This document discusses segmentation and clustering techniques for advertising. It defines segmentation as dividing a population into groups with similar characteristics that differ from other groups. The key steps are:
1. Analyzing a consumer database using regression analysis to form initial groups.
2. Clustering the groups to determine which solution best represents distinct segments based on behaviors, attitudes, size, and differences between groups.
3. Analyzing the segments to understand their characteristics and target the most attractive segments for advertising based on criteria like consumption levels and segment size. The goal is to customize messages to different audience segments.
2. Today
• Advertising
• Targetting The building blocks of our business
• Media + Messaging
• Segmentation
• Clustering
• Decision making based on segmentation
2
3. The only purpose of advertising is
to persuade someone of something
Idea
Action
Choice
Opinion
Try
Continue
Return
3
6. So let’s talk about the Universe
22 year-old women
Blonde 40 year-old who go to college
women who own a and live in a dorm
red Bentley
Continental GTC
Women who had a
28 year-old, Single baby less than 6
White females, who months ago
own a chocolate-
covered Labrador
6
7. Blonde 40 year-old
women who own a
red Bentley
Continental GTC
22 year-old
women who go to
college and live in
a dorm
28 year-old, Single
White females, who
own a chocolate-
covered Labrador
Women who had
a baby less than 6
months ago
7
9. Why go through the whole trouble?
• Eliminate waste
• Customize messages to increase response
9
10. What is segmentation?
• Dividing a heterogeneous population into groups where the
members are similar to each other and different from all
others
– Behavioral and attitudinal factors are the key factors in
determining clusters; demographic information is too general
• Not category-specific
– Should really take into account the
entire population or clusters might
not be real
– Can be a sub-cluster
• Ideally tied back to the database
that was used to define the clusters
in the same place
10
11. Building Blocks
• Database (Simmons, MRI, TGI…)
– Syndicated databases
– Robust samples
– Four key areas:
• Demographic information
• Attitudinal Batteries
• Product consumption (Behavioral)
• Media consumption (channel)
• Regression analysis
11
12. Databases
• Simmons, PRIZM, MRI, TGI… are all syndicated research:
companies subscribe and may add their own questions to
what is commonly known as an “omnibus” research
• Typically have thousands of respondents and thousands of
variables
– TGI for Latin America has 55,000 respondents and 4,300
variables: 236,500,000 data points that are analyzed via
regression analysis
• Variables describe a behavior not a response
– Drinking 24+ beers per month is a variable (heavy beer drinker)
– Drinking Presidente beer is not a variable, it is a response
12
13. Database Structures
Variables go in columns…
Respondents go in rows…
Responses: Each Respondent
will be defined by
the answer to all the variables
13
14. Respondents can belong to many
segments & clusters
… be an avid photographer…
… enjoy Golf…
… be a
CEO… … read magazines…
A woman can…
… and be a mom
This is why it’s important to
segment an entire population in
order to get the full picture
14
15. Regression Analysis
• Regression analysis is a pure statistical term which includes
the techniques for modeling and analyzing several variables,
when the focus is on the relationship between a dependent
variable and one or more independent variables.
• In layman’s term, it lets you see how value of one
variable(also called dependent variable) changes when any
one of the independent variables varies.
• You ask the software for a number of solutions containing a
number of predefined groups (26, 28, 30…)
• The software forms the groups. Typically, you can’t remove
one of the groups without affecting the total answer, which is
why you need to have several solutions
15
16. Regression Analysis
• At its simplest, regression analysis does the following
• Say it is forming a solution with 28 groups.
• It takes respondent #1 and assigns it Group #1
• Then it takes respondent #2 and checks whether it is so
similar to Respondent #1 that they should be grouped
together
– If they are… Respondent #2 goes into group #1
– If they are not… Respondent #2 forms Group #2
• And so on and so forth until it has grouped the thousands of
respondents from the database
16
18. Step 1 – Database to segments
Begin with a robust database. All
respondents in rows, one row per
respondent, all variables in
columns.
This database is exported,
typically in a CSV format, which
the software doing regression
analysis can process
18
19. Step 1 – Database to segments
Run the regression analysis.
For really small groups you can
use Excel –and there are several
great tutorials in YouTube
For a large database, where you
might process millions of
datapoints and have thousands of
columns and rows, you have to
use professional software and hire
specialists
19
20. Step 1 – Database to segments
Solutions with pre-formed groups.
The output from the regression
analysis software contains a
number of groups (which you pre-
determined). This could be 24, 26,
28… etc. groups. There is a limit
to the number of groups we can
really understand
At this stage, you only have a
mathematical construct which
might or might not make “sense”
20
22. Clustering (1)
• Once you have the set of solutions, you have to analyze all of
them to determine which one will fit your “model of the world”
• Parameters:
– Number of groups (not too many, not too few)
– How different are the groups from each other
– Do they “make sense” (the smell test)
– Are the groups significant either in numbers, purchasing power
or other metrics?
• By now, the segments have also been inserted back into the
original database and you are ready to go
22
23. Clustering (2)
• Once you have a final solution that everyone is happy with:
• Name the groups so that understanding them is instinctive
– Knee Deep in Toys
– Salt of the Earth
– Blueblood Estates
• Personal peeve: “Cutesy” names, though. I am worried that
they become shortcuts
• Recommendation: Pictorial Profile
– As a group choose a group of 3-5 pictures that you think
represents each segment to establish a visual identity
– Focuses efforts
23
24. Clustering (3)
• Analyzing the segments/clusters – What they look like
Group Variable Index
25 Have had a child in the past 12 months 450
25 Have had a child in the past 24 months 395
25 Have purchased toys in the past 1 month 360
25 Bought a digital camera 275
25 Watch daytime television 220
25 Rent/download movies 5+/week 215
25 "I would rather spend a quiet evening at home" 209
25 "We often entertain at home" 205
25 F 25-34 120
25 F 35-44 115
24
25. Is everyone familiar with an “index”?
Item Value Index An index relates a value to
Item 1 130 66.7 the average for the series
Item 2 120 61.6
Item 3 200 102.6
Item 4 400 205.2
Item 5 315 161.6
Item 6 221 113.4
Item 7 76 39.0
Item 8 95 48.7
Item 9 55 28.2
Item 10 32 16.4
Item 11 500 256.5
Average 195 100.0
25
26. Clustering (3)
• Analyzing the segments/clusters
1. Size – Ideally, there should be some sort of size continuum so
that we have large groups and small groups. But:
• No group should be huge compared to the others
• A situation where there are two or three large segments and many
small ones is also flawed (the large groups carry the weight)
2. Homogeneous – ideally, the members of each segment should
all index high against the main components of the group (e.g.,
have travelled to a foreign country 4+ times in the past 12
months) and this will not be shared by all the other groups.
1. This is why gender and age are NOT good discriminators
26
27. Clustering (3)
• Analyzing the segments/clusters
1. Heterogeneous – While a few groups can share some traits
(e.g., purchased a new car in the past 12 months) if a trait or
variable is shared by a large number of groups, then it fails to
really segment
2. Attitudinal Similarity – Since the end goal is to be able to
customize messages for each group, groups that share some
variables (e.g., purchased new clothing in the past month) but
differ in attitudes (e.g., some are very conservative, some are
very liberal) probably will not work
3. Extremes are good: high indices in the top 15% or so of the
variables is great
27
28. Clustering (3)
• Analyzing the segments/clusters
– Look for media habits… are there any media vehicles that index
high in this group but not on others? (e.g., visit radio/music sites
on the web more often)
– Bears repeating: does the group make sense? Sometimes you
will read through the variables and the group that emerges just
doesn’t make sense. For example, have taken 6+ foreign
business trips in the past 12 months AND index high to knitting
but are predominantly males. Discard it. Leave it in the attic.
28
29. Groups, segments & clusters
• Many times these terms are used interchangeably.
• Groups, in general, might have only one thing in common
(e.g., play soccer every Saturday morning)
• Segments –what we have been talking about– have more
things in common and tend to be different from other
segments. They are also unique.
• Clusters are groups of segments.
– For example, you may have four different segments that drink
wine often (12+ times/month), but one of the segments might be
heavily into cooking, another segment heavily into partying,
another might be rich and have wine with their dinner every
night, etc. You can conceivably create a cluster of “wine
lovers”
29
32. Attitudinal
Battery
Demographic
Information
Media
Consumption
Consumption
Behavior
32
33. Let’s look at this situation
Consumption CPC Size of the
Group per Capita Index Group As a % Consumption As a %
Group 1 50.5 218.8 120,000 7.8% 6,060,000 17.0%
Group 2 45.3 196.3 75,000 4.9% 3,397,500 9.6%
Group 3 30.2 130.8 180,000 11.7% 5,436,000 15.3%
Group 4 24.5 106.1 200,000 13.0% 4,900,000 13.8%
Group 5 20.0 86.6 150,000 9.7% 3,000,000 8.4%
Group 6 18.2 78.8 70,000 4.5% 1,274,000 3.6%
Group 7 17.5 75.8 300,000 19.5% 5,250,000 14.8%
Group 8 15.9 68.9 250,000 16.2% 3,975,000 11.2%
Group 9 12.0 52.0 110,000 7.1% 1,320,000 3.7%
Group 10 11.0 47.7 85,000 5.5% 935,000 2.6%
23.1 1,540,000 35,547,500
33
34. Step 1 – Determining target segments
Consumption CPC The first step is to run a quick analysis
Group per Capita Index showing per-capita consumption of the
Group 1 50.5 218.8 product in each one of the groups.
Group 2 45.3 196.3
The top groups will be more attractive;
Group 3 30.2 130.8
the bottom groups we discard unless
Group 4 24.5 106.1 they are big enough to really merit some
Group 5 20.0 86.6 resource allocation
Group 6 18.2 78.8
Group 7 17.5 75.8 This gives us a rough guideline of what
Group 8 15.9 68.9 groups we should be interested in
Group 9 12.0 52.0
Group 10 11.0 47.7
23.1
34
35. Step 2 – We then look at the entire
picture of the market
Consumption CPC Size of the
Group per Capita Index Group As a % Consumption As a %
Group 1 50.5 218.8 120,000 7.8% 6,060,000 17.0%
Group 2 45.3 196.3 75,000 4.9% 3,397,500 9.6%
56%
Group 3 30.2 130.8 180,000 11.7% 5,436,000 15.3%
Group 4 24.5 106.1 200,000 13.0% 4,900,000 13.8%
Group 5 20.0 86.6 150,000 9.7% 3,000,000 8.4%
Group 6 18.2 78.8 70,000 4.5% 1,274,000 3.6%
Group 7 17.5 75.8 300,000 19.5% 5,250,000 14.8%
Group 8 15.9 68.9 250,000 16.2% 3,975,000 11.2%
Group 9 12.0 52.0 110,000 7.1% 1,320,000 3.7%
Group 10 11.0 47.7 85,000 5.5% 935,000 2.6%
23.1 1,540,000 35,547,500
Let’s say we have a large brand, so we are
interested in a large base. The top 4 groups
accumulate over 50% of the consumption. So, right
off the bat, we would look at those in detail
35
36. Step 3 – In-depth analysis
Group Group 1 Group 2 Group 3 Group 4 Total
Women 18-34 25% 40% 30% 35% 32%
Women 35+ 30% 25% 30% 15% 24%
Men 18-34 26% 25% 25% 35% 29%
Men 35+ 19% 10% 15% 15% 15%
100% 100% 100% 100% 100%
Objective: Understand the physical makeup of the target groups.
From here we would reach a couple of conclusions:
3.Young product (both Men and Women)
4.Older men just do not like it
36
37. Group Group 1 Group 2 Group 3 Group 4 So we see that groups
Women 18-34 25% 40% 30% 35% with a high
Women 35+ 30% 25% 30% 15% concentration of young
people tend to index
Men 18-34 26% 25% 25% 35%
better in certain
Men 35+ 19% 10% 15% 15% purchases
Group (Index) Group 1 Group 2 Group 3 Group 4
One conclusion we
Divorced 85 65 85 160
could make is that
Married 115 150 140 95 product “X” is, in some
Purchased Flat TV 125 125 110 155 way, a life-transition
Purchased Furniture 105 150 110 170 product:
Purchased camera 110 125 105 160
Watch TV < 1 hr/day 125 115 95 150 5.Young people
moving into their first
Watch TV + 4 hr/day 95 80 115 90
apartment
6.Recently divorced
women now living
We would look at dozens and dozens of statements alone
in order to gain insight into the different groups
37
38. Group Group 1 Group 2 Group 3 Group 4
A/MA - A job should be more than work, it
should be a career 110 115 115 140
A/MA - There are still many opportunities for
advancements if one works hard 115 120 120 110
A/MA - I am religious 80 120 140 80
A/MA - It is important to take care of the
environment 90 130 150 75
A/MA - Speaking a second language is a
great advantage 140 125 90 125
A/MA - I like to celebrate traditional holidays
at home surrounded by my family 80 140 130 100
A/MA - My friends seek my advice before
buying electronic products 140 95 80 140
Liberal Traditional/Conservative Liberal
In looking at the attitudinal
battery, we then see the What emerges is two big groups, one more religious
psychological makeup of and conservative and the other more liberal. Ideally,
the groups we would want to craft focused messages that match
their attitudes and beliefs
38
39. Group (Index) Group 1 Group 2 Group 3 Group 4
Watch TV < 1 hr/day 125 115 95 150
Watch TV + 4 hr/day 95 80 115 90
Listen to talk radio 3+ times/week 60 50 115 30
Listen to web-radio 3+ times/week 140 90 40 120
Read daily newsp 3+ times/week 80 110 130 70
Web - 30+ hours/month 140 105 90 130
Web - 60+ hours/month 220 90 80 105
Finally, we look at media
For this group, for example, For this group – which was
habits to determine which we might consider more more conservative– we
liberaly messaging and a might consider a channel
channels we will be using media strategy that leans strategy that used talk radio
heavily on the web and newspapers.
to reach each group (websites, web-radio)
We would then choose the
Given the size of the group, messaging form, which
On a first pass, we would it might be that, a priori, we could include live reads,
won’t be recommending advertorials and regular
look at each group television advertising
individually
39
40. Wrapping it up
Group Group 1 Group 2 Group 3 Group 4
Cons/Capita 50.5 45.3 30.2 24.5
CPC Index 218.8 196.3 130.8 106.1
Size of the Group 120,000 75,000 180,000 200,000
As a % of the Universe 7.8% 4.9% 11.7% 13.0%
As a % of the Sub-groups 20.9% 13.0% 31.3% 34.8%
Consumption 6,060,000 3,397,500 5,436,000 4,900,000
As a % of the Universe 17.0% 9.6% 15.3% 13.8%
As a % of the Sub-groups 30.6% 17.2% 27.5% 24.8%
Watch TV < 1 hr/day 125 115 95 150
Watch TV + 4 hr/day 95 80 115 90
Listen to talk radio 3+ times/week 60 50 115 30
Listen to web-radio 3+ times/week 140 90 40 120
Read daily newsp 3+ times/week 80 110 130 70
Web - 30+ hours/month 140 105 90 130
Web - 60+ hours/month 220 90 80 105
40
41. As a reminder…
THERE’S NO POINT IN SEGMENTATION OR
CLUSTERING IF YOU ARE NOT GOING TO
CREATE SPECIFIC MESSAGES
41
42. Some conclusions
• We might consider television (e.g., cable) for broad, more
generic messaging.
• We identified 2 liberal groups with a high web indices:
– Web radio
– Regular websites
• One of the groups also had a high divorced index
• Conclusion:
– We might consider a mix of ad networks (for quick reach and
cheap CPM) and depending on our product, some premium sites
– We might also want to consider “dating” sites specialized on
divorced women offer a promotion of some sort to test
response
42
43. Some conclusions
• We also identified 2 conservative groups that account for half
of our sub-groups consumption (44%) with high indices for
– AM Talk radio
– Newspaper readership
• They will also be exposed to our TV campaign
• Conclusion:
– We might consider some live reads on AM radio
– Program sponsorship and, depending on the product itself, long-
form programming (e.g., 30 minute shows with live talent)
– We might also consider a newspaper campaign including
advertorials and coupons
43
45. Loyalty & Consumption
Consumers “bonded” with a brand spend more on that brand.
The lowest level for any brand is awareness. Awareness doesn’t translate into
buying. The main purpose of going through a segmentation exercise is to
create a true bond with the brand (often described as “this brand fits me”)
45
46. An increase among top consumers can have
an oversized result for the entire company
Consumption Size of Increase New
Group per Capita the Group Consumption Goal Consumption
Group 1 50.5 120,000 6,060,000 15.0% 6,969,000
Group 2 45.3 75,000 3,397,500 15.0% 3,907,125
Group 3 30.2 180,000 5,436,000 12.0% 6,088,320
Group 4 24.5 200,000 4,900,000 12.0% 5,488,000
Group 5 20.0 150,000 3,000,000 6.0% 3,180,000
Group 6 18.2 70,000 1,274,000 2.0% 1,299,480
Group 7 17.5 300,000 5,250,000 2.0% 5,355,000
Group 8 15.9 250,000 3,975,000 2.0% 4,054,500
Group 9 12.0 110,000 1,320,000 0.0% 1,320,000
Group 10 11.0 85,000 935,000 0.0% 935,000
23.1 1,540,000 35,547,500 8.6% 38,596,425
46
48. You need to do the math
From 1 Coke per week
(50 per capita) to 1 extra
Coke every 6 weeks
From 6 per year
(very loyal
consumer at 50%
of the category
share) to 7/year
What does a 15% From 120 transactions/year to
138. At 3.5% or about $84 per
increase in customer/year to $97 if the
consumption really average transaction is only $20
mean?
48
49. Conclusion
• Segmentation is a powerful tool
• Four cornerstones:
– Demographic
– Attitudinal
– Product consumption
– Media habits
• Purpose: Increase response
• Messaging must be customized
• You have to do the math
• Avoid introducing personal biases into the process
• Channel & Messaging must work in unison
49
51. The only purpose of pitching is selling.
Not teaching.
Not proving a point.
Selling.
51
52. The 7 elements of a successful pitch
• Definition
• Why this makes sense
• WIIFM
• Business Plan
• Why is the BP credible
• Ask for the order
• Next Steps
52
55. 2. Why this makes sense
• Tighter job market forces many young men to adopt a more
“business-like” demeanor
• Younger segment accepts logos more readily
• Nike well-known for manufacturing premium footwear
• Profitable niche
– Current margin for dress shoes averages 30%; Nike brand can
achieve 45% on average sales of $150 per pair
– Current margin for athletic shoes average 20% due to
discounting on average sales of $75/pair
• Minimum marketing costs:
– In-store posters
– Social, Internet
– Handful of business and/or high fashion magazines
55
56. 3. A profitable business for Nike
Average
Division Unit Sales Price Total Sales Margin Gross Profit
Athletic Shoes 120,000,000 $ 75.00 $ 9,000,000,000 20% $ 1,800,000,000
Clothing 100,000,000 $ 45.00 $ 4,500,000,000 17% $ 765,000,000
Events & Others 30,000,000 $ 50.00 $ 1,500,000,000 30% $ 450,000,000
Formal Shoes 15,000,000 $ 150.00 $ 2,250,000,000 45% $ 1,012,500,000
265,000,000 17,250,000,000 4,027,500,000
Athletic Shoes 52% 45%
Clothing 26% 19%
Events & Others 9% 11%
Formal Shoes 13% 25%
High inherent margins will increase gross profits of the
company by $1bn
56
57. 3. A profitable business for Nike
Unit Cost % of
Division Total Sales for Mktg Marketing Costs Sales
Athletic Shoes $ 9,000,000,000 $ 4.50 $ 540,000,000 6.0%
Clothing $ 4,500,000,000 $ 3.50 $ 350,000,000 7.8%
Events & Others $ 1,500,000,000 $ 4.00 $ 120,000,000 8.0%
Formal Shoes $ 2,250,000,000 $ 4.50 $ 67,500,000 3.0%
$ 17,250,000,000 $ 1,077,500,000 6.2%
Lower advertising costs (due to lower-cost media) also
increases contribution of formal shoe line to the bottom line
57
58. 4. Business Plan
• There are plenty of business-plan templates on the web
• Just do one
Note
• The business plan should contain the advertising plan which,
in turn, contains the targetting considerations including
segmentation
58
59. 5. Key success factors
• Nike brand
– Premium quality
– Logo is acceptable
– Distribution is assured
• Distribution
– Already built-in Nike already distributes to 85% of shoe stores
– No cannibalization: formal shoes and athletic shoes have
different display footprints
• Management Team
– Successfully transitioned Cole-Haan from 2nd tier to premium
– Excellent distribution expertise
– Marketing guru transition from Nike to Nike Formal Shoes
59
60. 6. Decision & Timing
• Timing & Pipeline have an urgency:
– 6 months from concept to store
– Must be in stores by September – a key psychological period for
“back to work”
• We should proceed within the next 2-4 weeks
60
61. 7. Next Steps
• Week 1, 2 – Review business plan in depth with CFO and
Chief Procurement officer
• Weeks 2, 3 – Timeline and Pipeline with foreign sources for
design, manufacturing and shipping
• Week 4 – Present to CEO for final approval
• Week 5 – Begin design & outsourcing manuf.
61