or sorcery, but simple psychology and
basic statistics that play. The basic idea of Conjoint Analysis is that a product is valued as a bundle of
attributes. Thus, if a computer has only three attributes, and this is purely for the sake of simplicity,
in reality, tens of attributes could define a product. Let's assume that this price, RAM and Hard disk,
then we assume that the value of a computer is value of price or the amount of money you pay, plus
the value of RAM, plus the value of Hard disk. Now you can, of course, see that as price increases,
the value of prices is going to decrease, they are going to be less satisfied with the product that costs
more, keeping everything else equal. On the other hand, RAM and Hard disk, if you keep everything
else equal and increase them the value should increase. Now let's assume that as a marketer for
Dell, you have this option of three price points. 35,000, 38,000, and 40,000. Three levels of RAM - 4,
8, and 12 GB and three levels of Hard disk 500, 1000, and 1500 GB. How many configurations can
you design, and which is the best configuration? Pause this video for a while and try to work out
each? The answer to question one is obviously there are three levels of each attribute. So, three
times three times three that is 27, while to question two you would prefer the lowest price, which is
35,000 rupees. Highest RAM for that price, which is 12 GB and 1500 GB Hard disk, which is also the
highest memory. So, the best levels of each essentially. So, why should Dell not offer this to you?
Pause this video and think for a while. Of course, cost and profit, right? RAM, memory etcetera they
cost money, and it may not be smart for Dell to give this to you at the lowest price. They also want
to make profits, obviously. Now, Conjoint Analysis will come in handy in helping this product
designer, design the most optimal product that is profitable to him as well as acceptable to you. So,
on one side you have the seller who wants to make profit, on the other side, you have the customer
who is willing to pay that price. Copyright © IIM Bangalore. All rights reserved Let's go back to the
laptop example and the simplest possible version of Conjoint Analysis. This would involve me asking
you to rate 27 hypothetical computer combinations and then make inferences about which
attributes you prefer, essentially by solving a set of simultaneous equations. But now imagine and
design problem with 30 attributes and 50 possible values for each. The same problem now boils
down to 50 times 50 times 50 up to 30 times. That's a huge number. This is an impossibly hard
number of questions to design, let alone expect a single human respondent to answer. In fact, try
answering a question, a survey with more than 30 questions. You will find yourself getting irritated
pretty soon and even entering random nonsense as answers. This is true for even the most sincere
people or amounts of compensation they have provided. Thus, one of the crucial problems is that of
shortening the survey as much as possible, but not too short as to be an informative

Conjoint Analysis - Apartment Example -

Setting the Background
CONJOINT ANALYSIS - THE APARTMENT EXAMPLE - SETTING THE
BACKGROUND

The product that we have considered here is an apartment. We have considered 4

characteristics of the apartment.

They are:

Area - 1000/1200/1500 sqft.

Metro accessibility - Yes/No
Swimming pool - Yes/No
Price - 70/60/50 lakhs
We try to find out how many apartment configurations are possible.
Title: Conjoint Analysis - Apartment

Example - Setting the Background Let us do a very simple Conjoint study now. We will try to find out
a single respondent’s relative preferences for different attributes of an apartment. So, let's think of
yourself as a real estate developer, and you have different options to build an apartment for and you
want to find out how to design the optimal apartment. So now you could have let's say three options
for area 1000 square feet, 1200 square feet, and 1500 square feet. You can also have built it near a
place with metro accessibility or not. As you know, people nowadays value apartments with near a
metro station a lot more than if it were not there. Similarly, you could also decide whether you
should build a swimming pool or not. Clearly, if the user does not value a swimming pool, then
there's no point building it. And finally, what price? So, 70, 60, 50 lakhs. Now, if you are in a big city
like Bombay or Delhi, this might seem ridiculously low price even Bangalore. But there are still places
in India where you can get apartments for this price. So how do you proceed? First of all, what we
will do is the following: We will try to find out how many possible apartment configurations are
there. Now, you can see there are three types of apartments in terms of area, you could build it with
or without metro access. So that's two. And with or without a swimming pool and at three price
points and if you multiply all of this a total of 36 possible apartment configurations are possible.
Now. you could have, for example, let's say 1200 square feet with the metro access, with the
swimming pool and let's say 60 lakhs. Now, what we will assume is that each attribute level has what
is called up part-worth. We will call that U. And it is called part-worth often part-worth utility, which
is why we're using the letter U. Now, what we will also assume is that there are baseline utilities for
the worst possible configuration. So, for example, if area is 1000 square feet, let's assume that U
1000 equals zero. Now, if you increase the area to U 1200 to 1200 square feet, you will have
something here and we will assume thatthisis more than 1000 square feet apartment keeping all
other attributes the same and finally another one for 1500. Similarly, let's call, you know metro as
this is zero and we will call it UNM and this is less than U with the metro. The assumption here being
that metro accessis desirable and better than no metro access. Similarly, we are making an
assumption here that swimming pools are valued. Now, you might have certain reasons to believe
that swimming pools may not be as valuable because of the maintenance costs et cetera. But
generally, people tend to prefer apartments with swimming pools than without given the same
price, of course. So, we will call this, U no pool equal zero and what we need to find is relative to that
what is the value of having a pool UP. Clearly again, the higher the price, the worse it is. So, we will
call this U70 equals zero and we are just setting a baseline here and seeing relatively if I were to
decrease the price to 60 lakhs. What would happen? How would this be valued? So, we have U60
and U50. Now, you may be asking, why don't we have 71, 72, 73, 74 etcetera? Yes, you can. But
instead of three price points here then you might have 9, 10, 11, 15, 20 depending on how granular
you wanted to be. For the sake of simplicity here, we have just kept it at three so that our total
number of possible configurations is low

Conjoint Analysis - Apartment Example - Final

Analysis
CONJOINT ANALYSIS - THE APARTMENT EXAMPLE - FINDING THE PART WORTHS
AND FINAL ANALYSIS

Let us do a very simple conjoint study now.

We will try to find out a single respondent’s relative preferences for apartments.

Rate the following apartments from 1 (lowest) to 10 (highest) according to their attractiveness
1. 1000 sqft, metro access, no swimming pool, price: Rs. 50 lakhs RATING: 3

2. 1000 sqft, no metro access, swimming pool, price: Rs. 60 lakhs RATING: 2

3. 1000 sqft, no metro access, no swimming pool, price: Rs. 60 lakhs RATING: 1

4. 1000 sqft, no metro access, no swimming pool, price: Rs. 50 lakhs RATING: 1

5. 1500 sqft, no metro access, swimming pool, price: Rs. 70 lakhs RATING: 8

6. 1200 sqft, no metro access, swimming pool, price: Rs. 60 lakhs RATING: 4

QUESTION 4
4. 1000 sqft, no metro access, no swimming pool, price: Rs. 50 lakhs RATING: 1
 U +U +U +U =1
1000 NM NP 50

0r, 0 + 0 + 0 + U = 1 50

Or, U = 1 -----------(a)
50

QUESTION 3
3. 1000 sqft, no metro access, no swimming pool, price: Rs. 60 lakhs RATING: 1

U +U +U +U =1
1000 NM NP 60

0r, 0 + 0 + 0 + U = 1 60

Or, U = 1 -----------(b)
60

QUESTION 1
1. 1000 sqft, metro access, no swimming pool, price: Rs. 50 lakhs RATING: 3

U +U +U +U =3
1000 M NP 50

0r, 0 + U + 0 + 1 = 3
M

Or, U = 2 -----------(c)
M

Title: Conjoint Analysis –

Apartment Example - Final Analysis So, here is a survey with 6 hypothetical apartment
configurations and I got a colleague to fill this up. And we will be analysing her own relative part
worths based on this. Here is what my colleague has filled. So, you can see that most numbers are in
the lower range below 5 except there is this one apartment which she seems to think is very
desirable to her. So, let’s figure out what her value system is and what is driving this set of ratings.
So basically, what we are going to do is try to populate this table. Now the worst levels of each
attribute, the lowest area, no metro access, no swimming pool, and highest price, remember high
price is supposed to be bad and you are usually happy when the price is low, are assumed to be 0.
And relative to this baseline, relative to 1000 sqft. for example, we are trying to figure out what
these are going to be. Relative to no metro, how do you value metro access. Relative to no
swimming pool, how you value swimming pool and relative to 70 lakhs, would you be happier? How
much happy will you be if your price goes down to 60 or 50 lakhs? – 10 lakh differences. So now let’s
first start with question 4. You may be wondering why question 4 and not question 1. The answer is
very simple. You will see that it is easier to hand calculations if we go in a certain order. In reality
when you are doing a large number of these computations, you are just going to input this into some
sort of computer solver and it doesn’t matter what order you are in or if you are good at algebra,
you could go in different orders or have different ways of formally solving these equations. I am just
going to do this order for the sake of convenience. Remember it is nothing more than that. So,
question 4 tells you, of a hypothetical apartment, of 1000 sqft., no metro access, no swimming pool
and price of 50 lakhs. You can see that the first 3 attributes of baseline attributes while this is not a
baseline attribute. So, lets formally write down the equation once and later we will take shortcuts.
U1000 which we know is 0 plus U no metro also 0 plus U no swimming pool is also 0 plus U50 which
is not 0 equals 1 or 0 plus 0 plus 0 plus U50 equals 1. Essentially, we have solved for U50. So, as you
can see here now, I have put U50 on U50 a value of 1 for its part worth. Now we go to question 3,
again remember for convenience, and now we will directly write this as 0. We know that 1000sqft. is
baseline, no metro access again baseline, no swimming pool again baseline and now the price is 60
lacs, this also equals 1 and we have now solved for U60 as well. So, lets populate this table again,
now we have solved for U60. So, now you can see interestingly that common sense would tell you
that reducing the price should make something more attractive to the consumer or more valuable.
However, consumer does not seem to place much value on a price decrease at least in this range. So,
now lets go to question 1 and you see that here 1000sqft is baseline, so we will directly put it as 0, U
metro is something we need to solve for, U no swimming pool we know is 0 and U50 we have solved
as 1 equals 3, or Um plus 1 equals 3, Um equals 2. So now let’s populate the table, so now we have
populated a price the part worth of metro here. Now interestingly you will note that getting metro
access seems to be valued by this customer more than a price drop of 10 lacs. So, price drop of 10
lacs from 70 to 60 will elicit a part worth value of 1, while just building your thing near a metro is
going to get you an extra relative value of 2 here. This is what its saying, now of course you want to
do your backend cost benefit analysis and see if buying land near a metro station is more expensive
or less expensive etc. So, now we go to question 2 and Copyright © IIM Bangalore. All rights
reserved question 2 again 1000sqft is baseline, no metro is baseline, swimming pool is not a baseline
anymore, U swimming pool we will write it as Us and U60 we have previously solved to be 1 equals
2. So, Us plus 1 equals 2, Us equals 1. So, now we have solved for the part worth of swimming pool,
so let’s populate that as well. So now we have filled in as swimming pool and you know that a lot of
consumers seem to like a swimming pool in their apartment complex. This particular person equates
it along with comparable to a price drop of 10 lacs from 70 to 60. But now we need to find relative
attribute to the part worths for area. Now area as you know is a very important consideration while
buying a flat. So, lets just take a look at that. Question 6 now is the first question without baseline
area. So, now U1200 is what we need to find out, U no metro you know is 0, U swimming pool you
have solved and it gives a value of 1 and U60 also value of 1 that you have solved previously equals
4. U1200 plus 1 plus 1 equals 4 and if you solve this U1200 equals 2. So, now let’s populate this and
now you see U1200 is 2. So, now U1500, 1500 sqft is what we need to solve for. No metro access is 0
baseline, swimming pool we have solved for Us equals 1 and baseline price here, right, so baseline
price is 0 equals 8 and you see this large number coming up, right. Before from 1,2,3 or something
like that we suddenly see this number coming up 8. So now U1500 plus 1 equals 8 and U1500 equals
7, very interesting large part worth coming up. So, now let us populate the table and do our final
analysis. So, lets fill this in and you will see this disproportionate jump from 2 to 7. So, just adding
300sqft. from 1200 to 1500 gives you 5-part worth points or just adding 500sqft from 1000 to 1500
gives you a jump of 7. Now intuition would tell you without data if you were to do this that a price
decrease would be valued by the customer, but this particular customer doesn’t seem to be valuing
price that much. There is some little value given to metro access and some little value given to
swimming pool but not comparable to the sensitivity that this person shows for an increase in area.
So you may want to at your backend based on your cost build near a metro, give a swimming pool
and of course that means sacrificing some area or reduce price which means that’s directly costing
you money but all this seems to pay in comparison to building a 1500sqft apartment or just try to
think what kind of person this might be, probably has children, probably married, doesn’t value the
metro that much, so, probably owns a car or probably doesn’t have time to use the swimming pool,
a lot of busy professionals have long working days and the swimming pools are usually closed at
night anyway, right. But this person might be having an extended family, maybe in-laws, maybe
parents also living over which means that a high area is important.
Week 03: Survey Design, Constructs and Scales
Learning Outcome Week 03
By the end of this week, you will be able to:

 Analyse different survey designs: Summarise their pros and cons

 Explain a Likert Scale
 Explain different validity tests
 Determine Cronbach's Alpha using a data set

o Title: Survey Design

Welcome back. Thus far, we used surveys to measure willingness to pay, which is a monetary
amount, as well as gauge ratings forspecific productsin a conjoint analysis. However, surveys are an
important method for gauging sentiments and collecting information over far more complicated
settings. For example, how would you gauge someone’s attitude towards recycling e-waste? What
about their attitudes towards online shopping? Is this as simple as asking someone their income,
age or even willingness-to-pay, all of which can be represented by a single number? A few years
ago, two students at IIMB, Arko and Nilotpal, wanted to help a small start-up in researching
potential customersfor an e-waste recycling service they had come up with. I will withhold the
name of the start-up for reasons of confidentiality. They would collect your e-waste for a small fee
and sell it to processing companies. Occasionally, they would also offer e-commerce coupons as
gifts to incentivize you. Beautiful social initiative, and with a profit motive as well. They came to
me, and together we designed a survey that we floated to an online panel of Indian consumers to
gauge their attitudes towards multiple variables of interest. So, let us take a look at one variable of
interest, which we will call Attitude towards e-waste recycling. How would you find out if someone
is inclined to recycle e-waste? Would you ask them an open-ended question something like what
do you think about e-waste recycling? If you did ask an open-ended question, which qualitative
researchers do, then you would have to observe their sentiments and interpret their answers. Can
you quantify these into a number? Probably, yes, but with some subjectivity. A lot of big data
analytics today focuses on extracting sentiments from such unstructured data. So now, think about
the following question instead, do you recycle your e-waste? Yes or No? This is easier to analyze,
but it still has a problem. You cannot really distinguish between someone who is very enthusiastic
or somewhat enthusiastic. Both will answer YES. Similarly, someone completely unwilling and
someone somewhat unwilling to try it will both answer NO. I will ask the following question now.
As an environmentally conscious citizen, how likely are you to recycle your e-waste? You can see
now there is a scale of 1 (extremely unlikely) - 7 (extremely likely) which is provided. This is called a
Likert Scale. So, you have solved the YES/NO problem here in giving some range to measure
relative willingness. So now you know the difference between somebody who is extremely willing
and somebody who is somewhat willing or completely unwilling et cetera. It is better than a YES or
NO in terms of richness. But is this ideal? Think about it. Look at the phrase “as an environmentally
conscious citizen…” You may have subconsciously transferred your own biases on to the
respondent. Is this a problem? Definitely! For the time being, you may have caused the respondent
to think that they care about the environment, biasing them to answer YES. In real life, somebody
who doesn't care about the environment would have answered No unless you had reminded them
and in real-life it will continue to be a No which is a fatal mistake for you and e-waste recycling
company that is making money of an e-waste recycling service, right? Because this person is not
going to use your service. He may or may not care about what you offer. Copyright © IIM
Bangalore. All rights reserved In a later module, we will discuss ethics of marketing research, but
such leading questions are often asked by political pollsters in attempts to sway public opinion.
Think about your own news channel guys who might have sometimes asked you, ‘as a patriot what
do you think of action XYZ of the Government etc…’ These are actually attempts to sway your
opinion rather than actually gauge it. The most infamous example of this would be Adolf Hitler,
whose 1938 ballot paper read as follows: You can see something in German here which translates
to roughly "Do you approve of the reunification of Austria with the German Reich that was enacted
on 13th March 1938 and do you vote for the party of our leader Adolf Hitler." And you can see the
word Ja or JA means Yes in German and the word nein which is much smaller means No. When in
doubt look at the above ballot paper and avoid everything it does. It is a classic example of leading
question, bad design, poor question framing etc. First of all, a larger yes and smaller no are
intended to bias you towards a yes, the outcome that these people want. In marketing research,
there is no desired outcome. You are measuring and not inducing opinion. Next, it asks two
separate questions. Imagine the following question instead of the more loaded political question.
How likely are you to recycle your plastic waste and e-waste? Any answer here is meaningless, as
we know that e-waste and plastic waste are two separate categories of waste. Recycling plastic
does not necessarily mean you will recycle e-waste and vice versa. It is better to ask these as two
separate questions. How likely are you to recycle plastic waste? Question 2, how likely are you to
recycle e-waste? Another type of leading question and this is something we discussed before
already. It is well known that the planet is in distress due to e-waste. How likely are you to recycle
your plastic waste and ewaste? This is both severely leading and has two questions in one. Try to
avoid this. Sometimes when you ask these questions it may not be very obvious to you that you are
leading the respondent on. As a good market researcher your job is to audit these questions, get
them privately audited, get them independently audited etcetera. Go through multiple rounds of
design to make sure that they are not leading and that they can gauge the participant's opinions
rather than influence the participant's responses. So now, let us ask the following question. How
likely are you to use an e-waste recycling program with a 7-point Likertscale? While no question is
perfect let us take a look at the framing of this question. At least it avoids some of the pitfalls we
just discussed. It is a single question. It tries its best not to lead you on and it gives you a range of
responses from 1 to 7 but is this sufficient? Pause the video and think. What if you were taking a
competitive exam like the Common Admission Test or CAT conducted by the Indian Institutes of
Management or the more general Graduate Record Examination or GRE for admission to graduate
programs. There could be many such tests and you know there are many competitive exams which
have multiple questions. Imagine you were in the quantitative section, and they asked you ONE
question, a single question on the basis of which your aptitude were to be judged. Would this be
reliable? Would you be happy about this? One single Math question would determine whether you
have good quantitative aptitude or not. Actually, you would not be happy, right? This is exactly
why these exams ask multiple questions. You may slip up or get lucky on one question, but with
many questions, the test would be a fairer assessment of your aptitude. In other words, if you
were to answer question 1, your answers to Copyright © IIM Bangalore. All rights reserved
question 3, 4 etcetera would be correlated. So, you can answer one type of question, multiple such
questions, you would be able to answer more likely and maybe a couple here and there you would
slip up. A similar principle is used in designing surveys

Validity and Likert Scale

Likert scale is used to collect data - data on the respondent attitudes and opinions.
What are the benefits of using the scale? Is odd likert scale better than an even one?

Watch the following video to know the answers to these questions.

Also, here nomological (name) validity has been described briefly (probably a
sentence or two) . This sets the basis for the next unit.
Title: Validity and Likert Scale Welcome back. In our last video, we talked about question framing
and the need to use multiple questions to get some variables. Let us elaborate on this now. Our
variable of interest is Attitude towards e-waste recycling as you can remember. One question we
used was: How likely are you to use an e-waste recycling program? As discussed, this one question
may not be enough. Reasons could vary from the respondent not paying enough attention and
randomly clicking an option, to actually having an understanding of the question that is different
from yours. Yes, this is a common problem in survey design. Like a quantitative exam where you
could goof up on one question or get lucky on it, a single question isthus not a reliable gauge of an
involved variable like Attitude to e-waste recycling. We thus ask the same question in many
different ways. So now, let us take a look at each of these questions, also asked in the same survey:
1. The first question, of course, is what we had before. How likely are you to use an e-waste
recycling program? 2. Now, look at the second question. How inclined are you to use an e-waste
recycling program? Sound very similar, right. How likely versus how inclined. It is basically asking
the same question in many different ways. 3. The third question, again asking the same question in
another different way by saying, how willing are you to use an e-waste recycling program? Take a
look at the next three questions. They are not framed as questions but are statements with which
you have to agree or disagree. 1. I am likely to recommend this to a colleague or friend. 2. I am
likely to say positive things about this service in general to other people. 3. I am likely to encourage
friends and relatives to use this service. The next three questions are actually different from the
first three. Pause this video and look carefully at each question. As we discussed earlier, no
question is perfect. Would you rephrase them? Before we move on, another feature of interest is
the 7-point Likert scale that we have used for each question. Why do we use this particular
formulation? As we discussed earlier, a simple YES/NO question may not yield sufficient details
about people’s relative attitudes, but then why not use a 100-point scale instead? While a 100-
point scale is indeed more granular, so-called granular, try answering the same question on a 100-
point scale. Can you really give a consistent answer? For example, is answer of 37 on a scale of 100
truly different from say 41 or 39? The truth is, the human brain is bad at processing large numbers,
even if you think otherwise. While you want your scale to be sufficiently granular and capture
differencesitshould not overwhelm the participant. The number 7 is convenientin thatsense.
Copyright © IIM Bangalore. All rights reserved We could have a 10-point scale as well, but with an
even number, there is no natural middle point. Thus, an odd number, in a 7-point scale that is 4
stands for something called a so-called neutral option. So, you neither agree nor disagree
somewhere in the middle. Additionally, the 7-point scale gives you sufficient granularity without
overwhelming the respondent. The cognitive load becomes more and more after 7. Try
memorizing 8- and 9-digit numbers, and you will find it extremely hard to do so. 7 on the other
hand, is just about ideal for most people. Coming back to the questions themselves. Do they
measure the same thing? The first test is that we imply is something called nomological or name
validity. Look at the first three questions and compare them with the next three. The first thing you
see is that they seem to ask two different things. The first three questions actually try to gauge
your own attitude towards e-waste recycling. The next three, however, gauge your willingness to
spread word of mouth. So, its own attitude versus word of mouth. Two distinct variables, both of
which are of interest to a startup that wants to do e-waste recycling! Our next test would be to see
if people actually think the same as you think. To do so, we need to see two things: 1. Responses to
the first three questions are highly correlated and Responses to the next three questions are also
highly correlated. The firstthree represent the one block and the next three represent another
block. 2. The second thing is responses to any of the first three questions should have low
correlation with responses to any of the next three. Ideally, that should be the case to show that
these are different constructs. Sometimes you find that there are high correlation in which case we
will go back to this idea of nomological validity. The second measure, the correlations are about
statistical validity. So now, we will have to take a look at the data. So, please download the data file
which we call “att_ewaste.xlsx.
Title: Nomological Validity Our next test would be to see if people actually think the same as you. To
do so, we need to see two things. One is that responses to the first three questions are highly
correlated and responses to the next three questions are also highly correlated. So, we treat the first
three questions as a single block, and they have to be correlated and the next three questions will
treat as another block so they should also be highly correlated. The second thing and this is very
important also from a statistical point of view is that responses to any of the first three questions
should have low correlation with responses to any of the next three. Sometimes, it doesn't happen,
but ideally, this is the case. And then you can say that they're talking about different measures. So,
first we have to take a look at the data. Please download the data file called "att_ewaste.xlsx”, it is
an excel file, and this session we will be using only excel. So, just take a look at your data set. So
now, what you're seeing is responses to six questions. Each of these is actually a single respondent
answering all six questions and responses are between one and seven. This is the Likert scale that we
have used. So, the firstthree questions and we have actually labelled them according to the
questions, likely to use e-waste recycling, inclined to use e-waste recycling, and willing to use e-
waste recycling. We discussed these questions right, and this we will call Block one. These three
questions are going to be Block one, and it is basically your own motivation or own intention to use
e-waste recycling. This is variable one or construct one with one block of three questions. Block two
has three more questions and now this is about likely to recommend, likely to say positive things,
and likely to encourage friends and relatives. So, first of all, take a look at the difference between
one and two. We already discussed this, but this is about some sort of intention to spread word of
mouth. You can phrase this better if you want, but I'm just going to say intention to spread word of
mouth. Both are important, your own intention and your intention to spread word of mouth. These
are both important to marketers and essentially, we're going to do two things now. We spoke about
nomological validity. We are now trying to do some statistical validity tests. So, the first thing we will
try to do here is to create whatis known as a correlation matrix. A correlation matrix is basically...
you take any two variables at a time. So, let's say these two, these two, it could be these two, and
you can see there are six variables here so you will have 6 times 5 divided by 2. You will have actually
15 different correlations. You take any two pairwise and find out its correlation. So, one way you
could do it is by using the function of correlations and excel has some function, and I think it's called
Correl, but we can take a look and then do A…but then you will have to do it multiple number of
times for each pair, right, and this is not very convenient to do by hand because you will have to do a
lot of correlations. So, we will use a feature of Excel called Analysis Tool pack, which actually
generates all these pairs of correlations. So, let's go to Excel Analysis Tool pack. So, any version of
Excel since 2016 has something called an Analysis Tool pack, you will find it in your data tab. It's
called Data Analysis. If you cannot find this, what you need to do is essentially and go and enable this
from your tools. This is a Mac version. You will have similar versions in windows. If you go to Tools
and go to Excel Add-ins here, you will have something called Analysis Tool pack, Solver Addin
etcetera. Copyright © IIM Bangalore. All rights reserved So, let's go to our Data Analysis and click on
something and find something called Correlation. So, we will click on correlation and now my input
range is going to be my six variables that I have here. So, A, B, C, D. So, A to F, essentially A, B, C, D, E,
F are part of my data input. Now, you can see that the first row here is a label. So, it is the label. So, I
will click on that and you can either output this to the same worksheet and choose a range, but I will
just take it to a new worksheet. So, it will add a sheet to Excel, and once I click okay, it will actually
take me to a new sheet. So, let's take a look at this. So, here is a correlation matrix that you see, and
we will first explore what the structure of this is. You see that the upper triangle here is blank
because it doesn't matter if you take correlation between X and Y, and correlation between Y and X
or the other thing is, if you know what a correlation is, you will see that a variable is perfectly
correlated with itself, so likely to use and likely to use are perfectly correlated, inclined to use and
inclined to use, a perfectly correlated etcetera. The next thing we want to do is look at correlations
between any two. So, let's take a look at correlation between likely to use, and I'm using a short-
form here the actual variable is much longer. If you expand these cells, you'll see them. But it's also
unwieldy to do that here, so likely to use and inclined to use, if you do, you will see it is this number
here. 0.5596 or if we keep it to two decimal places, this is about 0.56. So now, let us look at all the
pairs. The first column shows correlations between responses to the first question and the others.
So, you will get 1, 0.6, 0.75, 0.61, 0.59, and 0.58 respectively. These are all pretty high. Now, what is
high? It's difficult to actually say what is unambiguously high, and what is an unambiguously low.
High would be anything more than 0.5 or less than minus 0.5. So, you can also have negative
correlations. You don't see any negative correlations here though. Now, everything is above 0.5 here
and seems pretty high. It's a good thing that the first three are high, and that's what we wanted,
right. But the next three are also pretty high, and this is not very good, which means that there is
high correlation between the two blocks. So, before we go forward, just pause the video and try to
find out what are the second question correlations with the first, second and up to the sixth
etcetera. Just pause the video and try to do this yourself. Did you get 0.56, 1, 0.52, 0.38, 0.38, and
0.38, we are rounding up to two decimal places here. You see high correlations between the first
three, but much lower relations with the next three, and this is exactly what you would want in an
ideal scenario. Pause the video and do the same exercise for all the other variables. The correlations,
what I have here are for the third 0.75, 0.52, 1, 0.65, 0.68, and 0.63. For the fourth, 0.61, 0.39, 0.65,
1, 0.71, and 0.68. For the fifth, 0.59, 0.38, 0.68, 0.71, 1, and 0.8. And for the last, its 0.58, 0.38, 0.63,
0.68, 0.80, and 1. As I said, what qualifies as a high or low correlation is subjective, but let's assume
that 0.5 is high and most correlations between any two pairs except those 0.38 that you saw are
pretty high and this suggests that this may statistically speaking the part of the same construct or
variable. However, we will go back to our idea of nomological validity here. You can see that the
questions you have here are unambiguously trying to measure two different things from a language
perspective. So, we will actually go back to that and say that these are two different constructs.
Now, you might have been in trouble if the questions themselves were worded similarly. What this
also suggests is that these two constructs are heavily correlated, and they might even be causal
relations, A causes B or B causes A or things like that. There are things called... There is a method
called Structural Equation Modelling that actually takes a look at such causal relationships. We are
not going to go there in this course because it is much more advanced.

CRONBACH'S ALPHA

It tells you if the variable of interest is accurately measured by the test that you have
designed.

o Title: Cronbach’s Alpha Our next task will be to look at what is known as “unidimensionality” of
our two variables. Simply put, this means, is one scale enough? So, if yes, we can take the averages
of the three items and put a single number in our data sets. Kind of like, how your quant exam are
totalled and used as a proxy for your quantitative ability. To do so, we calculate something called
Cronbach’s Alpha for each of these two variables. To calculate Cronbach’s Alpha, we will actually
need to calculate variances and co-variances of each variable. Let's go back to data analysis and you
see something called co-variance matrix. Let's just generate that in a new page. So, we again select
A, B, C, D, E, F six columns, labels in the first row and will also come to a new worksheet as follows.
So, this is what a covariance matrix looks like. It's similar to correlations, except the diagonal now is
the variance. So, a covariance of a variable with itself is nothing but it's variance. So, diagonal and
the others are similar to correlations. You can take any pair of variables, and you will get covariance.
So, now we will define something called Cronbach’s Alpha. So, Cronbach is the guy who invented it
and it's Alpha. It's given by the following formula and I'll explain to you what this means N times C
bar over C bar plus N minus 1 times C bar. Now, this is looking like a fairly complicated formula. So,
let's just explain what each term means. Now, we are going to take a block. Now, you remember, I
will just do it for the first block, which is 3 questions and Cronbach’s Alpha is always calculated
within a single block. So now, let me call this again for the sake of convenience I am going to call this
variable 1, this I will call variable 2, and this I will call variable 3. So, only the first three. Remember,
we are not taking the next three. So, this block has three variables and N is the number of variables,
so N equals 3. The second thing we will do is remember I told you these are the variances, so V bar
equals average variance in block. So, you will actually take these three because there are three
variables in this block and calculate their average. Very simple. C bar equals average covariance. So
now, you have covariance between variables 1 and 2, covariance between variable 1 and 3. So, 1 and
2, one covariance another covariance between 2 and 3, another covariance between 1 and 3 and
you take their average and that becomes C bar. So, let's try calculating this. So now, what I am going
to do is extract each of these things. So, I'm going to call V1, V2 and V3 as the three variances. So,
what is V1, it will be the variance of the first response, I am just directly putting it in here to make
this easier for me later on. Are you along the diagonal and get 3 variances? Now, V bar I will call as
the average of these three, I will just take the average of these three and this is my term V bar. What
I now need is covariances and similar to how you located correlation we will do the same thing here.
So, I need the combinations here so there are three variables will actually have three covariances.
Since we have three here, C23 and C13. Remember, C23 and C32 are the same, the order doesn't
matter. Similarly, C13 and C31 are the same. I will do this now. The first one is C12. So, on the
column I will take the first one and here two, so I get this. Do you get the same? Verify. C23, what
would C23 be? C23 would be the second one here with the third one here. This is C23. Check if you
have the same and C13 would be the correlation between the first and the third, which comes here.
I will now calculate the average covariance as the average of these three. To calculate Alpha, let's
write the formula. You remember what the formula was, but I'll write it here it is N star C bar and
the bar is essentially what we wrote a bar on top, but I will just write it as C bar and then it is V bar
plus N minus one times C bar. This is the formula that we need to do. So, let's just Copyright © IIM
Bangalore. All rights reserved apply this formula here. You remember what N is. N is 3 times C bar
divided by V bar plus three minus one is two times C bar. We're not going to go into the math of
what, how the Cronbach’s Alpha comes about but we will take it as given. And you get this number
0.818. So now, we have Cronbach’s Alpha calculated and I will put it to two decimal places at 0.82.
Now, Cronbach’s Alpha is between zero and one. If you have low values, low Alpha, then you have
bad news. Low Alpha is not a good construct. What is high and what is low? Again, this is very
subjective, but a lot of market researchers take Alpha greater than 70% or 0.7 is considered usually a
benchmark for a high Alpha. Now, you can see here this is much greater than 0.7. So, we will take
this as a satisfactory Alpha.

Week 04: Segmentation, Targeting and Positioning

Title: Introducing STP Hello and welcome back. So now, this week, we will discuss something called
the STP approach or Segmentation, Targeting, and Positioning. Now, you may have heard of this,
especially the first part. The first two parts, if you are familiar with some course in marketing. You
may have taken some course in marketing either in college, if you have gone to such a college or
some of the MOOCs including IIMB's own MOOC done by one of my colleagues on marketing. Now,
segmentation, targeting, and positioning or STP as it is called is a predominant way of looking at
marketing that's very, very powerful. And most marketers today, look at marketing through the lens
of the STP framework. So, what do these terms mean? We will do a very simple example, illustrative
example here and then go on to show you how computers and machine learning can be used to use
actual survey data in segmentation, targeting, and positioning. But let's get familiar with some of the
terms. What is segmentation? Now, the idea of segmentation is that you have many customers in
your market and find groups of customers with similar tastes. So, this could be segment one, two,
three, would have multiple segments, et cetera, and then, based on your own business needs you
focus or target one of the segments. So, you first find segments, you find segments, focus on a target
segment, or segments. Finally, positioning is the entire set of marketing actions that you are going to
take based on what you've learned about preferences of your target segment. So, STP. This is a very,
very important framework and we will be doing this in the context of marketing research

o Title: Illustrative Example: Segmentation

So now, let's take a very simple, illustrative example, and, as we go along you will see that it will
seem almost ridiculous and to your common sense. But then we'll discuss, how this ridiculous
example can translate to other scenarios and suddenly you will find how this makes sense. Let's
imagine the following: You are running a cafe, something like coffee day, Starbucks that kind of cafe
and you have two offerings Tea and Coffee. Let's assume now, you have 20 customers, again very
simplistic scenario, this could happen on multiple dimensions, but we're going to take two
dimensions here. Let's say you have 20 customers, it could be much more, ideally, much more, but
20 for this scenario and you ask them, do you prefer Tea or Coffee? 10 people say Tea, 10 people say
Coffee. So far, so good. And I have put that on an axis. This axis, I will call temperature in degree
Celsius. Now, you know a lot of people like hot Tea and I just did a Google search hot tea is between
85 and 95 degrees Celsius. Five people and you ask them now what temperature would you like it
at? Let's assume that they know their preferences well. Five people put something here. So, there
are five people now and you see that the average is 90, let us say between 95 and 85. The average
comes to 90. So, five people have said what their ideal temperatures are, right? There are lots of
people who like Iced Tea. Again, there are five people who have... this is not very accurate and to
scale. As you can see, I'm not very good at that between 10 and 20 degrees Celsius, let us say again
based on a Google search, I found these are ideal temperatures for a cold beverage. So, there are
five people here who say this. So, this is about 15, 20, and 10. Similarly, there are five people at the
same temperatures. Of course, you can see this is a hypothetical example who like hot coffee and
five people who like cold coffee. Iced tea, cold coffee, these are also popular beverages and if you go
to most cafe's, you will see it. Now, remember, as human beings, we're trained to think in averages,
right. So, if a lot of people give you decisions, you say this is the average, right. So now you can see
equal number of people like tea and coffee. Equal number of people like it hot and cold at some
temperature. So, let's assume you say something like 52 degrees. Something close to that is the
average and you say I'm going to make a Cocktail, which is at 52-degree Celsius is the average
preference. This is the one beverage that I'm going to offer 52 degrees Celsius, which is lukewarm.
It's like a hot day in Delhi at the moment, sorry that was an exaggeration, but pretty close and 50%
Tea, so half tea plus half coffee, basically, I am going to take the average preferences. What do you
think will happen to this product? Who is going to buy it? Is anyone in this group of people going to
buy it here, here, here? Think about your own preferences? Do like lukewarm drinks? I don't know
about you, but most people have I have spoken to don't. They like it either hot or cold, especially
when it's tea or coffee, water may be different, and I have actually tried making a cocktail of tea and
coffee. It was one of the worst drinks that you can make. Now, maybe you are special, and you like
it, but most people probably won't. Now, this sounds ridiculous, right. Nobody really gives you half
tea or half coffee. It's obvious they give you the cold tea, hot tea, cold coffee, hot coffee. Now,
mathematically speaking for business purposes, an average or a mean only make sense when the
variation is low. You can see the variance is very high here, right? On the other hand, if you were to
make tea at 90-degree Celsius, this guy may not be completely happy, but he should be okay, just a
five-degree difference. It's still pretty hot, it is same here 95 to 90. It is still pretty hot, 85 to 90, it is a
little hotter that you wanted to be, etcetera. These people are more or less likely to buy. Same here
offer hot coffee. Coffee at 90 degrees C. Similarly, cold tea, iced tea, ice coffee. These four drinks, I
will just write it as cold now tea, cold coffee. So, let's call it as hot, and if you offer these four drinks,
there will be groups of people who buy it. Now, let's assume for some purpose, you are not able to
procure tea. Then you may focus only on coffee, or if you don't have good refrigeration facilities, you
may give only hot coffee, like a lot of places. If you have both, maybe you give both or for certain
business purposes, you may focus on only certain segments. This is a segment of people who like
coffee and who like it hot. This is a segment, this is a segment, and this is a segment. So, what we
have done effectively is segment people into four categories.
Illustrative Example: Targeting & Positioning
.Now, let's say you are something like the Starbucks. You're going to say I don't want to serve Tea; I
will focus on only Coffee. That becomes your target segment or if you are in Bangalore, you have a
lot of the so-called Darshini, that serve only hot coffee and hot tea. In that case, you will target
segments. There are specialised places which may serve only cold coffee. If you go to ice-cream
parlours, for example, they are going to serve only cold coffee. Not hot tea, not cold tea, not hot
coffee, only cold coffee. In which case, they're zeroing down on what we call as target segments.
Now, there may be a variety of reasons for this. There might be more people willing to drink coffee.
You may be in touch with a coffee estate and able to get coffee at a cheaper price. It might be that
people know you for coffee, something like Starbucks or Coffee Day. It would be ridiculous today if a
brand called Coffee Day starts specializing in Tea and things like that. That is the third step of
positioning. How do you position yourself so that the customer does business with you? Essentially,
so these are the three things of segmentation, targeting, and positioning.

Title: Few Examples: STP

So now, this seems very ridiculous, right. Clearly, you know that nobody makes tea and coffee,
cocktails, and nobody servers it lukewarm. But think about other contexts where company's figure
out that they are not doing the segmentation very well. For example, let's take a look at the idea of
computers. Now, there are many different segments of computer users. There might be basic users
who want cheap computers. They don't really care much about video graphics. They don't care much
about processing power, but they wanted cheap, right, basic laptops for 20,000 rupees or 22,000
rupees. That would be a very basic segment. There might be people who use it for high-performance
computing, they don't care about video. They can pay a little more price, but they want the
processor to be really, really good. There could be a third segment out there, who are Gamers, who
need a lot of video graphics, they're willing to pay a lot of money, and they need a lot of processing
power. Now, if you take those three and give something in the middle, intermediate price, but it
does not have either the best or low capacities or the same thing with the video graphics, maybe
video graphics, which is better than what we need for surfing, but not good enough to play the latest
three dimensional games. You're going to end up with nobody being happy. There are other
examples, the housing market for example, there might be people who want luxury homes, there
might be people who want affordable homes. Now if you price them somewhere in the middle and
give middle of the line facilities. Neither segment is going to buy. You can think of education sector.
Basically, any business sectors faces this problem. What kind of segment and whenever you have a
large number of customers, you're bound to have segments of customers? Very few products out
there or services out there can actually appeal to the entire market. So, remember, when we did the
survey design for an e-waste recycling company, sometime later on in one of the later on videos, we
will actually be doing segmentation of customers based on how their environmental concerns, how
their attitude towards recycling, awareness of recycling, etcetera are different and we will try to use
a machine learning algorithm called K-means clustering and try to find out these clusters, segment or
a cluster. So, this is another term for a segment. We will be doing that later on in this week. In the
process, we will also see how we can get information about these clusters of similar people and how
we can target those people, how we can identify segments based on their size and based on the
average preferences, which makes business sense to you, and then how to position your products
based on what you have learnt about your target segment
k means Clustering
Defining k means Clustering

It is a machine learning algorithm.

Here:

k = number of clusters/segments
means = statistical mean or average or centriod
Cluster = Group

Defining a Centriod

A centriod is nothing but an average or mean.

For example: Let us take two variables x1 and x2

Then, the centriod is: (x1+x2)/2. This is the formula of centriod for two variables.

For more than two variables, where n = number of variables,

then centriod is: (x1 + x2 + .... + xn)/n

Title: Defining k means Clustering So now that we have discussed what segmentation is. We are
going to discuss an algorithm, machine learning known as k-means clustering, in which, we will use a
machine-learning algorithm to implement segmentation. Now, before we go forward, what I will do
today is try to explain what this algorithm is in very basic terms. Now, this is advanced machine
learning and a lot of computer scientists actually research clustering algorithms, but very briefly
what do these three terms mean. First, you say something called k. k is the number of clusters and
when you say number of clusters, it also means number of segments. Basically, we will be telling a
computer that we want three segments, four segments, five segments, etcetera, and the computer
will then try to find your groups or clusters of people who are very similar and based on whatever k
you give it will find that many. Means, so this is same as a statistical mean or an average or an
average and we will be doing this in a... I will be explaining this to you in a graphical method. So,
another term for mean we will use here is centroid. And clustering, of course, a cluster is a group.
So, group, segment, cluster, they all mean the same. So, essentially, what I will be describing today in
a very intuitive manner is this process of k-means clustering for segmentation

Title: Defining a Centroid So, now assume you have a survey with many questions. Let's say you're
a cell phone manufacturer. You can have hundreds of questions in clustering analysis, but for the
sake of this explanation, I will do it only with two. So, let's say the first question is, how important is
price to you? Starting from 1, 2, ... Likert scale as we discussed up to 10: 1 being least important, so
there are some people who can buy very expensive phones, iPhones, etcetera, and don't really care
about the price or some people who have very price sensitive, most important, right? And I'll call this
variable V1. Why am I calling it V1, because I will try to plot it on a graph? So, how important is price
to you is a variable V1 and you are going to have hundreds or thousands of responses? Second
question is, how important is camera to you? Or I will make it more specific? How important is
picture quality to you? Again, same scale 1 to 10, similarly, V2. So, 10 is somebody who uses a lot of
social media etcetera, so they will find that the camera is really important. 1 is maybe somebody of
an older generation who still uses a phone for calling or some other purpose. Now, you can have
hundreds of questions like this. V1, V2, V100, V1000 etcetera. We will illustrate how k-means
clustering happens with these two variables. So now, assume you have several people and V1 was
price, V2 is camera. That's why I called them V1 and V2 so that I can... price sensitivity, price means
high price, V2 is camera and now you will find some people, who have answered anywhere like this
and I'm just marking out people in clusters here, you want to groups of people like this. So, for
example, this person has somewhat low on both a price sensitivity and camera quality. This person is
very price sensitive and doesn't care about camera, et cetera, etcetera. This is how you would read
any graph starting from zero. So now the question is, let's say I want to find three clusters. K = 3.
How would you do it? Now, think about this for a while. One way to do it is, use your intuition and
say that this is one cluster, this is one cluster, this is one cluster, the problem is your intuition is not
that great like the computers. So, you are actually going to miss out points or you will have to make
clusters that encompass everything efficiently. Now, this is not very easy, which is why we will use a
machine learning technique to do this. So, now we will use the first word. You remember, when I say
mean I used a word called centroid. This is an important concept. So, if you remember your
geometry lessons, three points of X1, Y1, X2, Y2, need not be three, it can be X3, Y3, many more
points Xn, Yn… Centroid is a point that is equidistant, somewhat like in the middle. So, that's why it's
like a mean… will be… it’s coordinates are going to be X1 plus X2 plus as many points as you have,
Xn... It's like a mean and plus and this can happen in as many dimensions as you want also. Z, K,
whatever dimensions you can have these

Title: Finding and Updating a Centroid

So now, what I'm going to explain is an iterative algorithm multiple steps. So, if you are familiar with
computer programming, it could be done using say a for loop or while loop or something like that.
So now, you have preset your number of clusters to be 3. At step one, and I will be changing the step
number, I am randomly going to assign three centroids. It could be anywhere, and I'll call this C1, C2,
and C3. Now, k equals 3 that is I am giving 3, if k was 4, I would have given 4, etcetera. That's the
kmeans. So, k centroids or k-means, means the same thing. Now, I will assume that each centroid is
the mean or centroid of a cluster. Now, what I'm going to do is take each point and see where whose
distance is closest to with centroid. So now, for this point, this is very obviously the closest, so I will
label this point as 1. This is also 1. This is also 1, simply because centroid one is the closest here. This
you can as 2. I don't know, let us say it is 1, 2, this is probably 3 here. So, each point I will label by
finding out who is the closest centroid that is step one. So now, I make a cluster one, I make a cluster
two, and I make a cluster three. This is step one. Now, what do you see is a problem. Clearly C1 is
not the centroid of cluster 1 anymore, right? It is very obvious, at the centroid would come
somewhere here actually. C3 also the centroid would actually come somewhere here, C2 also the
centroid would come actually somewhere here. It's like kind of the mean and why we are using the
centroid is geometrically it is easier to visualize. In a triangle, if you remember, the centroid is the
center of the circle. In a circle, it is also the center of the circle, etcetera, it is the center of any body.
So, clearly this is not the centroid anymore. So, we need to update the centroid. So, what do we do
now? Now that we have defined that each of these in a cluster one... so many ones, we find the
centroid of that and update the centroid. So, this is no longer the centroid, we said this is the
centroid, this is no longer the centroid, we said this is the centroid. So, you are updating the centroid
essentially, C2 comes somewhere here and this becomes step two. This becomes step one. Now,
that you have updated, you go to step two and repeat step one. Now, that this is a centroid maybe
somewhere here or somewhere here, maybe this becomes 3, maybe this is closer to three, etcetera.
These remain the same, these remain the same, etcetera. Some of the centroids get updated. What
did we do in step one remember, find the closest centroid. So now here maybe this is the 3 and
maybe this isthe 2 now? Remember, this centroid has come here, so maybe this is 2. So now, the
labeling of every point change and you have different ones, different two's, and different three's.
Now if you run this hundreds of times, let us say 100 and step 200 sometimes the k-means clustering
runs for hundreds of times. After a while, you will find that these centroids don't change much
anymore the centroids remain the same and you find clusters of points. Now, if you are familiar with
machine learning techniques, this is called an unsupervised learning technique. This would have
been easy for you, and if you're familiar with the programme in any language, say R, or C or C++, you
can try programming this on your own. So, if you are familiar with VBA scripting in Excel, you can
actually make your own k-means clustering program. Otherwise, a lot of packages like R have this
inbuilt, based on your comfort level, you can do that. We will also be uploading our own excel
template in the next session where we will do an example on K-means clustering where we will do
this over multiple variables.

o Title: K means Clustering – Setting the Background So now, what we will do is a live example of k-
means clustering. So, what you see here is an app that I have developed. It is online and the link is
shared with you. You can also follow the same steps with me. So, we have a website here where you
can see inputs, number of clusters. So, as we discussed, in k-means clustering, you have to specify
what k is. So, if you look here, the number of clusters has to be pre-specified. So, you can either go
up 4, 5, 6, 7, 8, 9. I have not provided more than 9 because typical marketing example usually have
at most 5 or 6 clusters. Usually, we may do with 2, 3, 4, 5, etcetera. As we go along, you will see
why? The default number here is 3. It will go up to 2. 1, of course, doesn't make any sense. As you
know, the entire data is one cluster, right. So, it starts from 2 and goes up to 9. What you need to do
here is choose an input CSV file. So, CSV means Comma Separated Values. It is like an Excel file but
since Excel of proprietary format. If you have Excel data and we will come to that, you need to save
it as a CSV file. Excel itself has that option. If you go to file and export, you can see the option of CSV.
It will look just like an Excel file. Here what you see is does your data have headers? We will go to the
data set and I will explain further what this means, and Comma Separated Variables or Comma
Separated Value files usually have separators in terms of comma. If you are in Europe and using a
European system, sometimes the semicolon is used. So, you can choose that if you want, but for
most cases, India, America, etcetera, the comma used as a separator between two values.
Sometimes tab-separated values, etcetera. Once we run k-means clustering, you can download the
output also. So, let us go ahead and take a look at what our data input file looks like

Explaining the variables

Explaining the Variables

To understand k means clustering better, we are using a data set. The data set can be
found here.

In the following video, the perception or the attitude variables of consumers have
been explained. These kind of data are very important for business as these will guide
your marketing decisions.

Title: K means Clustering – Explaining the Variables

So, this is what Comma Separated Value file looks like. We are opening this in excel and it looks just
like an excel file, except an excel file has dot xls or dot xlsx, while a Comma Separated Value file will
have dot CSV. Now CSV files can also be opened in things like text editors. Something like notepad
etcetera. That is the only difference. While an Excel file will not. Now, let us recall in the previous
week when we spoke about our recycling service, where a company comes, picks up your e-waste
and in return gives you shopping coupons, etcetera, right. Remember, we spoke about Cronbach's
Alpha and how multiple questions are first validated to see if a construct is consistent. The first
variable you see here is called recycle awareness. It had about three or four questions. We have
validated that in the past. So, once you have three or four questions, you take their average. We
have discussed that previously. So, for each respondent, the average is taken that is why you are
going to see something like 3.33, 4 etcetera? This is how a Likert scale of 1 to 7. ENV attitude, so if
the first variable is recycle awareness. Second variable is environment attitude. So, do you care
about the environment? Are you aware of what e-waste could do to the environment, etcetera?
There were questions about that. The third one is about how sensitive are you to shopping
convenience. So, it is called shop convenience. Do you avoid lines, do like shopping from home, and
things like that? So, because this company is giving e-commerce vouchers as gifts, right. The fourth
one you see is likely e-waste use. So, we will also be sharing these questions. You can download
them from a link given. How likely are you? How inclined are you? Etcetera, of using e-waste service,
which I am calling likely e-waste use. So, the underscore is... so that I can have multiple words, but a
variable name cannot be separated with spaces. That's why the underscore is being used. And
finally, how likely are you to spread word of mouth? So, WM means word of mouth. So you have
likely e-waste, word of mouth. Now, take a look at these variables. These are all perception or
attitude variables of consumers, which are very important for this business. Clearly, you may want
awareness, or if you see that a certain segment does not have awareness, you may want to do
something about it. Same thing environment attitude. You may want to target people who have high
sensitivity towards the environment. You're giving e-commerce vouchers, right. So again, maybe
people who are interested in e-commerce. So, the proxy for that is are you looking for convenience
shopping from home, variety, things like that? Which e-commerce has more than offline commerce?
How likely are you going to use e-waste service, of course, that's very important, and word of mouth
is also important, or you can have hundred more such variables. Just remember, the way the k-
means clustering has been done in this particular program requires that all variables are in the same
scale range. So, in the case, 1 to 7. If you don't have that, you may want to re-scale your variables
accordingly or design your survey so that they're all in the same range. The other thing you are going
to notice here is that I have not used the demographics of any consumer. Essentially, what I want to
do here is find out people's preferences then find out a target segment where the preferences seem
to match my business purposes, and in the third step, I will do marketing variables. I will make
marketing decisions that serve these tastes or preferences. Copyright © IIM Bangalore. All rights
reserved Now, it is immaterial to me, if this person is young, old, rich, poor, etcetera because the
preferences are already being covered in this survey. Now, I could use demographic data as the
second measure to see if it's correlated with something else age, gender, educational level, etcetera,
but this is not a very rigorous way of doing it. And this is why purposely, I have removed all
demographic data from this data file. It is usually not a good practice to use these in your clustering
analysis. Remember, what you are seeing here is preferences and these preferences will, in turn,
guide your marketing decisions and this is what we will be doing

Title: K means Clustering – 2 clusters What we need to is very simple now, browse in my file system,
and just import this file. It is as simple as that. Click and once it is uploaded, you are actually going to
see a table here. Now, this is a reactive place, so I can reduce the number of clusters and you can see
cluster number one and two come. I can increase clusters as well. 1, 2, 3, 4, 5, 6, 7, 8 up to 9. Let's
start with two at the moment and let's explain what this output means. So, if you go back to the data
set, you will see that there are 305 rows. What the k-means algorithm has done is to break this up
into two clusters. So, it has found two clusters with the algorithm that we explained yesterday. The
first cluster has 184, second cluster has 121. If I increase the number of clusters, you will see that the
total will still remain 305. So, the first column here is the cluster index. So, the cluster number 1, 2, 3,
4, 5 as you saw, the second one will also be an integer, it is number of people in each cluster. So, if
you have done your sampling correctly, this is the relative size. So here, clusters one is a little bigger
than cluster two is what you see. What you see here on the other columns is actually, remember, we
spoke about the names of each column, recycle awareness, environment attitude, shopping
convenience, etcetera. The average of those respondents that we have taken in each cluster that the
algorithm has assigned to each cluster. So, these are means, each of these are means or centroids,
as we discussed. So, this is the centroid of the first cluster. This is the centroid of the second cluster.
This is how we interpret this table. Let's take a look at each of these variables again. Recycle
awareness, cluster one mean one greater than mean two. It just means that it has discovered a
cluster where one cluster, the larger cluster, is actually more aware of the recycling. Similarly, and
this is not a surprise really if you look at environment attitude, you observe the same thing. This is
greater than this. If you look at all the clusters actually, you will see it just based on the means you
will see that cluster one seems to be better for your business purposes. Remember, you are trying to
sell an e-waste recycling program. Cluster one is bigger, which means potentially many more
customers. They're also more aware. Now more aware would mean that they are probably likely to
adopt it faster which also, by the way, likely to use the e-waste here shows they are more likely, and
these two are heavily correlated. And if they're more aware then your advertising, actually need not
focus on awareness. It can focus on something else. It can focus on what you're offering rather than
making them aware of the environment. Same thing here, you can look their attitude towards the
environment, it is also higher. Interestingly enough, these other people who seem to be liking
convenience of home shopping, shopping from home, and they're also probably the people who are
more on social media. This is bigger than this and more likely to spread word about environmental
initiatives. This is essentially what your analysis with two clusters is telling you. So, what we have just
done now is segmentation. What is segmentation? Discovering groups of people with similar tastes.
So, you do not want to go after the entire population, but for your business purposes, if you were to
use two segments, what I would say is go after cluster one. Now, this is your target statement. So,
you done segmentation followed by targeting. This is all it means. This one act of cluster analysis has
let you do both segmentation and targeting. Targeting also involves finding out or rather positioning
the next step would involve finding out what kind of marketing variables you will use

Title: K means Clustering – 3 clusters

Now, when you see that there are three clusters, you see that cluster one is now smaller, it has 121.
Now, of course, you're going to have a lot more customers than this hopefully. If you have done your
sampling right these are the relative sizes or roughly you could say that cluster one size is to cluster
two size is to cluster three size are in these ratios. That is all this means just compare the means.
Remember, these are the means within cluster, right or the other name is centroid. If you take a look
at it again, you will see cluster one has high awareness, much more than cluster two or three. Now,
you saw that the relative one and the previous one was six-point something and five-point
something. Here you can see a starker difference. So, you could actually narrow down a little more
with three clusters. Why would you narrow down? Maybe you have limited budgets. You don't want
to focus on people who you think are not going to adopt your product, etcetera. Their environment
attitudes if you look cluster one and cluster two are very similar, but again much higher than cluster
three. Looking for shopping convenience, again cluster one is the highest, very similar to cluster two
and again, cluster three does not seem to be looking for shopping convenience that much. You can
take a look on a lot of parameters. The next two parameters also, you will see that cluster one and
two are very similar but cluster three is highly different. But then, what is the difference between
cluster one and two? Awareness. So, by chance, if you were going to focus on cluster two, I would
still suggest based on this, just based on this data, your target is probably going to be cluster 1. Why?
It's larger, so more potential size. And if you look at it, it seems to score high on almost all of these
parameters. By chance, for some reason, if you had extra data, let's assume that cluster two was
more promising then you have to deal with the fact that the awareness of recycling is not that high,
and you may have to target them accordingly. And by target, I mean, you may have to make ads that
make them more aware, or you may have to send people to talk to them about recycling awareness,
etcetera. They may care a lot about the environment, but probably they don't know that much, as
much as cluster one about how to recycle, benefits of recycling, etcetera. What you will see is that as
you increase the number of clusters, it's obvious, the same number of people are there, right. So,
the cluster sizes themselves... look at the cluster sizes here, they all decrease. At some point, you
may want to be a bit careful because statistical analysis is better done on large samples rather than
small samples. If you have very small clusters, I would not really trust the statistics behind it. The
means. Let's take a look at this data file. Now, this particular person, which cluster does he belong
to? That is something we have not yet done, right. The cluster analysis, I have given you aggregates
of each clusters, but is this person in cluster one or two or three? We don't know. What about this
person? How is this person different from this? Etcetera. This is something we need to know, right?
Every cluster analysis program should, therefore, take each individual that you have in your dataset
and assign that person to one cluster. In this case, we have implemented that by allowing you to
download cluster data, so the same data file will get downloaded again. So, remember, we had the
same file here an extra column called cluster has been added. It tells you that person one has been
allocated to cluster three, person two has been allocated to cluster three, person three has been
allocated to cluster one, etcetera

Title: Final Words: STP So, in the past few videos, you saw the idea segmentation, targeting, and
positioning or the STP framework and how it is central to marketing. It is actually a very important
concept of marketing and we showed you one method, which is a machine learning method called k-
means clustering, which does segmentation of customers based on their perceptual, or needs, or
behavioral data, etc. Now, this is an extremely powerful concept in marketing. The idea that you do
not focus on everyone because everyone is not interested in your products is actually central to
marketing. So, you need to discover groups of people that are similar, then zeroing on some
particular group or groups and finally, make marketing decisions or positioning decisions that will let
you appeal to this group better and do business. So, let me just give you a few examples. If you're on
the road and an auto driver sees you are better dressed. He tries to assume that you're richer and
could probably ask you for higher prices. Based on your appearance, people might talk to you very
differently. That is also a kind of mental segmentation, targeting, positioning, etcetera. You could
also segment people based on your intuition. It won't be so perfect, but you could talk to hundreds
of people and then try to find out groups of people based on your own intuition, people who you
may think have different willingness to pay, who are looking for different things in life, if you are
running a cafe you might try to find out who likes cold coffee and things like that. These are all
informal methods of segmentation. Now, K-means clustering is not the only quantitative method of
doing a segmentation as well. Just take a look at your inbox, if you're using, for example, Gmail and
you will see Gmail tries to do some sort of segmentation. What does it do? If you look at your inbox
now, you will see things like primary inbox, notifications, social media and tabs like that. Basically,
what Google is doing, and Google can do this because it is very powerful computers and very
powerful machine learning algorithms is essentially taking your data and trying to find similar
groups. So, notifications you'll see your OTPs, your bank has sent you a statement, etc. Your reset
password messages from other websites. They end-up in your Notifications folder. On the other
hand, you get this kind of ads, right? Which says, 50% off and XYZ or let's say, some taxi companies
giving you 10% off today, etcetera. You will see all these go into promotions folders, while if you get
notifications from Facebook etcetera, those are going to a Social Media folder. And if you and I
interact, it will probably go to main inbox. So, Google's algorithms go through reams and reams of
text and try to figure out what is optimal. Why is this important? Especially, almost more than 90%
of the world's email today is Spam. So, the Spam folders that you see are again based on similar
clustering algorithms to some extent. You have other kinds of analysis. For example, I am doing a
project right now, which is a little beyond the scope of this core. Where you take newspaper articles
or online... Right now, online blog articles and online news articles and try to classify them as
political, or lifestyle, or entertainment, or health, things like that based on the words that are
appearing. Now, this is also done by a machine learning algorithm, not K-means clustering, but very
similar where it goes through the words and tries to find clusters of words, which are very similar.
So, Trump White House, etcetera, if those come, it will show them as a single clusters. And then I will
take a look at those words and say, this looks political to me. Copyright © IIM Bangalore. All rights
reserved Segmentation again, going back to a pseudo marketing use, segmentation becomes very
important in political campaigns. What kind of person are you going to target with your campaign?
What are their preferences? Are they looking for right-wing initiatives or left-wing initiatives, and
what do you have to offer? Based on all of these, you can again do segmentation, targeting, and
positioning. Many big countries across the world, now use data scientists and market researchers to
form their polls. Now, there are ethical issues associated with this, which we will speak about in the
next week. There are ethical issues associated with a lot of what we do in marketing research, and
marketing in general, which is what we will do in the next week. But segmentation is one of the core
and key concepts of marketing, and this is what we discussed this week

Week 5 : Marketing & Research , Ethics

Learning Outcome Week 05

By the end of this week, you will be able to:

Explain the concept of ethics in marketing research using theories of price

discrimination, targeted advertising and influencer marketing
Video Title: Ethics in Marketing Research

In the last few weeks, we focused on quantitative marketing research techniques like willingness to
pay estimation, survey design, conjoint analysis, cluster analysis, etc. While these are important
technical skills, we must also be aware of various ethical dilemmas in marketing research at various
steps. This week, we will discuss some of these issues. Ethical dilemmas can arise at various stages in
market research. Starting from why a certain project is being done, to designing a survey, to
implementing insights, marketing research in general, and marketing also are fraught with many
ethical issues. Going forward, you will be presented with some actual scenarios. While based on true
stories, we have often masked the name of the actual people, agencies, or companies involved
because our purpose here is not to put blame on any particular entity, but rather have a look at
possible ethical problems in market research. The first scenario here involves a rich, but autocratic
regime in a country called Atlantis. Atlantis is known for two things: A large deposit of natural
resources, and an autocratic regime which has captured these natural resources. This regime has
built a modern country with glittering buildings, fancy cars, and modern infrastructure. However, it
has come under the global scanner for allegedly enslaving poor immigrants from other countries and
cracking down on the press and internal dissidents who are trying to expose various human rights
abuses under its shining façade. The leaders of Atlantis are now concerned at the negative image of
the country and the associated fallout on business investment. It hires consulting company Globex
Associates to suggest how to contain this bad impression. Excited at the prospect of such a
prestigious assignment, Globex suggests that a lot of dissidents use social media and that its
advanced data mining systems can pinpoint the most influential and viral sources of dissent. As I
mentioned this is the completely real scenario with the names of the entities masked. Globex
actually carried out such an exercise and submitted a report to the Government of Atlantis.
Immediately afterwards, a lot of these dissident voices on social media were jailed. Some may have
even been killed. The Globex report was leaked to the media causing outrage worldwide. The top
management of Globex had to issue a formal apology in which they claimed that they had no
knowledge that their work would be used by Atlantis to further suppress individual freedoms. Think
about potential uses of your technical knowhow. Would you be comfortable knowing that your
market research project could be used for purposes you may not like or may not be comfortable
with? Let us go to the second scenario now. You are a manager at Globex Associates’ India office.
Your client, a bank in Mauritius, wants you to segment its customer base to optimize their offerings
like loans, deposits, etc. You have at your disposal, their entire customer database, including
customer complaints, feedback, etc. It will take you about 1000 man-hours to analyze this and come
up with recommendations. However, to increase your own revenue, you make a proposal for 1500
man-hours instead. You also schedule a few data collection trips to Mauritius, which is known as a
tourist paradise. Thus, you can thus mix work and pleasure, ticking off this destination from your
bucket list. Copyright © IIM Bangalore. All rights reserved This scenario is very common in the
market research industry. Some of the decisions you made may not have been in the best interest of
your client. Were the inflated expenses really necessary? Were your so-called business trips adding
any value to your client? As human beings, we often tend to justify things that may benefit us, even
if they need not be in the best interests of our clients. Think about the last time a doctor may have
prescribed unnecessary and expensive tests because he or she was receiving a commission from the
hospital for it. Is your situation similar? We now come to our third scenario. You are a manager of a
franchise of Burgers Inc., a major fastfood joint. At the end of every transaction, customers are asked
to provide feedback about your franchise, in a feedback slip. They rate your store’s cleanliness,
service efficiency, food quality, etc. The purpose of this exercise is so that you can improve your
offerings. You approach every customer and request them to give you good feedback, so that the
head office is pleased with you. If you see poor feedback, you confront the customer and ask them
to change their feedback. If they refuse, you surreptitiously throw away the slips with negative
feedback. You also go online to sites like Trip Advisor and Zomato and insert fake positive reviews
about yourself as well as negative reviews about your competitors. In this case, you seem to be
tampering with data – another kind of ethical violation. The feedback system is in place so that you
may improve your service, but you are tampering with this. Think about it – why are you doing so
and how can it backfire? For one, you are deliberately ignoring or tampering with negative feedback.
If you have systemic problems in place, they will continue to exist, whether or not they manifest in
your survey, and customers will continue to perceive these. Over time, you may actually see
customers avoiding your establishment altogether because of these negative points. This is a
common problem in a lot of businesses. Are you willing to risk the health of your business by
ignoring feedback? Governments often outsource some important services to private contractors. I
was availing of one such crucial service, very crucial to me, where the private service provider
insisted I fill up a feedback form in front of their eyes, before my form was sent out for processing. In
such a scenario, do you think I would tell the truth if I did not get satisfactory service? I don't know
what is going to happen with my form, right. Many government initiatives fail because of such
coercion in data collection by third-party vendors who are anxious to boost their own ratings.
Occasionally, entire governments across the world fail to get re-elected and businesses shut down
because they tried to coerce survey respondents to answer according to their own whims, rather
than seriously collect and act on feedback. A lot of regimes, especially authoritarian regimes have
collapsed simply because data collection agencies were too scared to report real economic figures to
their bosses. I distinctly remember an incident from my undergraduate days. One of my college
professors was angry with the entire class and thankfully not my class because someone had made
negative comments about his teaching in an anonymous survey, and the management had asked
him to address these issues. Rather than working on his material and teaching, this person
threatened to give low marks to the entire class unless the person who gave poor feedback was
outed. He even tried to match the person’s handwriting which was in an anonymous slip with other
people’s handwriting from their notebook. Thankfully, the person who had given the negative
feedback was smart enough to have written in capital letters, so he could not be identified –
however, the entire class suffered. Apart Copyright © IIM Bangalore. All rights reserved from the
professor himself, the college management clearly breached an ethical principle by showing him that
particular feedback slip, rather than summarize findings in a printed document. At IIM Bangalore,
and a lot of other colleges today, feedback is collected online now, with absolutely no provision to
trace the respondent’s identity. The survey results are shared with the instructor after the course
grades are given so that there is no chance that he or she may try to do what my own teacher had
done about two decades ago. Think about sensitive research dealing with online piracy which is
important for companies like Netflix and Amazon Prime. If respondents fear reprisal for sharing the
truth about their behavior, and you know that video downloading is often illegal then they may
never share insights about illegal streaming or torrent downloads. Would Netflix or Amazon then be
making optimal decisions in their own video offerings? Anonymity of feedback, along with integrity
of data and protection of participants is essential to a good market research process. If you are in
charge of designing a survey or other data gathering processes, this is a very important
consideration. If you are a business manager facilitating data gathering, your responsibility is to
ensure the integrity of this process.

Title: Price Discrimination

Welcome back. We will now be discussing another ethical issue in marketing that arises from
segmentation. Often, we see that different segments of people have different willingness to pay. For
example, people in Europe may be used to paying higher prices than in India and certain products
are much cheaper as a result in India. Have you seen some books which explicitly say that they are
forsale only in the Indian subcontinent? Check their prices in the UK, and you may be surprised that
they sometimes cost more than double. Similarly, you may have seen some expensive software
packages coming atspecialstudent discounts. These student discounts can be had at museums,
movie theaters, airlines, and a lot of other places. Thisidea is called price discrimination – the idea
that different groups can be charged different prices based on their willingness to pay. Reasons can
be many. For example, discounts to differently-abled people, senior citizens and army veterans can
be either for goodwill or social welfare. But sometimes it is just the idea that you can get as much
money as possible from a segment – thus bulk buyers get better deals from wholesalers, students in
India get cheaper books than their US counterparts, children often pay concessional fares to travel in
trains and so on. Can price discrimination be a problem though? Think about your local taxi or
autorickshaw driver who refuses to go by the meter. He assesses your income based on your looks,
and charges well-dressed people more. Some companies like taxi aggregators actually build price
discrimination into their algorithms, segmenting you on present circumstances and past behavior
and then charge you different prices. Consider now that you are the CEO of TravelMax, an online
travel booking website in the US. As usual, it is a real company, but its name has been disguised.
Your marketing research manager suggests that it is possible to track each machine’s IP address, and
show higher hotel and airline prices to people living in affluent or posh areas. Additionally, it is also
possible to charge higher prices to Mac and iPhone users, because Mac and iPhone are expensive so
people who have bought them should have higher willingness to pay given their choice of device and
you can track this online. While this may seem perfectly valid and even smart price discrimination
strategy, this is fraught with ethical pitfalls. Such a scenario actually happened in the US a few years
ago, where this strategy was revealed by an anonymous employee of TravelMax in a blog. There was
huge outrage from consumers, with threats of boycotts, and even legal action. So, first of all, do
check if a certain price discrimination policy is legal in your country. So, price discrimination based
on age, sex or race is not only immoral from an ethical point of view but also illegal in most countries
and then there are other issues even if it is legal – how will consumers react if they know you are
doing it? While it is a perfectly acceptable practice to price discriminate based on demand in hotels
and airlines, the idea of discriminating based on device-used for booking, orresidential address was
seen as completely unethical by US consumers. While it was not illegal, it caused huge damages in
public perception of the brand TravelMax and allowed competitors to gain some extra business.
Sometimes, companies are just experimenting with different prices by randomly showing a test
group some higher prices than the base price just to see if they will buy. It is a price calibration
experiment. A major online retailer once got into trouble for doing so, and they also had to issue a
public apology. So, what is the moral of the story? Price discrimination is a fairly straightforward
economic concept in theory but implementing it can have many legal and ethical pitfalls. A lot of
caution and deliberation is advised before embarking on any price discrimination scheme, especially,
in consumer-facing businesses.

Title: Privacy and Targeted Advertising

Let us now discuss another ethical dilemma that arises from segmentation, especially with sensitive
personal data. This is the problem of privacy and targeted advertising. Consider that you have gone
to a big hospital and done a series of medical checkups. The hospital has done a battery of blood
tests, heart rate tests, urine tests, etc. These are sensitive in nature and are usually meant to be
between you and your physician. Now, while you are walking home, you get an SMS from
OnlineMedicines dot com, a hypothetical large e-commerce retailer, saying, “Now that you have just
finished your medical checkups, and the doctor has prescribed X brand of tablets for you, we are
pleased to offer a special 30% discount on tablets X. While it is indeed true that you are getting a
substantial discount on your purchase, it is also true that your privacy has been compromised.
Sensitive medical information seems to have been shared by the hospital with a commercial third
party, who is now targeting you for their profit. This scenario seems to repeat a lot now in India. You
may register for a service somewhere, and your data is being passed on to third parties without your
knowledge. A common form of this is individual merchants who misuse data that they have got from
another service. For example, you may have seen restaurants bombard you with SMS after you
ordered food from them via a third-party food delivery app. They had your number only to
coordinate the delivery, but did you actually authorize them to send you promotional content?
Sometimes, even people issuing your sim card can be bribed to pass on your details to unauthorized
third parties. This is increasingly a problem today. Put yourself in the shoes of a business now.
Should you acquire data through such back channels? It may seem tempting to buy data on the black
market from service providers, including mobile phone service employees and even government
employees who often have accessto important and sensitive personal data like electricity usage,
school enrolment, bank transactions, etc. Think about it for a moment – even if you are not breaking
the law which you often are, would your conscience be ok with such back-channel activities. Now, let
us come to another kind of grey area – where consumers may have given consent to share their
data, but you are still in a moral; grey area. Most online apps have a huge legal agreement where we
agree to terms and conditions without actually reading anything. We have all done this, haven’t we?
Let ustalk about one specific case – that of Facebook and Cambridge Analytica. We will not be
masking these names, as it is a very famous case today. Cambridge Analytica is a political consulting
company that collected some user data on Facebook with users’ permission, using a plugin.
However, in the process, they also collected data about the said users’ friends, using the Facebook
API, which wasthen used for targeted political campaigning in the 2016 Presidential election. So, you
as a user have not only give your data but you have also given data about your friends which they
may not have consented to. The case is extremely high profile, with congressional enquiries in the
US about it, especially as foreigners may have used the data to influence American elections, which
is a serious allegation in any sovereign country! Copyright © IIM Bangalore. All rights reserved While
complete privacy may be neither feasible nor convenient in today’s world, it may be worthwhile to
know exactly what you are allowing your data to be used for. It is useful to know that there is no
such thing as a free meal, and for every service you are using, you are giving away valuable data to
be used for targeting. If you are someone selling this data, say from an app or a service you have
built, it is worthwhile to introspect yourself on the ethicality and legality of your actions. Going back
to the first video of this week, think about the user of your data – what purpose will they use it for?
More importantly, would you be comfortable if your own data were used the way you are allowing
other people’s data to be used? Finally, think of yourself as the targeted advertiser. The person who
is sending out the targeted ads. For all the hype, targeted advertising may often not be the best
strategy. A major US retailer once used very advanced statistical methods to determine that some
women were pregnant, even before the women themselves knew it. Apparently, some subtle body
changes meant that these women changed their shopping patterns without knowing it themselves.
This retailer then sent out targeted offers, saying congrats, you are pregnant, etc., which caused
quite a bit of controversy as these ads were perceived to be intrusive to the point of stalking. Would
you want your own brand to be seen as a stalker? For all the hype about hyperlocal targeting, how
much is too much is always a big concern. Finally, again think, are you using data that you have
ethically sourced? Are you using it in a manner that you would be comfortable with? These are
important questions that you need to consider while doing targeted advertising.

o Title: Influencer Marketing Imagine the following scenario now. You follow a popular
personality on Twitter or Instagram, as they are experts on a topic that interests you. For example,
let's say you are a food enthusiast, and you follow a popular food blogger. Now this person says that
he went to a restaurant and was wowed by the food, décor, and service. You respect this person’s
opinion because you follow this opinion for the expertise and decided to go to this restaurant. Now
imagine that you realize that this person was paid by the restaurant to post this so-called
spontaneousrecommendation. Would you be so happy knowing thatthis was notreal word of
mouth? Instead of a food blogger, think of a doctor who prescribes you medicines because they are
getting commissions from brands rather than what you absolutely need. Would you feel ok? A lot of
these people are called influencers – people who have an influence on consumers’ purchasing
behavior. Some of these are experts in a given domain like gaming, food, fashion, etc., while others
may just be popular actors, models, or other celebrities. Some influencers are even just attractive
people who post-holiday pictures of themselves and occasionally slip in promoted content.
Influencer marketing is an ethically grey area in marketing when the financial relationship between
the brand and influencer is not disclosed. Influencer messages often post as word of mouth
recommendations rather than ads. In a way, it thus misuses the trust placed between the social
media user and the influencer. However, many brands explicitly do not want this relationship to be
disclosed, because word of mouth works differently than advertising. In fact, they tell these so-called
influencers not to disclose that they were paid to make these recommendations. The US federal
trade commission actually has rules for influencer marketing today. It mandates that influencers put
a hashtag like #paid or #sponsored when putting out paid content. Instagram even has algorithms
that puts these tags automatically if it suspects that some content may be sponsored. In India, there
is not yet such an explicit law at least to the best of my knowledge, but many ethical influencers
declare their relationship with brands that are sponsoring them. However, many do not. There have
been stories of influencers threatening brands to pay them either money or free gifts like hotel stays
and meals or get poor reviews so either give me a free meal or I will bad mouth your restaurant on
Zomato that kind of thing. So, online platforms like Zomato even try to detect such unethically
posted reviews by suspending offending influencers. So, if you have paid reviews on Zomato and
Zomato finds out they will suspend your account, they have done it to multiple people. The idea is
simple – word of mouth platforms like review sites and social media would like to sanctity of
reviews, and not let unscrupulous people benefit unethically. If you are a brand using influencers,
think about disclosing your relationship with the influencer – if consumers get to know that you have
been doing something underhand, not only will they distrust the influencer, but this distrust may
spill over to you. Do you want that as a brand? If you are an influencer, think about the trust you
have built with your followers – would you be ok exploiting this relationship by pushing undisclosed
sponsored content. Finally, as a consumer, take suspicious reviews with a pinch of salt. A lot of what
seems to be word of mouth today is actually sponsored content out there. So be careful before
taking word of mouth recommendations.