This document discusses modeling customer relationships through a flexible, integrated data architecture. It describes evolving approaches to customer databases, from flat files to more sophisticated dimensional models. The key elements of a customer relationship management database are outlined, including tracking customer, product, contact and event details over time through a customer-centric model. This approach facilitates iterative questioning, contact list refinement, and integration with campaign management and analytical tools to better understand customer relationships and behavior.
2. A b s t r a c t
This white paper draws upon the lessons learned by Sequent
Computer Systems in implementing large-scale technology platforms
to support Customer Relationship Management (CRM) strategies
within major organizations worldwide. It is assumed that the reader
is familiar with the strategic direction of most customer-focused
organizations and understands the cyclical and repeatable nature of
a technology-driven marketing strategy. Interested readers are likely
to be concerned with how they might model customer relationships
in a way which will support the transition from present product-
focused views of marketing intelligence to more useful (and prof-
itable)
customer-focused views. This paper provides the data architect or
modeler with a generic template for modeling customer data. This
approach is not product or technology specific but does provide
a flexible data architecture for integrating the various technology
components that use data to drive marketing. This paper also
highlights commonly faced problems that occur when modeling
customer data.
Editor: David Puckey
Sequent Computer Systems
UK Professional Services
1
3. Introduction used to ensure that the core data struc-
The strategic importance of managing tures in the CRM technology layer
customer relationships both drives support the integration of the various
and is driven by technology. In par- components of the marketing process
ticular, this applies to data and the and reduce the time required to design
increasingly sophisticated and useful and execute a campaign. This approach
ways in which data is used to model enables the creation of a complete model
relationships and to drive contact of customer relationships over time.
strategies. At the core of any technology
enabler for CRM is the customer The Evolution of the
database. The customer database rep- Customer Database
resents the data hub that integrates Current approaches to the design of the
the various statistical modeling, cam- customer database fall broadly into two
paign management, contact history camps. The first—the flat earth view of
and response tracking components of the world—hails from the glory days of
the marketing campaign lifecycle. target marketing in the late 1980s.
This is true whether the database is This approach, which is popular with
used for the execution of marketing list providers and bureau operations,
strategies (e.g., generates mailing lists), utilizes the concept of the customer
or whether it exists purely as an file or list. Such a list tends to offer a
analysis engine that passes contact current snapshot of the customer or
strategies and information to a separate prospect base and is often the product
customer interaction platform for of much tortuous cleansing, de-duplication
execution (e.g., customer call centers). and point-in-time segmentation. This
approach makes it difficult to analyze the
The technology layer and its integration ups-and-downs of an organization’s
with emerging business processes is relationship with a customer over time
therefore key to the successful imple- due to its current snapshot view of the
mentation of a data-driven Customer customer and prospect base. Further, it
Relationship Management strategy. This typically lacks the transaction-level detail
paper describes, in a generic way, an and promotional history needed to model
approach that Sequent has successfully customer behavior.
Evolution of Customer Database
Campaign Customer Relationship
Management Management
s contact horizon s one-shot s sequence
s output s offer s information
s systems s mail/phone s touchpoint
s execution s manual s scheduled
s departments s marketing s front-office
s data types s purchases s contacts
s update interval s monthly s daily
s reaction time s billing cycle s transaction
s goal s reduce waste s add revenue
Source: Raab and Associates
Figure 1: Customer databases evolve with integration of technology and business processes
1
4. The second approach has evolved from approach incorporates the maximum
the data warehousing movement and degree of analytical flexibility for the
Sequent’s experience in helping hundreds marketer and marketing analyst with
of organizations design and implement the efficient scoring, segmentation and
data warehouses. Sequent has developed extraction of data to execute marketing
a mature methodology for delivering campaigns or contact strategies. It also
rapid business benefit by integrating places the customer or prospect at the
sophisticated analytical tools with center of the model and seeks to model
subject-oriented and time-consistent all facets of a relationship with that
central databases. Such systems customer over the known lifetime of
typically concentrate on the delivery the relationship. Sequent’s approach to
of business intelligence and are generally the design of customer databases is not
not designed to plug directly into an list based and is not designed to simply
organization’s day-to-day operations. support ad-hoc, point-in-time marketing
However, the modeling techniques solutions. Rather, the objective is to give
employed by Sequent for the delivery the marketer true insight into the vari-
of successful data warehousing projects ability of his relationship with a customer
represent a radical shift in emphasis or customer segment over time and to
from both flat earth views of data and deliver seamless integration with the
the microscopic views of data used in widest possible choice of campaign
online transaction processing (OLTP) management and statistical modeling
systems. Sequent’s dimensional view of tools available.
data provides the optimum combination
of analysis of facts over time and high Elements of a Customer
system performance when dealing with Relationship Management
large data volumes. Database
There are a number of required features
Sequent’s approach to successfully of a CRM database that the architect
delivering large-scale technology must integrate in order to support the
platforms to support CRM strategies marketing lifecycle. These are (in no
uses the best attributes from both particular order):
of the previous approaches. This
PRODUCT HOLDINGS
PRODUCT USAGE
CONTACTS WITH CUSTOMERS
EVENTS
TIME
Figure 2: Typical facets of a customer relationship that need to be tracked over time
2
5. CONTACTS EVENTS
CUSTOMER
s AGE
s GENDER
s ADDRESS
s SEGMENT_ID
s PROPENSITY SCORE
s SUPPRESSIONS
PRODUCT PRODUCT
HOLDINGS USAGE
Figure 3: Customer focus is key. Each facet of a relationship may be treated as an island of analysis,
linked centrally to the customer
s Customer or prospect focus Taking a Lifetime View
s All relevant facets of the relationship of the Customer
over time In order to fully realize a CRM strategy,
s Integration of external prospect lists the marketer must have information
s Integration of external data classifi- that enables him to take a lifetime view
cations of the relationship. A relationship is
s Integration of external data most usefully defined as the starting
enrichment point at which the organization has an
s Ability to directly score the database initial interaction with a prospect. This
and segment the database many times relationship then needs to be tracked
s Ability to evaluate different campaigns as the prospect is encouraged to climb
and treatment strategies over time the loyalty ladder from prospect to
and across millions of transactions customer and eventually to highly
and customers valued customer. The marketer needs
s Campaign management, prioritiza- to see and understand past events,
tion, etc. contacts and purchase information in
s Ability to predict future customer order to assess the current and future
behavior based on past behavior profitability of the relationship. The
commonly used marketing analysis of
It is not possible to achieve all of the recency, frequency and monetary value
above features using either a flat file of transactions indicates some of the
approach or a standard data ware- facets of the relationship that should
housing approach alone. be tracked.
3
6. In Sequent’s experience of facilitating at a given point in time. For example,
client workshops to establish the busi- the marketer may take a point-in-time
ness requirements for a CRM solution, view of the relationship, a view over
four relationship facets appear common time or make prescient predictions for
to most organizations. These facets are: the future. Information about these four
facets of a customer relationship enable
Product Holding–What products has a the marketer to answer questions such
customer purchased and what products as: How many customers have bought
do they currently hold ? product X? How many customers
display a repeatable purchasing pattern?
Product Usage–How has the customer How often have I contacted this customer
used that product? For example, can an and when? Who are my most profitable
increase in credit card usage be attributed customers? What events or contacts
to some prior interaction with the cus- occurred prior to customer defection?
tomer or some promotional activity?
The approach taken by Sequent to sup-
Contacts–What has the organization’s port this kind of questioning is to place
interaction with the customer been over a customer table at the center of the
time and what were the outcomes? model and to surround it with satellite
dimensional schema (star schema) rep-
Events–What other events have resenting each facet of the relationship
occurred, either within the life of the to be modeled. Modeling the facets of
customer (e.g., marriage) or externally the relationship dimensionally allows
to the relationship (e.g., competitor who, what, when, where style analysis.
activity)? For example: Which segment bought
which products and what contacts
Each of these facets may be treated by preceded which purchase? Where do
the modeler as an island of analysis the contacts live, and how do they
linked centrally to an individual customer like to be addressed?
Customer Cancellation of Terminate
Customer Initial Service part of policy
Behavior Inquiry Call
No Activity Purchase No Activity Re-initiate
Customer Acquisition Mail Information Customer "Next to Buy" New Product Customer Valuation/ Winback
Action Campaign Kit/Thank You Valuation model Solicitation Solicitation
Sequent Campaign Query/ Query/ Campaign
Campaign Data Mining Campaign
Decision Management/ Reporting/ Reporting/ Management/
Management Application Management
Advantage Call Center OLAP OLAP Call Center
Application Application Application
Figure 4: Example—The Customer-Centric Model at an Insurance Company
4
7. The customer-centric nature of the Mr. Jones does not respond to the
model also lends itself well to the receipt of the information pack, and
prudent de-normalization of often- after three months the marketer plans a
used facts, such as disposable income campaign targeted at Mr. Jones and all
estimates, onto the customer table and the other Mr. Joneses who have inter-
helps facilitate the efficient extraction acted with the organization but not
of contact lists and integration with purchased any products in the last
statistical modeling tools, such as SAS three months.
or Unica. The customer-centric model
also supports very well the iterative In this case, a query can be run against
nature of the marketer’s questioning, the database asking, Who has contacted
such as: How many customers hold us in the last three months with a con-
product Y? Which of those customers tact type of inquiry? This query will
are profitable? Which of those customers generate a list of keys into the customer
did I contact last week and which of or prospect table, which, without further
them complained about the contact? It refinement, could be used to generate a
is also possible to assess what behavioral contact list. However, it is more likely
changes are exhibited as a result of that the marketer’s questioning will
identifiable interactions with the cus- continue further—How many of these
tomer. Once the marketer has exhausted customers or prospects were sent an
his questioning, which helps refine the information pack? The result set from
contact list names, addresses and saluta- this query will be matched against the
tions may be simply extracted from the result set from the last query to further
customer table using the relevant keys. refine the list of keys. This process may
Current suppression indicators and be further refined by asking, How many
propensity scores may also be stored people in this list do not have a product
against the central customer record, holding? Once the marketer has com-
allowing the possible automation of pleted his refinement of the list, it is a
standard hygiene filtering. simple, and highly performant, exercise
to take the resulting list of keys and
The Customer-Centric Model extract the name, address, salutation
at an Insurance Company data, etc. from the central customer
To see how this model might work, or prospect table and perform further
take the example of an insurance filtering based on suppressions on the
business. The firm’s relationship with customer table or assigning customers
Mr. Jones begins when he makes an to campaign cells for different treatments
initial inquiry about health insurance based on segmentation keys on the cus-
via the organization’s call center. This tomer record. Once the contact list is
initial inquiry is the result of a press finalized, the customer keys are used to
advertising campaign that reached populate the contact table and to record
Mr. Jones; this fact is recorded. the fact of the outbound contact. By
storing all of this data in a centralized
In response to his interest in the company’s relational database management system
health insurance offering, the insurance (RDBMS), it is a relatively simple matter
business sends Mr. Jones an information to make this data available to sophisti-
pack. This step is also captured and cated campaign management tools and
recorded in the database. At this point, statistical modeling tools. These tools
Mr. Jones does not have a product hold- interface easily with an open RDBMS,
ing, but his name and address and con- such as Oracle, and almost without
tact records exist within the database. exception, such tools feature native
connectivity options.
5
8. CAMPAIGN
MANAGEMENT REPORTING
BUSINESS
CUSTOMER INTELLIGENCE
SEGMENTS DATABASE
SCORING
MODEL
CONTACT LISTS
EXTRACTION TOOL
DATA MINING
Figure 5: An integrated architecture reduces the marketing cycle
Those readers familiar with the pro- marketing analysis or campaign man-
cessing dynamics of most RDBMS will agement) and to temporarily satisfy
immediately spot a major dependency parochial needs, it has left a troublesome
of this model—the various software legacy for the integrator of the technol-
components deployed to support the ogy layer who seeks to accelerate the
marketing lifecycle must allow the marketing cycle, empower the marketer
generation of interim result sets. This and reduce the marketing department’s
is absolutely crucial in order to support dependency on highly skilled and
the marketer’s analytical processes as he expensive (and often obstructive)
constantly shrinks and expands potential database experts. Such function-focused
target lists, possibly to generate the solutions have ensured that the walls
required list size to match a budget that block the implementation of a
allocation. Already, a number of tools virtuous circle of continuous improve-
vendors are acutely in-tune with the ment in the marketing process remain
mindset and thought processes of the solid. The proliferation of file formats,
modern marketer. APIs and unnecessary processing layers
needed to integrate these elements have
Integrated Infrastructure delivered a full employment charter for
Supports Marketing Process those who wrangle with the complexity
In the past, database marketing solutions of the technology layer at the expense
often focused on individual user com- of marketing responsiveness and creativity.
munities participating in the overall
marketing process. While this focus Sequent’s solution to such technical
has managed to hit the sweet spots of anarchy is to focus firmly on a techno-
these often isolated communities (e.g., logical infrastructure that supports and
6
9. integrates the overall marketing process, s Identification of significant life
and underpins the progressive develop- events (coming of age, birth,
ment of a relationship management marriage, etc.)
strategy. The use of a centralized s Analysis of geodemographic data
relational database and open systems to by household
manage customer data, contact history
and relationship history allows the easy Multiple households can be problematic
integration, at the data level, of the var- for both the marketer and the system
ious technologies deployed at different designer. Individual customers may
stages in the marketing process. Analysts’ have multiple addresses, each of which
models may be stored alongside the is related to the customer via the product
actual data, and scoring and segmentation holding. For example, Mr. Jones has a
keys can be made directly available to main residence in the city and a weekend
campaign management and campaign retreat by the coast. Mr. Jones has a
scheduling software. The automation of household insurance policy for each
routine communications is simplified address. An insurance marketer may
and database triggers can be utilized to wish to sell Mr. Jones a life insurance
make marketing more event driven. policy. However, for the modeler, a
household is just a simple grouping of
Typical Data Modeling individuals. Specific business questions
Challenges must be answered in order to track
This section details some of the data the household movements of individuals.
modeling challenges, which, in The difficulty is in the actual identification
Sequent’s experience, are common of a household—particularly in high-
across a number of industries and density urban residential areas or areas
organizations. with a highly transient population.
Householding There are several approaches to handling
The grouping of individuals by house- customer householding, de-duping and
hold or relationship patterns is often geocoding challenges. These include:
a difficult process in product-focused s Service Bureau operations
legacy systems. These systems often s Integrating specialized software
have great difficulty in even identifying the tools to perform this function on
individual responsible for purchasing a a regular basis (this also requires
given product. The benefits of groupings process integration for proper and
for the relationship marketer are many: effective handling)
s Avoidance of unnecessary duplicate
contacts per household A number of marketing data processing
s Understanding loyalty patterns bureau services perform household
among relationship groups identification, based on, for example,
s Identification of cross-sell and electoral register information, etc.
up-sell opportunities (e.g., family However, such matching is never
policies, etc.) 100 percent accurate.
7
10. Products Held by Groups grouping. In some cases, Sequent has
of People allowed a “degree of confidence” value
Certain types of products, for example to be assigned to the grouping record to
joint bank accounts, introduce a many- provide the marketer with a coefficient
to-many relationship between product that validates assumptions. The business
holdings and persons. This fact, if rules for deriving this coefficient clearly
modeled literally, can cause performance evolve over time, and can result in the
problems in the database and confuse creation of specific profiling questions
campaign management and extraction targeted to specific customers during
tools seeking to identify a single interactions.
prospect. This is particularly true in
cases where organizations are transi- As with householding, some marketing
tioning to a customer-focused marketing data providers can perform unique
strategy yet still require the ability to person identification based on postal
market in the interim period based on lists, real estate listings, electoral rolls,
product holding attributes. This situation and other data. This identification
is common in large businesses that cannot activity can be cumbersome as it
possibly switch from a product to a involves exporting and re-importing
customer focus overnight. The only data periodically. If the grouping of
answer to this problem is a business one. seemingly multiple individuals into
Identifying a primary marketing contact one is handled as a grouping table,
for a product holding can simplify the the impact on, for example, referential
problem in some cases. integrity within the database can be
minimized. However, this kind of
Person Matching group can also make the model more
Another key challenge for the designer complex—with a possible impact on
of a CRM database is the identification performance.
of individuals. Often, seemingly multiple
individuals on the database are in fact Unfortunately, there are no magic cures
the same person, albeit at a different for the problem of person matching,
point-in-time, or with a different product and the database modeler should be
holding, or at a different address. wary of the purveyors of such cures.
Organizations with multiple operational
systems serving multiple customer touch Classing and Banding
points often find that the non-uniformity A number of marketing database designs
of input validation across these systems use fields such as “date of birth” or
leads to situations where Mr. John Jones, “age” on the customer record. Though
Mr. J. Jones and Mr. J. B. Jones at the there is a clear use for such fields, mar-
same address could perhaps be one, keters rarely wish to contact people who
two or three actual people. This problem are, for example, 51 or 23 years of age.
is further exacerbated when external Usually, the marketer wants to target
prospect lists are brought into the people aged between 25 and 35 or
database. Once again, the modeler can those who are past retirement age.
incorporate a simple grouping of people Such targeting calls for some sort of
within the database design but the banding of customers to reduce wasted
problem is identifying the actual processing and simplify the process for
the marketer.
8
11. Age is not the only candidate attribute approach will, on its own, support the
for banding. The modeler should seek management of customer relationships
to understand other candidates and over time. Likewise, neither will integrate
include these in the model. all components of the marketing
process in the most efficient way.
Regularly Used Measures
Initially, and over time, the modeler of The template presented in this paper
the customer database should seek to may form the basis of the data architect
identify those frequently asked market- or analyst’s initial attempts to define
ing questions, such as: Who earns more data structures, which will support both
than $20,000? Who has made more of the above objectives. This template
than four insurance claims in the last reflects the work Sequent has done with
period, etc.? It makes little sense to a number of major organizations to
have multiple marketing campaign support their database marketing activities
designers all scanning the product usage and to drive the strategic implementation
table over and over again. This can be of Customer Relationship Management
avoided by denormalizing regularly at both the business and the systems
used measures directly onto the cus- levels.
tomer or prospect record.
CRM is an emerging strategy and as
Suppressions such requires a fresh approach to sys-
Most organizations are able to identify tems design, along with the flexibility to
a number of standard reasons for sup- accommodate unexpected change.
pressing marketing communications. Many piecemeal or point solutions in
Suppressions can range from blanket the market fail to take an integrated
“do not communicate at all” indicators view of the entire marketing lifecycle
to “do not market a specific product” and focus only on data structures to
to this individual. These suppressions support their own specific components
should be held directly on the customer of that lifecycle. As CRM matures as an
or prospect record to enable swift and operational reality, it is imperative that
easy filtering of targets. organizations have an integrated view
of business processes and data. Failure
Summary to take an integrated view of requirements
will lead to significant effort and cost
While both flat file and standard data
reengineering the organization’s market-
warehousing approaches to the customer
ing databases—sometimes comprising
database will allow analysis of customers
many terabytes of data.
and the selection of target lists, neither
9