Good and Bad Market Research: A Critical Review of Net Promoter Score

Abstract

Net Promoter Score, touted as the “single customer metric you need” and calculated from customers' answer to one simple question about their loyalty, has been in use since 2003 and adopted in a wide variety of settings. However, it has not lived up to its claimed benefits. This article evaluates the NPS approach in terms of its positive and negative results.
This article is for people interested in NPS, still considering implementing NPS in their company, or interested in its technical underpinnings. It points out the benefits and shortcomings and explains why, and it describes what can be done to achieve the outcomes NPS theory claimed it would produce, but has not.
The article is written in two parts for quite distinct audiences: firstly, for executives and managers who need customer data and information to make marketing decisions; and secondly, for market researchers, statisticians, and business analysts who are responsible for capturing and providing reliable, understandable, and meaningful customer data to the executives and managers who need the information. Consequently, the two sections are written in two different styles. The first section takes the form of a summary for managers and executives of our findings and recommendations in language aimed at business leaders; the second section provides a detailed analysis and critical review of NPS for market researchers, statisticians, and business analysts. Both sections present a better solution than NPS for understanding what customers value, delivering the best value to customers, winning market share, and creating truly loyal customers.

Full text

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
1
Version: 23 October 2018
Good and Bad Market Research: A Critical Review of Net Promoter Score
1
Nicholas I. Fisher
2
and Raymond E. Kordupleski
ABSTRACT
Net Promoter Score, touted as the “single customer metric you need” and calculated from
customers’ answer to one simple question about their loyalty, has been in use since 2003 and
adopted in a wide variety of settings. However, it has not lived up to its claimed benefits. This
article evaluates the NPS approach in terms of its positive and negative results.
This article is for people interested in NPS, still considering implementing NPS in their company,
or interested in its technical underpinnings. It points out the benefits and shortcomings and
explains why, and it describes what can be done to achieve the outcomes NPS theory claimed it
would produce, but has not.
The paper is written in two parts for quite distinct audiences: firstly, for executives and
managers who need customer data and information to make marketing decisions; and secondly,
for market researchers, statisticians, and business analysts who are responsible for capturing
and providing reliable, understandable, and meaningful customer data to the executives and
managers who need the information. Consequently, the two sections are written in two
different styles. The first section takes the form of a summary for managers and executives of
our findings and recommendations in language aimed at business leaders; the second section
1
 Nicholas I Fisher & Raymond E Kordupleski, June 2018
2
Nicholas Fisher is Visiting Professor of Statistics, School of Mathematics & Statistics, University of Sydney, NSW
2006 AUSTRALIA (email Nicholas.Fisher@sydney.edu.au), and Principal, ValueMetrics Australia; and Raymond
Kordupleski was formerly AT&T Customer Satisfaction Director; now retired in Louisville, Colorado, 80027 USA. The
authors thank Stephen Sasse for helpful comments.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
2
Version: 23 October 2018
provides a detailed analysis and critical review of the NPS for market researchers, statisticians,
and business analysts. Both sections present a better solution than NPS for understanding what
customers value, delivering the best value to customers, winning market share, and creating
truly loyal customers.
KEYWORDS Customer Value Management, Market Research, Satisfaction surveys, Transaction
surveys

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
3
Version: 23 October 2018
1. Summary for Executives and Managers
Net Promoter Score
3
. This is a splendid term for marketing purposes. It resonates similarly to
Net Income, a very well-known financial term that all executives know and respect. From the
name itself, Net Promoter sounds as if it must be important. And when you add the business
consultants’ sales pitch that “It’s the only number you ever need to succeed in the marketplace”
it’s extremely tempting to buy into the concept. But should you?
The answer is simple. No. A little critical thought can provide the answer. Suppose a business
consultant or academic or research paper from a leading business school asserted that:
“Net Income is the only number you will ever need to manage your business, forget
cash flow, assets, income analysis, production costs, process improvement cost,
investments, liabilities, share price, dividends, interest rates, etc. Stop spending
needless money on expensive accounting, reporting, tracking, and analysing these
unnecessary metrics.”
You would tell the people – politely or perhaps not so politely – to leave the premises. And you
would pitch the research paper into the recycling bin.
Everyone knows you cannot manage and improve the financial results of your business with
only one financial marketplace number such as Net Income. If you tried, you would not be in
3
If you don’t know how it’s actually calculated, here is its original definition. After an interaction with a company’s
products or services, people are asked “How likely is it that you would recommend our company to a friend or
colleague?” Based on their responses on a 0 to 10 rating scale, group the respondents into “promoters” (9–10
rating – extremely likely to recommend), “passively satisfied” (7–8 rating), and “detractors” (0–6 rating – extremely
unlikely to recommend). Then subtract the percentage of detractors from the percentage of promoters.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
4
Version: 23 October 2018
business very long. Why would anyone think you could manage and improve your results in the
customer marketplace with one customer number, Net Promoter Score?
Yet many businesses are trying this simplistic approach. Remember how childish it was to do
something just because everyone else was doing it and it was the latest thing, only to learn
there is a wrong way and a right way. There are no shortcuts.
In this summary we will cover five problems with NPS and recommendations to avoid them:
a) NPS provides no data on what to do to improve
b) NPS focuses only on keeping customers, not on winning new customers
c) There is no such thing as a “passive” customer
d) NPS provides no competitive data
e) NPS is internally focused not externally focused
a) NPS is an indicator of how you are doing, but provides no data to help you know what to
do
NPS is one way of calculating one customer loyalty score. It’s important to know the percentage
of customers who are loyal, will buy again, and recommend. And as one of a number of loyalty
metrics, it may well be useful. But it is unsatisfactory for business decision-making because it
does not tell you why they are loyal or disloyal. NPS does not provide any statistically reliable
data and information on what it is that customers value, so that you can both retain them and
win additional customers. A customer loyalty metric helps answer the question, “How are we
doing with our own customers?” But it gives no information to help you answer the question,
“What should we do?” let alone “What should we do and in what order?”

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
5
Version: 23 October 2018
Currently, Bain and Company has two meanings
1
for the acronym NPS. One is the classical
metric called the Net Promoter Score. According to Bain, the new meaning of NPS is the Net
Promoter System. It was developed because the NPS score itself has no value for helping
companies decide how to improve. So Bain needed to improve the usefulness of their NPS
metric. The Net Promoter Score is based on the concept of finding out if your customers are
loyal. Nice to know, but not helpful. The new Net Promoter System is based on a follow-up
customer inquiry process when, if they don’t say they love you, ask more questions and find out
why not? What originally was promoted as a simple “all you need to know is the answer to one
question” system has morphed into an expensive and intricate gathering of anecdotal
information by personnel at all levels of the company, leading to a complex summary of non-
statistically reliable customer complaints. Whilst modern text mining and machine learning
techniques can find nuggets in such material, the data are by definition low-level, operational
(at the level of the user), based on a transitory experience and not representative of the
decision-makers in the market.,
b) NPS focuses on keeping customers not on winning new and retaining old customers
Keeping customers is very important. But without winning new customers a business cannot
grow. No matter what, you will always lose some percentage of customers. Winning new
customers is imperative. NPS is one reasonable customer loyalty metric, but is not good enough
for business success because it is not a competitive metric. A measure directly linked to and
predictive of winning customers is critical.
What drives customer growth? Simply stated, it is customers choosing your products and
services over the competition’s products and services. Revenue growth or decline is driven by

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
6
Version: 23 October 2018
customer perception of the worth of your products and services compared to the competition.
It’s all relative, and it’s all perception. NPS does not measure the perceived worth of your
products and services compared to that of your competition.
c) NPS provides no competitive data
NPS is not a competitive measure … but the marketplace is all about competition! Customers
have all the votes and they vote with their dollars for the products and services that they feel
are worth it. A metric that accurately measures customer perceptions of the relative value of
one firm’s offer compared to the alternatives is what is critical but missing with NPS.
There are numerous examples of companies having high or increasing customer loyalty while
losing customers and not winning new customers. One example is Oldsmobile. Oldsmobile,
founded by Ransom E. Olds in 1897, was one of the oldest companies in the world before it
went out of business in 2004. At that time, its remaining old customers were extremely loyal to
the Oldsmobile brand. Oldsmobile had customer surveys and measured customer loyalty by
tracking the percentage of customers responding with “very willing to buy again” and/or “very
willing to recommend” on their surveys. The data were solid and predictive of repurchase rates.
Over time, the percentage of their customers who were loyal increased, but total sales revenue
decreased! How could that be? The answer was simple math. The increase in percentage of
their customers who were loyal (number of loyal customers divided by the total number of
customers) was a result of the loss of the customers who were not loyal and left for the
competition. Only the loyal customers were left. The decrease in the denominator led to the
increase in the percent of loyal customers. Their key customer satisfaction metric was
misleading. Even if Oldsmobile had been using the NPS metric, it too would be misleading. The

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
7
Version: 23 October 2018
NPS score will improve as the number of detractors decreases because they are no longer
customers available for surveying. Not only will detractors leave, so will a significant amount of
the “passives”. So the percentage of total customers surveyed who are promoters can easily
increase as you lose the “detractors” and a fair amount of “passives.”
d) There is no such thing as a “passive” customer
The NPS concept puts customers into three categories, Promoters, Passives, and Detractors.
The NPS categorisation theory is absolutely correct about the promoters and detractors, but
absolutely wrong about the passive category. Promoters are loyal and will speak favourably
about a company. They are not looking for a better, more valuable alternative. Detractors are
disloyal and speak unfavourably about a company. They are looking for a better offer and will
leave for the competition. However the NPS concept of “passive customers” is completely
erroneous. These customers may be passive about speaking favourably or unfavourably, but
they are very willing to shop for better value. Good is not good enough. There is no such thing
as a passive customer. By definition, passive customers are not loyal customers. They are very
willing to consider competitors and readily switch should they be presented with better
competitive value.
Keeping customers is important. But winning new customers is also very important because
growth is important. To win new customers, a firm must beat the competitors’ offers. NPS does
not measure a firm’s competitive customer perception. Let’s go back to the Oldsmobile
experience. New younger customers were entering the automobile market. They were
shopping for the best car at the best price. Oldsmobile worked very hard to improve its cars
and indeed they made major improvements. But in spite of their continuous improvements,

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
8
Version: 23 October 2018
their cars were not perceived to be the best value in the automobile marketplace. Oldsmobile
continued to lose market share in spite of their older customers’ loyalty to the car they had
driven most of their lives. Young potential customers were not buying the much improved car.
Oldsmobile invested in a major marketing and advertising campaign. The slogan and theme
was “This is not your father’s Oldsmobile!” It was true enough, but their new cars were not as
good and not worth as much as other cars in the market. Before too long, General Motors killed
the Oldsmobile brand, in spite of their remaining customers’ fierce loyalty. After 107 years,
competition killed Oldsmobile.
e) NPS is internally focused not externally focused
The NPS metric and the NPS system are still internally focused. They are based on the notion
that the purpose of a business, its reason for being, is to create profit for the business and its
shareholders, and that customers’ loyalty is required to do so. It preaches and teaches the
concept that we need customer loyalty because the purpose of a business is to make money
and create shareholder value.
By way of comparison, the Customer Value
4
system is based on the principles of understanding
what customers value and using that knowledge to manage your systems and process to deliver
greater value than is offered by any competitor. By delivering a superior product and service at
a fair price a firm attracts, wins, and retains customers.
The Customer Value concept is based on the truth that the real purpose of a business ― the
reason a business is even allowed to exist ― is to fill customer needs and to improve customers’
4
Described in the next section of this article.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
9
Version: 23 October 2018
quality of life with its products and services. Shareholder value is the reward for fulfilling
customer needs the best. Customers vote with their pocketbooks as they evaluate the products
and services in the marketplace. To win their vote a business must provide products and
services that are worth what they ask customers to pay and are worth more than what your
competition is offering. Tomáš Baťa, founder of the great Baťa Shoe Company, recognised this a
long time ago, when he said
5
: “Do not pursue money. He who pursues money will never achieve
it. Serve! If you serve as best as you can, you will not be able to escape money.”
Senior leaders and bosses who preach “we need all customers to be very willing to recommend
us” set an internal focus. They measure and manage their employees on customer loyalty
metrics. They want customers to be loyal to the firm (them) because customers create value for
the business (them). But these executives have the cause and effect backwards. The customer is
the real boss. Customers demand value from the company before they are willing to give their
loyalty and money to the company.
The new Net Promoter System has it backwards too. It also puts loyalty to the firm over a firm’s
loyalty to the customer. The “system” first ask the customer if they are willing to recommend.
Then for customers who are not very willing to recommend, the “system” probes further and
asks customers why not? That is too late. It’s much better to first find out what is important to
customers and design your products, services and business processes to satisfy their needs.
Think of it this way. Personal relationships are rewarding and lasting when each person puts the
other’s needs first rather their own first. Each person understands what is important to their
5
https://en.wikiquote.org/wiki/Tom%C3%A1%C5%A1_Ba%C5%A5a

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
10
Version: 23 October 2018
partner and does what’s necessary to fulfil their partner’s needs and desires. To ensure they
do, they ask for occasional feedback with sincere questions on how satisfied their partner is. In
return they receive love, respect, and a rewarding relationship. The Net Promoter System
essentially puts the customer’s loyalty to the firm over the firm’s loyalty to the customer. To
maintain great customer relationships and a strong vibrant business, focus on the customer.
Or, put more simply, NPS is measuring what customers do for you. CVM is measuring what you
do for customers.
Thus there are no shortcuts to winning and keeping customers, but there is a proven path.
We conclude this section with a comparative summary of the two methods:
NPS approach
CVM approach
 A customer satisfaction metric
 A marketplace perception survey
 Triggered by recent events
 Not triggered by an event, but measures the
general market perception at the time
 Taken of the person who had the recent
experience, even if they are not decision
makers or purchase-influencers
 Surveys decision makers and key influencers
who determine which company they buy
from
 Does not survey competitors’ customers
 Measures both the sponsoring company’s
value and that of the competitors
 Only focused on the recent transaction and
experiences, no matter how unimportant or
important the event.
 Studies and provides data on the market’s
perception of the competitive value of a
company’s products, services, image and
prices which influence purchases.
 Used by middle managers to measure
performance of service centres such as
technical support, and individuals such as
service reps.
 Used by senior managers to determine where
to focus a company’s scare resources to
improve the competitive value of their
products services.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
11
Version: 23 October 2018
2. Technical evaluation of Net Promoter Score
2.1 Introduction
Here is a scenario from everyday life.
You’ve just completed a small online transaction. Before you log out, you are presented with a
request to answer a one question customer survey:
On a scale of 0 to 10, where 0 = Not at all and 10 = Definitely, please rate your
willingness to recommend us to others.
As a statistician or market researcher, this should trouble you.
This familiar scene is an example of the use of NPS (Net Promoter Score), a ubiquitous metric
collected routinely by companies after most transactions. We contend that statisticians should
be taking a stand against the use of NPS as a panacea for the problems it purports to address,
including its use as a stand-alone metric in assessing customer satisfaction or staff satisfaction.
To expose NPS’s failings, we shall focus primarily on a customer (rather than a staff member) as
the stakeholder being surveyed. The customer may be a consumer, or may be a corporation.
We will be making a critical distinction between two different types of survey respondents: on
the one hand, the person who actually makes the purchasing decision about a product or
service; and on the other, the person who actually uses it. These may be coincident, or maybe
not. Thus, in a corporation, the purchasing decision about an automobile may be made by a
senior executive, but the transaction carried out by an administrative assistant. In the
consumer world, a purchasing decision about a teenager’s first car may be made by a parent

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
12
Version: 23 October 2018
(taking account of price, safety, …) whereas the teenager may be making the actual choice of
vehicle and have much of the experience of interacting with the vendor.
Section 2.2 outlines key goals of customer satisfaction surveys, and provides a brief description
of a long-established and proven process (Customer Value Management, or CVM) to achieve
these goals. Section 2.3 gives a description of NPS and evaluates it as market research metric.
Section 2.4 provides some concluding remarks. Full details of how statistical modelling and
analysis for CVM are carried out can be found in Kordupleski (2003) and Fisher (2013, 2019) and
so will not be discussed here.
2.2 Carrying out market research: Why and How?
The customer survey process is an important means of managing a company’s relationship with
customers. Let’s start by asking why we carry out market research. Here are some of the main
reasons:
a. Find out what's important to people about the product or service they are seeking to
purchase.
b. Ascertain what people think about your company’s offerings.
c. Obtain timely feedback about what you need to fix, and with what priority.
d. Detect changes in the market — e.g. emerging preferences, styles, technologies, ….
e. Find out how the competition is viewed.
f. Ultimately, improve your business bottom line!
If these are important, then there are strong implications for how the market research should
be carried out:

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
13
Version: 23 October 2018
a. Use a statistically sound method that provides assurance that no attributes of the
product or services that are important to the customer have been omitted from the
survey, and produces reliable data.
b. Have a means of linking survey results to higher-level business drivers.
c. Ensure that the survey results are actionable, including the ability to drill down.
d. Ensure that there is a means of identifying where to focus improvement priorities in
order to have the greatest beneficial impact on both customers and the business bottom
line.
e. Create a design that will provide comparable and useful benchmarking metrics.
A process to design and conduct customer satisfaction surveys that meets these criteria was
developed some 30 years ago at AT&T … in response to a crisis. The crisis is described in detail
by Kordupleski (2003, xv et seq.), and subsequently summarized in Fisher (2013, Chapter 4) as
follows:
In the mid-1980s, AT&T was confronted by a paradox: on the one hand, customer satisfaction
levels were running at about 95%; on the other hand, they lost 6% market share, where 1% was
worth $600,000,000. For the first time in corporate history, AT&T laid people off — 25,000
worldwide from an overall staff of 300,000 — including managers recently rewarded for the
apparently outstanding customer satisfaction performance.
An AT&T trouble-shooting team discovered that one of the critical factors explaining the paradox
was the way in which Customer Satisfaction was being measured …

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
14
Version: 23 October 2018
AT&T assembled a team to find out why there was no apparent connection between customer
satisfaction (95%) and business performance (market share down 6%), and to fix it. As
described by Kordupleski (2003), this team identified three core issues:
 The first was what they were doing with their raw data. Customer satisfaction was
being measured on a 4-point scale: Poor, Fair, Good and Excellent. The Good and
Excellent responses were being combined into a single Satisfied Customer category,
giving rise to the 95% score for customer satisfaction. This was a major mistake. Of
those who had rated them Excellent, almost all were very willing to repurchase from
AT&T. In contrast, of customers who had rated them Good, some 40% were not very
willing to buy again and were shopping for an alternative provider.
 The second mistake was to not benchmark their customer satisfaction scores against
those of the competition. After all, business is a competition and customers have
choices. If competitors do a better job at satisfying customers you will lose market
share.
 The third and most important mistake was a failure to focus on Value as the ultimate
metric, where Value was defined as the trade-off between people’s satisfaction with
the Quality of the product or service they were receiving balanced against their
satisfaction with the Price paid. In simple words, did customers perceive the products
and services received to be “worth what they paid”? AT&T developed a value metric
called CVA or Customer Value Added and deployed a process called Customer Value
Management (CVM). It lead to a major turn-around in AT&T’s fortunes (an increase of 7
points of market share in a year and a half as reported by the Wall Street Journal) and,

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
15
Version: 23 October 2018
subsequently, the fortunes of many enterprises world-wide who learned and adopted
the approach . Fuller descriptions of the following very terse summary of CVM can be
found in Kordupleski (2003) and Fisher (2013, 2019).
At the heart of the CVM toolset is the Customer Value Tree. Let’s take as an example the
purchase of an automobile. The overall concept of Value (Worth What Paid For) is modelled as
having two principal drivers, Quality and Price, each of which can also be elaborated, as shown
in Figure 1.
The Customer Value tree forms the basis of a survey of the market (both your customer and
those of your competitors); importantly, it is a survey of people who make the purchasing
decisions. The Customer Value tree also forms the basis of the report of the results displayed in
a simple to understand and use format. Respondents are asked to rate the performance of their
supplier on Automobile Attributes (on a scale of 1 to 10, where 1 = Poor and 10 = Excellent).
Then they are requested to provide an overall rating of the Automobile, together with the main
reason for assigning this overall rating. The rating process continues for the whole tree, up to
an overall rating of Worth What Paid For. At this point, it is useful to request higher-level
ratings of business impact, such as Willingness to repurchase or Willingness to recommend
your company to someone else. Also, after being asked for a summary rating for each main
branch, a respondent is invited to provide reasons for assigning this rating.
Thus the overall response from a respondent takes the form of a tree-structured set of ratings.
A sample of such tree-structured data can then be analyzed by a fitting a sequence of
hierarchical regression models: Automobile as a function of its Attributes, …, Delivery Process

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
16
Version: 23 October 2018
as a function of its constituent sub-processes, …, all the way up to Value as a function of Quality
and Price. This modeling process yields two critical sets of information:
 the hierarchy of fitted models, which provide confirmation, or otherwise, that no
important factor affecting the market’s overall perception of Value has been omitted;
and
 for each model, the relative rating of each explanatory variable and its impact weight.
Figure 1. A prototypical Customer Value tree for buying and using an automobile. Value (Worth What
Paid For) is represented as having two main Drivers, Quality and Price. Quality has as its Drivers the
Product (in this case, an automobile) and Delivery Process – the sequence of experiences (service sub-
processes) when the customer interacts with the supplier. Automobile, Direct Costs and Indirect Costs
each have 6 – 7 Attributes determined from market focus groups. In some cases, Brand Image may be
sufficiently important as to be elevated to the same status as Quality and Price as a Driver of overall
Value.
The first endows this approach to perception surveys with a unique advantage over other
approaches. And the second provides the basis for a very powerful management decision
process to focus improvement efforts. We sketch a simple synthetic example.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
17
Version: 23 October 2018
Table 1 shows the top-level profile for your company and the average of your competitors.
Overall, you are somewhat below par on Relative Value and Relative Satisfaction with Quality,
and around par on Relative Satisfaction with Price.
So, we look at two issues:
 how to connect the current rating of Value to business performance
 how to use the results to select and act on improvement priorities
Impact
weights (%)
Mean ratings (± 0.2)
Relative
rating (%)
Driver
Our company
Competitors
Quality
51
7.4
7.7
96
Price
35
7.1
7.0
101
Value
(R2 = 81%)
7.3
7.5
CVA = 97
Table 1. Top-level table of impact weights and comparative ratings for Value and its main drivers. The
Relative Value metric is known as CVA (Customer Value Added). (Note that the weights sum to R2, not to
100. Quality and Price do not totally explain Value.)
Figure 2 is an example of a Loyalty curve. Recall that we had collected respondent data on
Willingness to recommend. Conventionally, a rating of 8, 9 or 10 is regarded as in indicative of
a respondent being very willing to recommend your product or service (although it could be
defined as just a rating of 9 or 10). Figure 2 indicates that the current Value score of 7.3
corresponds to about 63% of your customers being very willing to recommend. To see an
increase in this figure to, say, 80% will require an increase in the Value score to around 7.8.
A critical issue in applying CVM is that at any one time, we only look for a few high-priority areas
for improvement. We can look at less important things next time after the ones with the
greatest impact on the business (as judged by the change in Figure 2) have been attended to.
So, where should improvements be focused in this case?

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
18
Version: 23 October 2018
Figure 2. The current Value rating of 7.3 corresponds to just 63% of your customers being very willing to
re-purchase from you. If you want this to be at least 80% within 12 months, you’ll need to raise your
Value score to about 7.8.
We have a statistical model for Value as a function of Quality and Price: Value = 0.51  Quality +
0.39  Price. Of course, the model is imperfect: 19% of the variation is unexplained. However,
if we’re just looking for the most important improvements, the following course of action has
proved effective in many CVM applications. Suppose that by working on improving Quality, and
communicating these improvements to the market, you can achieve an increase in the overall
rating of Quality of 0.6 when you re-survey the market in 12 months’ time. (An analogous
analysis for Price is also appropriate.) Then the predicted increase in Value would be simply
0.51  0.6  0.31. (Of course, this argument assumes that the mentioned statistical model is not
merely association between Value and Quality ratings, but a causal link, and such a link typically
cannot be established through data fitting alone. The use of the model in this way has proved
useful in numerous case studies; e.g. Kordupleski 2003, Fisher 2013, 2019.)
Table 2 shows the corresponding profile for Quality –

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
19
Version: 23 October 2018
Impact
weights (%)
Mean ratings (± 0.2)
Relative
rating (%)
Driver
Our company
Competitors
Automobile
39
7.8
7.5
104
Del. process
59
6.9
7.8
88
Quality
(R2 = 89%)
7.4
7.7
96
Table 2. Profile of impact weights and comparative ratings for Quality and its main drivers.
– and drilling down further we get Table 3, the profile for the Delivery process and its sub-
processes. It appears that there are serious problems for customers interacting with your
finance department. Now you know where to focus attention.
Impact
weights (%)
Mean ratings (± 0.2)
Relative
rating (%)
Driver
Our company
Competitors
Initial contact
…
…
…
…
…
…
Billing
40
6.1
7.5
81
Delivery process
(R2 = 86%)
7.4
7.7
96
Table 3. Profile of impact weights and comparative ratings for Delivery process and its sub-processes.
The actions to take are:
 Develop a lower-level tree for Billing (Figure 3), and put in place some internal process
metrics to track improvements. Note that, in contrast with a Value survey which is
focused on decision-makers, this survey is actually focused on actual users, a critical issue
when we come to clarifying the characteristics of NPS.
 Carry out a Transaction survey, focused just on your own customers, to ascertain where
the specific issues are with Billing.
 Make appropriate improvements, using the internal metrics to confirm stabilization of
the Billing process and monitor it; communicate them to the market, and re-survey.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
20
Version: 23 October 2018
Figure 2. Billing tree for a transaction survey, together with internal metrics that can be tracked to
monitor improvements.
We are now in a position to capture the CVM continuous improvement process in a simple
diagram (Figure 4).
Figure 3. The CVM improvement cycle.
It is easily verified that this process meets the five criteria (a) – (e) outlined at the beginning of
the Section (but see Section 4 for further discussion of the benchmarking issue). In summary,
then, CVM is a robust, rigorous and proven process for gaining and sustaining a company’s
competitive position in the market. See Kordupleski (2003, 2018) for numerous applications. It
is based on an ongoing cycle of continuous improvement, building and sustaining the
relationship between the company and its market by monitoring and responding to current and
emerging market needs and its competitive performance in relation to these needs.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
21
Version: 23 October 2018
It may be helpful to explain a few more details about the CVM process before passing to a
comparison with NPS.
(a) Typically, a representative sample of at least 30 – 50 respondents is needed to obtain
sufficiently precise results to be actionable.
(b) An online CVM survey takes some 10 – 12 minutes to complete, including providing
comments. (By way of contrast, an NPS request might take just seconds to respond to, or
perhaps a minute if a lengthy comment is being added. Actually, a CVM survey will typically
include an NPS-type request as just one of its 25 – 30 rating requests.)
(c) In the CVM approach, the quantitative data are used to identify where to make
improvements that are likely to have the most beneficial impact on the business. Once
these areas have been selected, the comments can provide valuable insight into what might
need to be fixed.
(d) The statistical analysis requires the usual level of practical skills as do other areas of
application. For example, not all respondents complete all requests in a CVM survey.
However, the sort of imputation techniques commonly used in regression and other settings
can be applied to handle such complications.
Full details are available in Fisher (2013, 2019).
2.3 Comparison with Net Promoter Score
Net Promoter Score (hereafter NPS) was introduced by Reichheld in a 2003 issue of Harvard
Business Review, in an article entitled “The One Number You Need to Grow”. The article started
out with a striking assertion:

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
22
Version: 23 October 2018
If growth is what you are after, you won’t learn much from complex measurements of customer
satisfaction or retention. You simply need to know what your customers tell their friends about you.
The full description of NPS in the article reads as follows:
Asking a statistically valid sample of customers “How likely is it that you would recommend our company
to a friend or colleague?” enables you to calculate your Net Promoter score: the ratio [sic] of promoters to
detractors.
Based on their responses on a 0 to 10 rating scale, group your customers into “promoters” (9–10 rating –
extremely likely to recommend), “passively satisfied” (7–8 rating), and “detractors” (0–6 rating –
extremely unlikely to recommend). Then subtract the percentage of detractors from the percentage of
promoters.
And readers were advised of the very substantial benefits likely to flow:
Many companies – striving for unprecedented growth by cultivating intensely loyal customers – invest lots
of time and money measuring customer satisfaction. But most of the yardsticks they use are complex,
yield ambiguous results, and don’t necessarily correlate to profits or growth.
The good news is: you don’t need expensive surveys and complex statistical models … By asking this one
simple question you collect simple and timely data that correlate with growth. You also get responses you
can easily interpret and communicate. Your message to employees – “Get more promoters and fewer
detractors” – becomes clear-cut, actionable, and motivating, especially when tied to incentives.
Leaders of numerous enterprises that were using rigorous ongoing market research to gain and
maintain competitive advantage were attracted by the twin prospects of massive savings and
greatly improved business performance, cancelled their market research campaigns, and signed
up to NPS. For example, at the time of writing, Australia’s four largest banks are literally in the
dock (in the form of a Royal Commission into Misconduct in the Banking, Superannuation and

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
23
Version: 23 October 2018
Financial Services Industry), because of their treatment of their customers. The principal metric
banks were using to judge their performance with customers was NPS. As is evident from the
definition of NPS, it is not customer friendly. For example, simply by discarding their detractors,
a bank can improve its NPS rating yet it has done nothing to improve how it treats customers.
The interim report of the Royal Commission has also commented on the unsatisfactory nature
of NPS as a customer satisfaction metric.
Much of Reichheld’s article amounts to a simple statement of the obvious: the importance of
customer loyalty – whether expressed as repeat purchasing of the same product or service,
purchasing across the range, or recommendations to others – and how this “correlates to”
higher-level business drivers such a market share and profitability. However, there is some
space devoted to debunking other approaches. For example:
One of the main takeaways from our research is that companies can keep customer surveys simple. The
most basic surveys employing the right questions – can allow companies to report timely data that are
easy to act on. Too many of today’s satisfaction survey process yield complex information that’s months
out of date by the time it reaches frontline management. Good luck to the branch manager who tries to
help an employee interpret a score from a complex weighting algorithm based on feedback from
anonymous customers, many of whom were surveyed before the employee got his current job.
Where do we find fault with NPS?
i. Many examples in Reichheld’s article are of operational symptoms, without insight into
root causes (systemic or otherwise). See for example the first main paragraph in the
second column on page 4:
Even the most sophisticated satisfaction measurement systems have serious flaws. I saw
this first hand at one of the Big Three car manufacturers. The marketing executive at the

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
24
Version: 23 October 2018
company wanted to understand why, after the firm had spent millions of dollars on
customer satisfaction surveys, satisfaction ratings for individual dealers did not relate
very closely to dealer profits or growth. When I interviewed dealers, they agreed that
customer satisfaction seemed a reasonable goal. But they also pointed out that other
factors were far more important to their profits and growth, such as keeping pressure on
salespeople to close a high percentage of leads, filling showrooms with prospects
through aggressive advertising, and charging customers the highest possible price for a
car.
Indeed, as our introductory scenario suggested, an invitation to provide an NPS rating
can be triggered by the most low-level of customer experiences. Thus, an administrative
assistant using an Enquiry service (e.g. Figure 3) who was frustrated by being placed on
hold for a lengthy period might vent that frustration by a very low NPS rating when
asked to say on the line at the end of the call and answer a question. Does a company
really want to have its whole performance judged by an instant of minor irritation? The
contrast with a Customer Value survey is critically important:
 In a Customer Value process, the person being surveyed is a decision-maker.
The respondent is asked to rate all aspects of the customer experience before
assigning a rating for overall Value. Only at that point is the respondent asked
about Willingness to Recommend.
 In an NPS process, the person being surveyed is typically a user. Rather than
being led through the whole customer experience before being asked about
Willingness to Recommend, the respondent is being asked to make a decision
based on one particular interaction with the company, no matter how trivial.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
25
Version: 23 October 2018
There is no overall judgement of Value – after all, this is a transactional moment,
and the user may well not be the decision-maker – and so no way to get at
Relative Value, which is the quantity that has a well-established track record as a
lead indicator of business success (Kordupleski 2003).
As noted in Section 1, an NPS request is usually accompanied by a request for a
comment on the reason for assigning a particular Loyalty rating. These
comments can yield a large amount of unstructured data which, through
application of text mining, machine learning and other procedures, might yield
some insights into reasons for customer dissatisfaction. However, there are
significant limitations associated with such data and associated findings. For
example, the data are completely observational, with no account taken of
whether or not they are representative; thus they are not a reliable guide to the
most important cause of problems. They provide information about one single
user experience – a user, but quite possibly not the person who made the
purchasing decision. And they provide no information about Value, let alone
Relative value, and it is Relative Value that has been clearly established as a lead
indicator of superior business performance. See also the comments in Fisher
(2019) on extracting information from such data.
ii. Table 4 provides an assessment of NPS against the five criteria in Section 2:

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
26
Version: 23 October 2018
a. A statistically sound method that ensures that
no important attributes of the product or
services have been omitted from the survey.
Totally ignored. The current routine collection of
NPS scores for operational activities (cf. the
introductory example in Section 2) is purely
observational, with little or no understanding of
demographic factors let alone sampling biases.
b. A means of linking survey results to higher-
level business drivers.
Assertion of “correlation with growth and
profitability”.
c. Actionable survey results, including the ability
to drill down.
No.
d. A means of identifying where to focus
improvement priorities so as to have the
greatest beneficial impact on the business
bottom line.
No defensible method.
e. Meaningful benchmarking metrics.
Nothing below level of NPS. Also, there are no
agreed standards about how NPS should be
aggregated in a company to produce an all-of-
company metric.
Table 4. Evaluation of NPS against desiderata for satisfaction surveys.
iii. Of course, there are plenty of poorly constructed market research surveys. However
there are good ones as well, and any comparison should be with what is best, not what
is worst.
iv. Correlation doesn’t equate to causation, yet the evidence for the efficacy of NPs
presented in Reichheld (2003) is based only on correlation with business results.
v. A company using NPS is basically not distinguishing those of its customers rating them 0
from those rating them 6, and is totally ignoring ‘Passives’, who are rating them 7 or 8.
This is a recipe for losing market share. Those rating them 0 – 3 or 4 are almost certainly
going to leave. However, with more information about customer needs, those rating
them 5 or 6 could possibly be converted to Passives, and those currently categorized as
Passives could be boosted to Promoter status.
vi. As is well-known from the AT&T work 30 years before, what is critical is the perception
of Relative Value: there are many ways of measuring ‘customer satisfaction’, some OK,
many others not OK, but none of which Reichheld mentions.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
27
Version: 23 October 2018
vii. In the case of AT&T, some of the 9 Business units were operating Business to Business, a
few were purely just Business to Consumer, but most were both. The CVA metric was
predictive of market share for all industries, markets and countries. For Business to
Business, NPS is even weaker because the decision maker is hardly ever surveyed. AT&T
standards required CVA to be calculated from decision-maker surveys of the market that
were random in time and sample, and not triggered by a recent event.
viii. Many executives use NPS as a way to pass on the responsibility for satisfying customer
needs to a coal-face employee so that the executives can be free to focus on satisfying
the business bottom line and the company's investors. However no one employee can
fulfil a customer's total and real needs, especially in the absence of guidance about
where the real problems lie. It is the senior leadership who need to be able to see the
whole picture, rather than just a single, often self-serving metric. Only by carefully
evaluating what can be done to improve products, services and costs can a company
improve the quality of life for its customers.
ix. As is described by Kordupleski (2018), the realization by AT&T in 1986 that they needed
to focus on Value, and Relative Value resulted from extensive statistical exploration of a
very large market research data base, with considerable time being spent seeking
correlations and cross-correlations to high-level business performance indicators such as
Market Share and Return on Invested Capital (ROIC). Thus, for example, CVA emerged
as a lead indicator of Market Share, not simply as a metric correlated to Market Share.
CVA, a customer satisfaction-derived marketing metric could, using a metaphor, “predict

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
28
Version: 23 October 2018
the weather.” What is even more important, a firm could use the data and information
to “improve the weather.”
x. It is important to note that indicators such as Market Share and ROIC relate to the entire
market, not just to one’s own customers, whereas Loyalty, which is fine as far as it goes,
still relates just to one’s own customers. This brings to mind a series of annual Quality
conferences in Australia where marketing for the next meeting was focused primarily on
attendees at the previous meeting. (The principal organizer of the conferences couldn’t
understand how, if every conference had nearly 90% satisfaction rating, attendance had
fallen from around 1200 at an initial meeting to around 450 ten years later. 1200 
(0.9)9  464.)
We note that NPS has not been entirely overlooked by statisticians as an object of study. Jeske,
Callanan and Guo (2011) studied NPS and the claims made for it by Reichheld, and made the
following insightful observations:
… the hope is that movements in NPS are positively correlated with revenue growth for the company.
While Reichheld's research presented some evidence of that, other findings are not as corroborative … .
Regardless of whether there is a predictive relationship between NPS and revenue growth, implementing
policies and programs within a company that improve NPS is an intuitively sensible thing to do … . A
difficult and important question, however, is how to identify key drivers of NPS. Calculating NPS alone
does not do this.
They then proceed to show how to gain insight into what these key drivers might be by applying
statistical modeling to an existing customer survey. This analysis is fine, as far as it goes.
However, it was conducted (seemingly) in the absence of knowledge of the CVM process and

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
29
Version: 23 October 2018
thus of the knowledge gained from a wealth of case studies identifying Value, and Relative
Value, as the crucial lead indicators of high-level business drivers (market share, ROIC, …) as
spelt out by Kordupleski (2003, 2018). Further, amongst the valuable tools associated with CVM
is the concept of a Value Map (Figure 5), which a company can use for strategic positioning.
Inter alia, the Value Map gives meaning to the concept of a Value Proposition. First a company
selects which market (Economy, Average or Premium) it is targeting on the Value Map. Then it
decides whether it will gain market superiority by being superior on Quality and at par on Price,
or by being superior on Price and at par on Quality, or both.
In fact, Kordupleski (2018) is largely a published version of an unpublished article prepared in
1989. To quote from its Abstract:
The following article, “The Right Choice – What Does It Mean” by R.E. Kordupleski and W. C. Vogel, Jr. is a
1989 paper that reported on some of the most significant findings in the early days of customer value
measurement and management. It is based on one of the largest empirical data bases available at that
time. AT&T was doing over 60,000 customer surveys per month. Three years of monthly findings were
analyzed by some of the best researchers and scientists in the 300k employee and $85 billion annual
revenue company. The paper presented empirical evidence of the power of the consumer’s perception of
customer value, its impact on market share, growth and customer loyalty and ultimately its impact on
shareholder value and employee value. The paper was never published, but it was released throughout
AT&T and also to AT&T’s strategic business partners. Its content was presented and discussed at national
conferences hosted by the American Marketing Association and the US Conference Board. Its findings
have stood the test of time.
Clearly, there need to be more bridges built between the statistical and market research
literatures!

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
30
Version: 23 October 2018
Figure 4. Value Map. The scores for Relative Quality and Relative Price are plotted against those of the
competitors. The Fair Value zone relates to which market sector you are targeting. Adapted from Fisher
(2013, Figure 5.6), based on the concept described by Kordupleski (2003); see also Kordupleski (2018).
2.4 Final comments
Customer satisfaction research needs to answer two simple questions
 How are we going?
and
 What should we do?
NPS does only a modest job answering the first but generally contributes little towards
answering the second.
For Customer Value Management,
A. Extensive statistical analysis from a large amount of market research data has
established that Relative Value is a lead indicator for competitive business outcomes
such as Market Share and ROIC.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
31
Version: 23 October 2018
B. Relative Value is derived from data collected from decision-makers (people making the
purchasing decisions), and necessarily involves acquiring competitive data as well as data
from your own company. It is calibrated by competitive information.
C. The decision-maker’s rating of Value is arrived at only after consideration of the entire
customer experience, involving both Quality and Price. The initial failure by AT&T to
obtain a connection between customer satisfaction and business performance was due
in part to overlooking satisfaction with Price.
D. There is a clearly defined and proven process for identifying improvement priorities
likely to have the greatest beneficial impact on the business.
E. Statistical analysis and consequent action are based on representative samples of
market data, with statistical uncertainty of current market position appropriately
quantified.
For NPS,
a. There appears to be little hard evidence that NPS is a lead indicator of business
outcomes; indeed the AT&T experience suggests that there may be no connection.
b. NPS focuses only on the user who, in many cases, will not be the person making the
purchasing decision. It is uncalibrated by competitive information.
c. It often derives from a single customer experience with unknown influence on a
decision-maker’s overall perception of Value.
d. There is no sound approach to selection of improvement priorities.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
32
Version: 23 October 2018
There are well-established ‘best practice’ approaches to creating and delivering superior value
to customers and so gaining and sustaining market share, and there are associated metrics. NPS
is not such a metric.
3. Acknowledgement
We thank Spencer Imel for some incisive comments on an earlier draft, and the constructive
advice of the Editor and reviewers.

Applied Stochastic Models in Business and Industry. Volume 35 (1), January 2019. To appear.
33
Version: 23 October 2018
4. References
1. Fisher NI. Analytics for Leaders. A Performance Measurement System for Business Success.
Cambridge: Cambridge University Press; 2013.
2. Fisher NI. A Comprehensive Approach to Problems of Performance Measurement. Journal
of the Royal Statistical Society Series A. To be read before the Royal Statistical Society, 16
January 2019.
3. Kordupleski R. Mastering Customer Value Management. Cincinnati, OH: Pinnaflex
Educational Resources, Inc.; 2003.
4. Kordupleski R. The Right Choice – What Does It Mean? Ground breaking research from in
the early days of Customer Value Management 2018. Journal of Creating Value. To appear.
5. Jeske Daniel R, Callanan Terrance P, Guo Li. Identification of Key Drivers of Net Promoter
Score Using a Statistical Classification Model. pp 8–25 in Efficient Decision Support Systems
– Practice and Challenges from Current to Future, edited by Chiang S Jao. Croatia: InTech;
2015.
6. Reichheld Frederick F. The One Number You Need to Grow”. Harvard Business Review
December 2003: 1–10.