Segmentation (final) 6/17/99 2:25 PM Page 6
ALAN COBER
Segmentation (final) 6/17/99 2:25 PM Page 7
M A R K E T I N G 7
A segmentation
you can
act on
John Forsyth, Sunil Gupta, Sudeep Haldar,
Anil Kaul, and Keith Kettle
Value-based segments usually don’t fit neatly into demographic ones |
Some solutions step around the problem; others meet it head on
R emember when marketing was simple? Your division operated
in a manageable geographic region. You defined your consumer
targets by age, say, and by income. If you were in a business-to-business
market, you divided up companies by size.
But the wild proliferation of brands and channels in rapidly globalizing
markets now flusters even the most sophisticated marketers. In this envi-
ronment, how should your sales force tailor its strategies to its accounts?
Different customers have different attitudes, needs, and preferences, but the
old distinctions no longer take you very far. What should you be looking at
today? The current purchasing behavior of your customers? The benefits
they seek to obtain? Demographics or its business-to-business equivalent:
“firmographics”?
Ford’s Model T strategy—any color you wanted, so long as it was black—
worked until customers had an alternative. Soon-to-be-deregulated utilities,
among other companies, are now miserably aware of this reality. How will
the utilities build loyalty among the most profitable customers before com-
petition takes them away? Utilities had so little need for marketing in the
past that some know very little about them and have no idea what products
and services might keep them loyal after the coming of choice.
John Forsyth is a principal in McKinsey’s Stamford office; Sunil Gupta is a professor at the
Columbia Business School; Sudeep Haldar is a consultant in the Chicago office; Anil Kaul
is an alumnus of the Chicago office; and Keith Kettle is senior vice president of The M/A/R/C
Group. Copyright © 1999 McKinsey & Company. All rights reserved.
Segmentation (final) 6/17/99 2:25 PM Page 8
8 T H E M c K I N S E Y Q U A R T E R LY 1 9 9 9 N U M B E R 3
Companies often address this problem by developing segmentation schemes
breaking down markets into sets of customers or potential customers who
share attributes that might be based on demography (income, say, or age)
or on values or needs.1 Consider some obvious examples. Video camera
manufacturers capitalize on the fact that families expecting their first chil-
dren are likely customers. Telephone companies try to sell call waiting to
families with teenage children. The USAA insurance agency targets military
personnel because it has come to believe, correctly, that this group is likely
to be significantly more loyal, and therefore more profitable, than others.
Unfortunately, easy cases permitting marketers to establish meaningful
differences among groups of customers and then to identify them— a
phenomenon we call “actionable segmentation”—are rare. More often,
despite decades of research and many refinements in the basic model, the
segmentation process creates very real difficulties for marketers. No doubt
such methodologies as conjoint or latent-class analyses permit them to use
values, needs, and attitudes to devise groups (for instance, price-, service-,
and quality-oriented segments) that
include almost all customers. Yet it
How do you find customers usually turns out to be very difficult
who care mainly about service to identify the flesh-and-blood people
without interrogating actually inhabiting the segments.
them all? How do you find customers who
care mainly about service or quality
without interrogating them all?
A leading insurance company based in the United States spent a lot of time,
trouble, and money dividing its world into segments, only to run into exactly
this problem. In the end, the company abandoned segmentation entirely.
The basic difficulty is that value-based segments generally don’t fit neatly
into demographic ones. Many companies therefore start with the simpler
task of identifying differences based on demography or on the different
attributes of different companies. Companies in consumer markets, for
example, typically divide their customers into baby boomers, generation
Xers, and so forth. Likewise, many companies that sell to other businesses
1
It has become fashionable to suggest that customer-relationship marketing, the interactive
character of the Internet, and the ability of today’s computers to collect, process, and
retrieve detailed information about huge numbers of customers have moved business toward
segment-of-one marketing and away from marketing by traditional segmentation. But we
think it unlikely that this kind of “mass customization”—for example, custom-fitted Levi’s
jeans—will replace segmentation as the basis of strategic decision making. First, designing
products and services for individual customers probably won’t be cost effective anytime
soon. Second, even if companies have sufficiently large databases to customize services
for their current customers, they rarely have enough information to do so for their com-
petitors’ customers or for nonusers. The information revolution thus will probably strengthen
the tendency of companies to use information about thousands or millions of customers
by organizing them into segments.
Segmentation (final) 6/17/99 2:25 PM Page 9
A S E G M E N TAT I O N Y O U C A N A C T O N 9
segment customers on the basis of such characteristics as their size, the
volumes of their accounts, and the industries in which they compete.
Unfortunately, though advertising agencies and sales forces find this
approach easy to understand and to implement, it really is no more
effective than value-based segmentation schemes: by no means do all
baby boomers have the same preferences and purchasing behavior, and
businesses of the same size, account volume, and industry may very well
have rather different values and needs.
Segmentation based on demographics or company characteristics thus is
not very actionable; these approaches don’t help you get to your customer
with the right offer. In what follows, we describe four ways of solving the
segmentation dilemma. Targeting and self-selection, the simplest of them,
step around the problem; scoring models and dual-objective segmentation
deal with it head on.
Targeting
Sometimes a segmentation strategy works even if you can’t identify who
is in which segment.
In the early 1990s, price wars at the pump threatened the profitability
of oil companies. To turn the situation around, Mobil Oil queried 2,000
customers in a segmentation study revealing that only 20 percent of gaso-
line buyers were price shoppers,
who spent an average of $700
annually, while customers in other Segmentation schemes
segments spent as much as $1,200. based on demographics or
Although Mobil could not distin- company characteristics are
guish price-sensitive shoppers from not extremely actionable
price-insensitive ones, the news
that 80 percent of its customers
were price-insensitive, heavier users shifted the company’s focus away
from pricing. As a result, Mobil reaped an extra $118 million a year
in earnings from an additional two cents a gallon on its gas—a major
accomplishment.2
In general, selecting targets is the first task of any segmentation strategy.
Before worrying about how to identify and reach individual customers
in any given segment, it is worth attempting to determine whether the
collective traits of a market segment might themselves suggest profitable
strategies.
2
Wall Street Journal, January 30, 1995.
Segmentation (final) 6/17/99 2:25 PM Page 10
10 T H E M c K I N S E Y Q U A R T E R LY 1 9 9 9 N U M B E R 3
Self-selection
The basic idea of self-selection is to reverse the roles of a company and
its customers: instead of trying to find, say, price-sensitive people, the com-
pany figures out what segments it wants to reach and gives the consumers in
them ways of finding it.
Companies most commonly try to get customers to select themselves by
multiplying stock-keeping units (SKUs); different sizes of cereal packets
or washing powders are the most obvious example. The segments walk
up to the supermarket shelf and buy the most appropriate offering.
Coupons, which allow consumers to select themselves on the basis
of price sensitivity, are another classic mechanism. Although everyone
receiving coupons has the option of getting a discount, only customers
who are both price-sensitive and relatively indifferent to the trouble
of clipping and saving coupons tend to redeem them. As a result, super-
markets earn smaller margins—mostly on the fraction of their transactions
precipitated by the coupons—and get the full price from consumers happy
to pay it.
Similarly, airlines often have lower airfares for people willing to include
Saturday night stays in their trips. These companies realize that consumers
in the price-sensitive segment will sacrifice flexibility for low cost but may
not know who these customers are.
EXHIBIT 1
Sniffing a discount, however, these
Segmentation through product design people come looking for the airline.
The hopefuls The fearfuls
Packaging the same product in
Brand name Conceive RapidVue
Price $9.99 different ways can also do the trick.
$6.99
Packaging Quidel, a company based in San
Pink box, smiling baby Mauve background, no baby
Shelf position Near ovulation-testing kits Near condoms
Diego, California, specializes in
developing rapid diagnostic tests.
One of its products can detect
pregnancies in their earliest stages. Until recently, Quidel undertook almost
no consumer marketing, focusing instead on doctors. In 1993–94, its preg-
nancy and ovulation products had almost an 80 percent share of the med-
ical market but just 18 percent of its consumer counterpart.
Quidel conducted a market segmentation study whose findings prompted
it to target two kinds of women separately: those who want to get preg-
nant (the “hopefuls”) and those afraid that they might be pregnant (the
“fearfuls”). Not surprisingly, demographics and related characteristics
didn’t help the company distinguish between the two segments, so it
created different packages for them. Exhibit 1 shows the correspondence
Segmentation (final) 6/17/99 2:25 PM Page 11
A S E G M E N TAT I O N Y O U C A N A C T O N 11
between the concerns of both segments and the brand names, prices, box
designs, and shelf placements of the products created for them.3
Companies can thus promote self-selection through a variety of mechanisms,
the most popular being coupons, pricing structures based on times of the day
or days of the week (telephone and airline pricing, for example), and different
versions of products. These approaches are most suitable when the base of
customers is large but the dollar volume for each of them is too small to make
other approaches to segmentation or mass customization economical.
Scoring models
A telecommunications company that wanted to segment its market undertook
a survey of telecommunications managers to identify the groups into which
its market naturally divided. It identified several groups defined by market
need, such as the price-, convenience-, and quality-oriented segments. As
usual, the needs of these managers could not be predicted from the nature
of the companies employing them. When the telecommunications company’s
salespeople used their own judgment to place customers in segments, the
results were better but still too weak to make a real difference.
To make it possible to act on the segmentation scheme, the company
developed a scoring model—a tool, based on a statistical approach called
discriminant function analysis (DFA), that lets marketers use the answers
of customers to a few key questions to place them in appropriate segments.
Credit card companies employ scoring models to classify customers as
good or bad risks.
The telecommunications company analyzed the answers of half of the
respondents to the original market research survey to establish as precisely
as possible the mathematical relationships between their answers and their
segments. The answers of the other half of the respondents were then used
to test the model’s predictive accuracy; in other words, these responses were
input to the scoring model and its segment predictions tested accordingly.
When the intuitive segmentation of EXHIBIT 2
the salespeople was compared with The effect of scoring models
the outcome of the model, the most
Segmentation system Hit rate
significant finding was that they were
Firmographics 33
much too ready to put accounts into Unaided salesperson’s judgment 47
the price-oriented segment and were Scoring model 72
therefore giving value away. Exhibit 2
3
Forbes, August 29, 1994.
Segmentation (final) 6/17/99 2:25 PM Page 12
12 T H E M c K I N S E Y Q U A R T E R LY 1 9 9 9 N U M B E R 3
compares the predictive accuracy of simple firmographic information, the
unaided judgment of salespeople, and the scoring model.
A similar approach helped a US company sell office machines to business
customers in Japan. Because the company had to compete with a deeply
entrenched competitor, the company’s market share was much lower
there than in the United States. Market research suggested that the
Japanese market was divided among three segments whose preferences
were the competitor’s product, convenience, and price, respectively.
Because the apparent differences among the business customers who
made up this company’s potential market were not significant, it was
almost impossible to predict the segment to which any particular cus-
tomer belonged.
EXHIBIT 3 The company first created a scoring
Scoring models: An example model to identify customers likely
to favor the competitor’s products.
A sales representative of a company wishing to sell “equipment A” to
company X, with 200 employees, six departments, and no equipment A, Another scoring model placed
calls it to ask questions that will help create a scoring model. the remaining customers in the
Possible Score for Score for convenience or the price segment.
Question answers answer company X
How many <9 1 1 These scoring models were based
departments does 10–49 2 on publicly available information
your company have? 50 or more 3
Do you own Yes 1 (such as the revenue of a company
equipment A? No 0 0 and the number of its full-time
How many employees <10 1 employees) and answers to
do you have? 10–50 2
>50 3 3 key questions salespeople put
to prospects over the telephone
Calculations
Score = 1.59 + 0.046(Q1) – 0.166(Q2) + 0.08(Q3) = 0.24 before meeting them (Exhibit 3).
Conclusion Having assigned those companies
Since the score is less than 1.5, company X is in the convenience segment
that did not prefer buying office
equipment from the competitor
to the convenience or the price segment, the sales force knew which to
target with a full-featured version of the product and which with a no-frills
version. Although the company did not expand its sales force, it increased
sales by 40 percent.
Dual-objective segmentation
Recent advances in market research and modeling techniques have begun
to make it possible to convert unactionable segmentation schemes into
actionable ones.4 In essence, these models of dual-objective segmentation
4
See Abba M. Krieger and Paul E. Green, “Modifying cluster-based segments to enhance agreement with
an exogenous response variable,” Journal of Marketing Research, 33 (August), 1996, pp. 351– 63.
Segmentation (final) 6/17/99 2:25 PM Page 13
A S E G M E N TAT I O N Y O U C A N A C T O N 13
EXHIBIT 4
(DOS) trade a small amount of
precision in delineating segments Dual-objective segmentation
for a significantly increased
High
ability to identify the customers Value-based segmentation
Actionable segmentation
who belong in each of them by
optimizing a function that is a
Identification of value
weighted sum of value-based
Joint optimization
segmentation and demographic
segmentation (Exhibit 4).
Demography-based segmentation
Consider an example. Recognizing
that many companies in the highly
Low
fragmented European market were Low High
Identification of customers
eager to outsource their information
technology activities—both to
reduce costs and to focus on their core businesses—a well-known technology
company became interested in offering them its network systems manage-
ment services. Before starting out, the company invited senior technology
managers to identify the factors that would dispose them to outsource. Their
answers permitted them to be grouped in six distinct segments, a number that
struck the optimal balance between precision and breadth.
Analysis quickly established that the alignment between the size of an
account and the receptivity of the customer was weak. This was not unex-
pected. More problematic was the
fact that four of the six segments
had an outsourcing probability of Dual-objective segmentation
40 to 50 percent — too many seg- trades a little precision in
ments with too low a probability delineating segments for
of outsourcing. Having distilled six a much greater ability
segments and discovered promise to identify the customers
in four of them (performance in them
seekers, people who valued the
ability to manage an entire network,
those focused on operations, and technology seekers) the company was
at a loss to know which one or two it should pursue.
Now, the segmentation scheme originally required the technology com-
pany to assign all customers to the segments that best described them.
As with just about any classification system, however, each segment con-
tained customers that clearly belonged in it and customers that were more
peripheral. In the DOS process, a computer works through an algorithm
that makes tiny changes in the segmentation system by experimentally
reclassifying customers on the periphery of one segment into another
Segmentation (final) 6/17/99 2:25 PM Page 14
14 T H E M c K I N S E Y Q U A R T E R LY 1 9 9 9 N U M B E R 3
EXHIBIT 5
segment. Then the
Probability of outsourcing — before and after computer runs the
Percent
analysis that correlates
33 membership in a seg-
Brand loyalists 15 ment with the likeli-
49
Operations-focused 53
hood of outsourcing.
49
Technology seekers
Of course, pushing 67
31 customers from one
Price-sensitive 4
46
segment into another
“Manage my network” 56 that seems less obvi-
41 ously appropriate
Performance seekers 73
for them has a price:
Before dual-objective
segmentation
After dual-objective
segmentation
blurring the clarity
of the segments. The
upside, however, is
that the process typically generates a much higher correlation between
membership in a segment and the likelihood of outsourcing. Eventually,
34 percent of the respondents were reclassified. The post-DOS numbers
clearly suggested that the company should target the performance seekers
and the technology seekers —an actionable outcome (Exhibit 5). As a
result, the salespeople, who now realized that only members of these two
segments should be pursued aggressively, were trained to ask all new
clients a few questions that would make it possible to place them in the
appropriate segment.
DOS offers the greatest gains when firmographics alone has very little
ability to predict behavior. We applied the DOS approach to three other
EXHIBIT 6
case studies (Exhibit 6).
The first estimated the
Trading off segment clarity for correlation with buying behavior
size of the market for
50 dietary supplements by
45 Network services analyzing consumers’
40 views about which
Fit of segmentation, percent
35 Dietary supplements
factors contribute
30 to a healthy lifestyle.
Group long-term disabilities
25 The second aimed
20 to size the potential
15 Chlorinated solvents market for chlorinated
10 solvents by examining
5 chemists’ prospective
0
0 5 10 15 20 25 30 35 40 45 50
need for the product.
Fit of identification, percent The third was an effort
to decide which benefits
Segmentation (final) 6/17/99 2:25 PM Page 15
A S E G M E N TAT I O N Y O U C A N A C T O N 15
managers were most likely to extend the insurance coverage of their com-
panies to include long-term disabilities.
As in the example of the network systems company, the pre-DOS correla-
tion between the traits and attitudes of potential customers for solvents
was very low; after DOS, it more than doubled. For dietary supplements
and long-term disabilities, the pre-DOS correlation was high, so the
improvements achieved through DOS were less compelling—an increase
of 34 percent and of only 3 percent, respectively. Most important, in
every case the improvement entailed a sacrifice of only 5 percent in the
fit between company and segment.
Getting the science right is a big part of the battle, but only a part. Many
segmentation schemes are carried out and then ignored because key deci-
sion makers were not involved in the process of segmentation and did not
understand how it was carried out. It is critical to discuss the impact that
the segmentation exercise of any company is meant to have on the way its
key people do business. One way to draw these people into the research
and to give them a stake in it is to get them to speculate about what the
segments of a given market might be and what can be done to attract each.
Remember also to talk to decision makers in your target segment after the
segmentation exercise and before you begin to market to them. They will
reveal—in their own language, not yours—what they want and how to
reach them. Their views also act as a reality check on the segmentation
scheme, which of course will never be totally effective. Given the inevitable
limitations of such a scheme, be sure to undertake a pilot project before
rolling out a national or international marketing campaign. This gradual
approach also helps bring the organization on board before the stakes
become perilously high.
Finally, be prepared to think beyond the current model. Typically, segmenta-
tion exercises reveal several new ways to serve customers—new channels,
new ways of doing business, new ways to train and support the sales force.
You may find that your goals expand along with your customer base.
