A Pragmatic Approach to Analyzing Customers

A Pragmatic Approach to Analyzing Customers
Mark Madsen
www.ThirdNature.net
@markmadsen

What happened in the overall market?

© Third Nature Inc.
The Internet happened and companies are still reacting
The internet is unlike all prior media channels: it allows for two-
way communication, enabling entirely new practices.

From the brand as controller of messages in a channel…

…to the brand as participant in interactions in an arena.

© Third Nature Inc.© Third Nature Inc.
Bad news if you want to reach people via advertising
Number of prime-time 60
second TV commercials required
to reach 80% of 18-49 year-olds
▪ In 1965: 3
▪ In 2002: 117
People with DVRs watch 12%
more TV 
90% of them skip the ads 

The obvious solution hasn’t worked very well.

One-to-One marketing concepts fared little better.

Changes to Communication Channels
Old model:
▪ One-way
▪ Outbound
▪ Interrupt-driven marketing
▪ Messaging
New model:
▪ Two-way
▪ Inbound, outbound and
between peers
▪ Event-driven marketing
▪ Messaging and listening

Social Software Affects Buying Behavior for Business

Everyone is a Direct Marketer Now

The (mass) Marketing Process
This is designed for a mass market where you create a few
to dozens of stable segments, with hundreds of thousands
or millions of members in each.
Analyze and segment the
market and/or customers*
Create offer
Execute
campaign
Record
results
Feedback
*or not

Hard because measuring results is hard
?
The assumption is this:
What happens in the
middle is largely a mystery
because of the nature of
mass communication and
transaction channels.
Mass marketing treats people similarly, but with slightly more
refinement owing to segments associated with channels.

Direct Marketing is Not Mass Marketing
Responses differ, so craft offers for segment-specific goals.
More offers, segments, faster feedback, constant resegmenting.
Many segments, and most
often derived from behavior.
Individuals uniquely identified.
Create
customized
offers
Execute
campaign
Record
results
Feedback

Better, but more complex, result measurement
?
The assumption in
direct marketing is this:
There is more visibility
into which campaigns
best lead to actions
and which customers
are best* but still a
visibility gap.
Direct marketers still treat people similarly, but with more
refinement owing to direct response visibility.
?
?

The Obligatory Funnel Diagram
Most of marketing’s efforts today are
in the area of customer acquisition.
Marketing tries to generate awareness
within an audience, some of which
become prospects.
When prospects are interested they
become sales opportunities.
When they are interested enough to
consider your product they are leads.
If they take action (purchase or
donate) they are customers.
Satisfied (happy) customers can
become proponents.
Audience
Prospect
Opportunity
Lead
Customer
Proponent

Conversion Rate: Where linking to the DW becomes important
Most marketing to sales metrics (particularly with web
analytics) track back to a single core transaction metric:
conversion rate.
Rates usually bounce around between 1% and 5%,
depending on industry.

0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Bounce rate
Take rate
Qualification rate
Abandon rate
Online conversion
Viewing Conversion Rates Properly
Mis-targeted
Opportunites
Leads Losses
Customers

The funnel concept is too narrow
It ignores non-transactional marketing. It ignores the
customer’s post-acquisition interactions with the organization.
After spending money to acquire a customer, attention must
be paid to service aspects of the relationship, and there are
multiple feedback loops.
Marketing analysis for brand
campaigns and customer acquisition.
Analysis for customer retention and
customer lifecycle management.
Audience
Prospect
Opportunity
Lead
Customer
Proponent
And these areas are not normally marketing’s
responsibility so not in their customer repository.

Post-purchase data is needed for experience measurement
You need to get post-purchase data and support tracking
of marketing activities for the post-sale period.
▪ 60% of consumers of facial skin care products
research more after purchase.
▪ 20% of evaluated CPG purchase decisions are made
differently at the point of purchase
You have to look at data from the perspective of
customer phases of activity:
Useful beyond marketing: product features, design, sales
experience (store, web, phone), service
Consider Evaluate Purchase Post-purchase

It’s a Multi-Channel World
There isn’t one
funnel, there are
many small funnels
that combine to
form the big
picture.
Each of these new
channels has
measurement data
coming from
different sources,
often external and
out of your control.

High Level View of Common Primary Channels
Communication
Channels
Radio
TV
Print
Online
Web site
Email
IVR
Social networks
SEM
Display ads
Mobile
Games
Each of these has individual metrics, some of
which are standardized in the traditional
channels but only for some online.

Communication channels are not transaction channels.
One drives people to the other.
▪ Measuring communication channels is about the communication
process and the expected outcomes in transaction channels.
▪ Measuring transaction channels is about the performance and
processes of those channels.
TransactionCommunication
Marketing Measurement = Channel Measurement
Online
Print
Radio
TV
Online
In-person
Phone
Campaigns can cross
communication channels.
TX channels overlap and
cannibalize each other.

The Online Channels Have Advantages
The online channel can provide realtime feedback
where old marketing channels sometimes have a
data lag of months.
Messages and offers can be
tested while a campaign is
running, rather than between
campaigns.
This requires that you collect
the data to manage the
process, while the process is
running.

Customer data is usually functionally aligned
Each function has specific applications and processes,
usually supported by local reporting and BI.
Sales automation
Audience
Prospect
Opportunity
Lead
Customer
Proponent
Customer master
Marketing automation
Call center automation
Sales BI
Marketing BI
Call center BI
Unifying customer information and contact processes
was something CRM was supposed to do for us.
Web?
Social media?
Mobile?

What Marketers (and Organizations) Say They Need
Individual, cross-channel, lifetime history: 360o

The customer experience is of supreme importance
…and it’s missing from the 360o view.

People do not experience the world in channels. “Online”
is just part of everyday life. They see one organization with
multiple touchpoints, not fragmented channels. This is how
you have to view your organization’s interactions.
Your view of them Their view of you

Customer experience has fuzzy definitions
A measure of quality, the total
of all interactions a customer
has with your organization’s
product or service.
▪ Utility
▪ Quality / reliability
▪ Ease of use / usability
▪ Aesthetics
An attempt to measure how
good your product or service is.
The basic thing being measured
is (quality of) interactions.

Experience is about integrating individual interaction preferences

Customer
Experience
Experience drives other metrics
Loyalty and satisfaction have a direct impact on financials
(revenue, profitability, market share), so the goal should
be to improve experience, which means (a) fix problems
(b) make things better
Utility
Quality
Usability
Aesthetics
Loyalty
Satisfaction
Likelihood to:
Repurchase
Discontinue
Defect
Recommend

What about Word of Mouth?
People talk about experiences.
Online measurement and social
software allows for watching
something which could only be
surveyed for in the past: what
people say.
▪ Awareness
▪ Sentiment
▪ Reputation
▪ Buzz
▪ Feedback
But these things are harder,
fragmented across multiple
external channels and sources.

Measurement? What measurement?
forrester

Kinds of customer analysis questions:
acquisition, management, retention, reactivation
What are the characteristics of our customers?
Who are our most profitable customers and how do we retain
them?
How to improve customer satisfaction?
How do we reactivate lost customers?
Attract new customers?
Increase sales?
Make our less profitable customers more profitable?
Manage high -risk customers?
Reduce acquisition or customer management expenses?
Dubai
April 28th,
2013

There’s a lot of hype related to big data and analytics

Closed loop process to monitor marketing
Create an offer
(promotion, call
to action, thing
of great beauty)
Publish it Monitor it React to it
Make
adjustments
(push it, tweak it,
fix it)
Stop it or
move on
Unlike older media models,
online/digital allows for
rapid adjustment.
We’re interested in the monitor/publish part, which is
more complicated than the usual BI work

Why is this more complicated?
The data sources
▪ Often (usually) not a DB
▪ Frequently want to see up-to-date results
▪ Many different sources / elements for one campaign
The data
▪ Need to see data for elements as well as the aggregate
(which might mean different/missing data for different
elements, e.g. video, twitter)
▪ Reaction / adjustment may change data collected
▪ Implies multiple, possibly changing data models
▪ Adjustments should be annotated, this is text on a timeline
▪ May throw it all away at end of a campaign
Create an offer
(promotion, call to
action, thing of great
beauty)
Publish it Monitor it React to it
Make adjustments
(push it, tweak it, fix it)
Stop it or move
on

3 common customer measures
Only sales taken into account. No valuation of differing
product margins, service costs, lifetime value of
customer. (Average Revenue Per User)
Revenue minus the cost of the product / service
(Average Margin Per User) Average across users, not
an individual metric
Total revenue revenue over the time the person has been
a customer to date (LTV) and forecast to an expected
date when they will stop being a customer (pLTV)
Averages are easy to calculate in BI, but these are usually an average,
e.g. sum(customer sales) / count(customers). They are not often
calculated for each individual in the database and they are missing
important information that would change how they are served.
Revenue
(ARPU)
Profitability
(AMPU)
Lifetime
value (LTV)

It’s always better to start simple, e.g. RFM
RFM: Table-based statistics without the stats
>40 year old technique used in direct marketing
Recency: Recent transactions indicate a customer is
more likely to transact again
Frequency: More frequent customers are more likely to
respond to an offer
Monetary: Customers who spend above average are
likely to spend more this time
The model assumes that R > F > M in importance.
This may not always be true.
43

RFM: Technique
Based on N quantiles for each attribute, where N is
commonly 5 but can vary.
Sort the data by recency, defined as the last time they
transacted with you on a voluntary basis (i.e. paying a
phone bill isn’t a good example) and assign each
record a quantile number
This numeric label is the recency
column in the table and is their
recency “score”
44
Record Set
1
2
3
4
5

RFM: Technique
Now do the same sort and assignment for frequency,
where frequency is by some meaningful time period for
your organization and customers.
Then repeat with monetary, e.g. avg order size, total
lifetime spend, etc.
The same row will have three numbers now.
Record Set
1
2
3
4
5
Record Set
1
2
3
4
5
Record Set
1
2
3
4
5
Recency Frequency Monetary

RFM: Technique
You can think of this is constructing a cube and
assigning one index value along each axis.
In this model, the cells will having varying
numbers of customer records corresponding to
their RFM score.
Recency
Monetary
1-1-1• Test a representative sample.
• Record the response rate by cell.
• Based on this and p(r) * v you can
now calculate breakeven by cell
and identify the ones to mail to
or avoid.

Why is RFM Used?
1. It’s better than random.
2. It’s simple.
3. It’s fast.
4. It’s cheap.
It’s usually not as good as other
models – until you factor in the
above points.
You can do several campaigns
with RFM for 1 run of a complex
scoring model.

Why simple might be better: Video Store case
Key Figures:
▪ 200 stores, 200.000 transactions per week
▪ Appr. 1 million active customers, 1.3 million inactive
▪ Average frequency 8-10x per year
Adapted RFM Model: FMR
▪ 'F' determines 'M', 'M' class influences 'R' class
▪ M = Revenue in 52 weeks prior to last visit
▪ Less than 52 weeks of data: weighted total
Assumptions (tested):
▪ No visit in last 52 weeks : inactive customer
▪ First 12 weeks: new customer
Similar model results, 2 minutes vs 2 days, SPSS consulting and
licenses and server vs 0 recurring external costs, better overall result

Analyzing the entire customer base another way
Pareto analysis of the share of buyers who make
up 80% of sales volume for products.
Data source: CMO council

What makes these customers different? How does this
affect a new product launch, or line extensions?
The idea of a mass
market is not really
true. More a set of
small markets within
the mass.
This is as true of
commercial supplies
as it is consumer
goods.
Data source: CMO council

Customer metrics today
should all be individual
Metrics go through periods
of being fashionable.
Average customer metrics
were fashionable in the 90s.
Today it’s all about
measuring the individual.
But can you? Probabilities
assigned to groups are not
an individual predictor.
Both of those rank each
individual rather than
providing averages.

Survival Analysis, Key to Lifetime Value Calculations
Based on a hazard function (mortality model).
Things survival analysis tells you:
▪ When a customer is likely to leave
▪ When a customer is likely to move to a new segment
▪ When a customer is likely to broaden or narrow the
relationship
▪ The factors in the relationship that may increase or
decrease tenure
▪ The quantitative effect of various factors on
customer tenure
Why it’s worth it: predict churn, risk of defection,
predict places to intercede to make things better,

Lifetime Value Calculations
Customer Value (CV) – the actual amount to now
Lifetime Value (LTV) – the predicted value to the
date determined from the mortality calculation
(or actual value in the case of inactive customers)
▪ Can be calculated based on Revenue or Margin*
▪ Can be absolute or discounted to current dollars
Customer Acquisition Cost (CAC) – the variable
and fixed costs to acquire the customer
Comparing any two customers requires knowing
their start date and expected tenure.
LTV = (CV – CAC) * Tenure

Satisfaction, Loyalty: CSI and NPS

LTV misuse
There was a flawed article about 5 years ago in a
business magazine with the idea that you should
“fire your worst customers.” First, find the LTV,
then get rid of the ones who are unprofitable.

GETTING STARTED

Evolution of the Customer Master
Demographic
• Your customer master
• Geography
• Group memberships
• Social network profiles
• Syndicated data
Psychographic
• Clickstream (visits, views,
timing)
• Communications
• Content (blogs, tweets,
comments, ratings)
• Transactions
• Syndicated data
Socialgraphic?
• Profiles of connections
• Communities of interest
• Connection likes and
dislikes
• Syndicated data
The detail and breadth expand as collection improves

Individual profile
Metrics (LTV, Satisfaction,
loyalty, influence, etc.)
Product preferences
Brand preferences
Price preferences
Seasonal preferences
Communication
preferences
Service preferences
Transaction preferences
What we need to build: the real customer master,
not a customer dimension
A lot more than just demographics. Includes every transaction,
interaction, observation about the customer. Not built all at once.
Household Household profile
Transaction history
Service history
Outbound marketing
contact history
Inbound marketing
contact history
Social networks*

The Advanced Companies Are Integrating Social Data into
Customer Databases
Do you have a process to record and integrate data from social interactions with customers into existing
customer databases?
Source: Survey for Social Media Program
Managers, conducted by Altimeter Group (Q1-
Q2 2011)
143 respondents, all over 1000 employees

Measuring customer interactions
State of practice:
▪ Limited customer master, in a silo. Partial customer contact
histories, outbound only.
State of the art:
▪ A system to gather data about any customer interactions,
inbound or outbound, from marketing to service.
Sources are extensive:
▪ web analytics, CRM, sales automation, lead scoring, marketing
automation, email providers, online advertising, call management,
IVR, bug reporting, warranty, service management, enterprise
feedback mgmt, decision automation, sentiment analysis,
conversation management, dynamic case management, VoC, social
media, surveys, focus groups, syndicated market data…
Many types and structures, not all relational-friendly, may require
new storage and processing platforms.

Different Data and
Different Usage Patterns
Be prepared for changed
assumptions:
▪ Marketing campaigns and
tactics change frequently,
particularly online marketing.
▪ This means data sources
change frequently, as well as
reporting needs.
▪ Much of marketing is like
experimental science, and
unlike the read-only BI usage
model.

Value
1. Web alone, not that much
2. Web plus transactions, big
increase
3. Web plus email plus
transactions, bigger increase
But the real goal in customer
analysis is a full picture, which
means an interaction master for
everything: online ads, mobile
app use, inbound & outbound
call center, physical point of
presence, direct mail contacts.
i.e. You have to say yes to all
new data requests…

The data warehouse has to evolve to support these
kinds of uses. This means a more complex data platform
Transient data
Raw data
Infrastructure layer:
Store and manage
Refine and deliver
Application layer:
Analyze and consume
The new model encompasses data at rest and data in motion
Multiple access methods
Standardized
data
Multiple ingest methods
BI, data extracts,
analytics, applications
The platform has to do more than serve queries; it has to be read-write.

Define the goal
Decide how to measure it
Baseline
Plan your actions and monitor
outcomes
Iterate
Define goal-driven actionable metrics as the starting
point, don’t boil the ocean

Getting started
You can’t build everything at once, so don’t try.
1. Build out the convenient: for a few channels that are
important, assemble the raw interaction data for each.
2. Store that data in a place where it can be be easily
refined from its raw state, and hopefully further linked
to DW data on TXs (which will show you outcomes).
3. Add some inexpensive tools to explore the data with.
4. Do some simple modeling to understand a few key
customer channel interactions.
5. Keep extending the data and capabilities.

“The future, according to some scientists, will be exactly like
the past, only far more expensive.” ~ John Sladek

Creative Commons
Thanks to the people who made their images available via creative commons:
Outdated gumshoe.jpg – http://flickr.com/photos/olivander/372385317/
Card catalog – http://flickr.com/photos/deborahfitchett/2372385317/
Book of hours manuscript2.jpg – http://flickr.com/photos/jeffrey/89461374/
Royal library san lorenzo.jpg – http://flickr.com/photos/cuellar/370663920/

Image Attributions
Thanks to the people who supplied the images used in this presentation:
outdated gumshoe.jpg - http://flickr.com/photos/olivander/372385317/
laptop face.jpg - http://flickr.com/photos/sd/7746599/
anne hathaway.jpg - http://flickr.com/photos/barbaradoduk/177959197/
desert ibex - michael polizia wheat_field.jpg - http://www.flickr.com/photos/ecstaticist/1120119742/
Open air market - http://flickr.com/photos/baboon/309793875/
child workers - http://www.flickr.com/photos/mtowber/4222787490/
whack-a-mole_door.jpg - http://www.flickr.com/photos/mtowber/4222787490/
uniform_umbrellas.jpg - http://www.flickr.com/photos/mortimer/221051561/
train_to_sea.jpg - http://www.flickr.com/photos/innoxiuss/457069767/
riot police line small - http://flickr.com/photos/73594239@N00/25719098/
changing of the guard.jpg - http://flickr.com/photos/mambo1935/160739264/
pyramid_camel_rider.jpg - http://www.flickr.com/photos/khalid-almasoud/1528054134/
chinatown little color gate.jpg - http://www.flickr.com/photos/paullikespics/3248133830/
teapot.jpg - http://flickr.com/photos/joi/411403/
motionless in crowd.jpg - http://flickr.com/photos/cactusmelba/1065738186/
well town hall - http://flickr.com/photos/tuinkabouter/1135560976/
febo amsterdam.jpg - http://flickr.com/photos/jshyun/1573065713/
cadillac ranch line.jpg - http://flickr.com/photos/whatknot/179655095/
Slide 77

CC Image Attributions
Thanks to the people who supplied the creative commons licensed images used in this presentation:
Veyron - http://www.flickr.com/photos/guano/414733876/
unrecycle.- http://www.flickr.com/photos/xiaming/94328953
wheat_field - http://www.flickr.com/photos/ecstaticist/1120119742/
Open air market - http://flickr.com/photos/baboon/309793875/
riot police line small - http://flickr.com/photos/73594239@N00/25719098/
Sandblaster - http://flickr.com/photos/stigeredoo/11643424/
subway dc metro - http://flickr.com/photos/musaeum/509899161/
fast kids truck peru - http://flickr.com/photos/zerega/1029076197/
well town hall - http://flickr.com/photos/tuinkabouter/1135560976/
Tokyo forum - http://flickr.com/photos/fukagawa/2004106475/

About Third Nature
Third Nature is a consulting and advisory firm focused on new and
emerging technology and practices in information strategy, analytics,
business intelligence and data management. If your question is related to
data, analytics, information strategy and technology infrastructure then
you‘re at the right place.
Our goal is to help organizations solve problems using data. We offer
education, consulting and research services to support business and IT
organizations as well as technology vendors.
We fill the gap between what the industry analyst firms cover and what IT
needs. We specialize in strategy and architecture, so we look at emerging
technologies and markets, evaluating how technologies are applied to
solve problems rather than evaluating product features and vendor market
positions.

A Pragmatic Approach to Analyzing Customers

More Related Content

What's hot (20)

Viewers also liked (11)

Similar to A Pragmatic Approach to Analyzing Customers (20)

More from mark madsen (20)

Recently uploaded (20)

A Pragmatic Approach to Analyzing Customers