|
THE FIVE STAGES OF AN
ANALYTIC CRM EVOLUTION
Mass marketing doesn't fit anymore. It's time to customize.
Stephen Brobst and Joe Rarey
THE
EMERGENCE OF ACTIVE DATA WAREHOUSE architectures has enabled
a new generation of analytic applications in the business intelligence
arena. With traditional data warehouse implementations, the focus
is on reporting and strategic decision-making applications. But
with active data warehouse deployments, strategic decision support
is augmented by tactical decision-making applications.1
The architecture of an active data warehouse is designed to acquire
data in near real-time, deliver analytics for tactical decision
support queries in seconds and provide 24/7 availability. This
provides opportunities for delivering applications that require
more aggressive service levels in the areas of performance, availability
and data freshness.
Analytic customer relationship management (CRM)
applications have been among those most benefited by the emergence
of active data warehouse implementations. The ability to go beyond
strategy development for CRM using a traditional data warehouse
to using an active data warehouse delivers huge advantages for
increasing effectiveness of CRM deployment. The evolution of analytic
CRM applications involves five distinct stages, each closely tied
to the underlying capability of the data warehouse foundation.2
The five stages are mass marketing, segment marketing, targetmarketing,
event-based marketing and interactive marketing.
 Each
stage demands increasingly sophisticated capabilities from the
data warehouse to support the analytic foundation for CRM decision-making.
This article describes each stage along with the underlying data
warehouse prerequisites for evolving to subsequent stages (Figure
1).
Stage 1: Mass Marketing
The primary focus of analytic CRM applications for mass marketing
is automation of list selections. The data warehouse repository
is used to store customer names, addresses, product ownership
and channel preferences. The data warehouse is refreshed on a
periodic basis, usually weekly or monthly. Data extracts are used
to deliver name and address information to mail house operations,
call centers, e-mail servers and other channels. Product ownership
information is used to exclude customers who should not receive
offers for products that they already own.
In this initial stage of analytic CRM, almost
all access to the data warehouse is batch oriented. Data loading
is performed on a batch basis with input files extracted from
source systems across the enterprise. List selections are also
executed as batch jobs, often using hand-coded SQL programs. The
biggest challenge for most organizations in Stage 1 deployments
is data quality. Constructing a single source of truth whereby
each customer appears once and only once-even those with
multiple product relationships-can be quite a challenge
if there is no "easy" match key, such as a taxpayer
identification code or enterprise customer identification number.
Name and address matching is fraught with pitfalls and should
never be attempted with "homegrown" coding. Use a
package that understands the heuristic nuances for name and address
matching (e.g. Trillium, FirstLogic, GroupOne, etc.), and make
sure the package is CAS (Certified Address Standardization) compliant
if your customers are in the U.S. marketplace.
The goal of the mass marketing stage of analytic CRM is to reduce
the cost for blanket marketing campaigns. Automated mass marketing
allows for efficient contact with anyone and everyone for whom
a name and contact information (address, telephone number and
e-mail address) can be obtained. In mass-marketing operations,
products (and services) are standardized and customers are interchangeable.
However, getting the right offer to the right individuals requires
a bit more sophistication than the mass marketing approach. The
problem with mass marketing is that most individuals are completely
over-saturated with direct marketing offers, and reasonable response
rates will not be obtained from a "one size fits all"
approach.
Stage 2: Segment Marketing
The segment marketing approach is focused on delivering product
offerings differentiated by customer segment. Rather than using
the blanket approach of Stage 1 analytic CRM, in Stage 2 customers
are not interchangeable when delivering product offers. The data
warehouse is more than just a big database of names and addresses.
Ad hoc analysis is used to develop an understanding of customer
segments and design product offerings appropriate to each segment.
Marketing campaigns are then targeted accordingly.
There will be an increase in the number of list
selections in Stage 2, but the number of names for each selection
will be smaller because targeting is more precise than with the
mass-marketing approach. A packaged tool for facilitating segment
definitions, product analysis and campaign management becomes
essential in Stage 2. In addition, a richer set of data about
customer demographics, synchographics and psychographics is desirable
to accurately formulate appropriate segment definitions. This
data can be acquired from external data sources (e.g. Experian,
Axciom, InfoUSA, etc.), or internal processes can be redesigned
to capture the desired customer attributes (a combination of these
two approaches is typically the best strategy).
The requirements of the data warehouse change
dramatically in Stage 2 deployment because direct access to data
by marketing analysts is essential. Packaged tools in the analytic
CRM space allow knowledge workers to interactively define and
refine customer segments and product offerings. Batch access via
SQL programmed by the IT department, as is often used in Stage
1 implementations, becomes unacceptable.
Performance management and proactive capacity
planning for the data warehouse is a prerequisite to success in
Stage 2 because knowledge workers are directly impacted by performance
problems. Moreover, the ability of the underlying data warehouse
to efficiently execute increasingly complex and unpredictable
SQL statements generated by point-and-click marketing analysis
and campaign management tools will put a lot of stress on the
underlying cost-based optimizer for the RDBMS engine.
The well structured SQL with hand-optimized
access paths (and possibly programmer-inserted hints) goes out
the door once knowledge workers have point-and-click access to
data via the semantic metadata layer provided by a packaged tool
for analytic CRM.
Stage 3: Target Marketing
The fundamental innovation in Stage 3 analytic CRM is to treat
each customer as an individual. The introduction of data mining
technology into the analytic CRM environment plays a crucial role
in target marketing deployment. Predictive models are constructed
to understand individual customer behaviors and to begin developing
one-to-one customer relationships. In the target marketing environment,
every customer and prospect is scored based on their propensity
to buy-even before a marketing communication takes place-to
determine if it makes economic sense to initiate the communication
to that individual. A simple break-even calculation using propensity
to buy, multiplied by expected value upon purchase, can be compared
to the variable costs associated with initiating contact to determine
if the communication should be undertaken. The key point is that
the calculation is performed on each individual customer with
unique results.
The predictive analytics involved in Stage 3
usually include clustering techniques for more advanced segmentation
than is the case in Stage 2 analytic CRM deployment. In addition,
data mining techniques are used to predict the lifetime value
of customers, which customers are at risk of attrition, cross-sell
opportunities and so on. The result is higher-resolution targeting
in marketing campaigns. Typically, an organization will reduce
its direct marketing (variable) costs by 50% or more when evolving
from Stage 2 to Stage 3 of analytic CRM. More importantly, however,
is that conversion rates will skyrocket as a result of more effective
targeting.The data warehouse supports data mining efforts by providing
direct access to detailed data. In the past, data mining was often
performed using data extracted from the warehouse into proprietary
file structures (e.g. SAS data sets). The problem with this approach
is that there is significant overhead in performing the data mining
outside of the data warehouse. The storage costs for duplicating
data and the burden of extracting data and moving it on to a separate
analytic server can be significant. For small training sets, this
overhead is not such a big issue. However, as data mining efforts
are increasingly targeted at identifying subtle (infrequent) customer
behaviors, the training sets for the data mining algorithms necessarily
increase in size (and cost).
New developments in the SQL standard and built-in
analytics within the RDBMS products are clearly shifting the primary
data mining approach from external data sets to "in-database"
data mining. All major RDBMS players in the data warehouse marketplace
(e.g. IBM, Oracle and Teradata) have introduced data mining products
that work directly within their engines. Moreover, the current
generation of third-party data mining tools entering the marketplace
(e.g. KXEN and Data Distilleries) are largely focused on building
and scoring models directly on top of relational data warehouses
rather than using proprietary file structures. This makes accessibility
to scoring output much easier, but the process must be considered
in capacity planning for the data warehouse platform.
Stage 4: Event-based Marketing
Stages 1 to 3 of analytic CRM deployment are largely focused on
a batch-oriented model for marketing automation. Data is periodically
extracted from source systems, integrated into the data warehouse
and analyzed in order to target marketing communications to specific
customers (with varying degrees of accuracy, depending on the
stage of implementation). In Stage Four, a new paradigm is introduced.
Event-based marketing focuses on enabling personalized marketing
communication driven by individual behavior patterns.
A key architectural component required for the
realization of event-based marketing is a framework for enabling
"software event detectives." Such architecture allows
you to capture business rules for identifying events relevant
to a customer relationship and then initiate very personalized
communications in response to such events. Events can involve
detected changes in customer behaviors, such as a difference in
banking activities or calling patterns. Event detectives for proactive
retention programs might relate to airline customers who are subjected
to delayed flights, wireless customers who experience dropped
calls, or retail customers who return merchandise for quality
reasons. Notice that event detection can be arbitrarily complex:
Looking for large withdrawals from a bank account might seem simple,
but can be complex to individualize because of seasonal considerations
(e.g. tax season) and the fact that what is large for one customer
might be small for another.
An important dimension of relevance is the timeliness
with which events are detected and action is taken. I recently
had a letter of apology and travel voucher waiting for me when
I got home after having a long flight delay at the beginning of
a business trip. That made a greater impact on me than if the
letter had come a month later. Best practices for most analytic
CRM activities dictate that event detectives should be executed
at least on a daily basis. This means that data warehouse content
must be incrementally refreshed on a daily basis to make customer
activities visible to the software event detectives. Moreover,
depending on the nature of the business scenario, even more frequent
event detection and outbound communications might be demanded.
Many organizations use software event detectives
to identify opportunities for contact by personal relationship
managers and forward leads to customer service representatives
within minutes or hours of the relevant event. These more aggressive
service levels for event detection imply more frequent (continuous)
data acquisition into the warehouse. This is consistent with the
trend toward extreme data freshness in active data warehouse deployments.3
Stage 5: Interactive Marketing
The interactive marketing stage of the analytic CRM evolution
involves integration between analytic and operational CRM. Operational
CRM is focused on managing the workflow of a customer interaction.
However, simply automating a call center with operational CRM
packages such as Siebel or PeopleSoft is not enough. Making a
customer interaction more efficient without making it more intelligent
is a business disaster. This is why integration between analytic
and operational CRM is so critical.
 The
trend toward component-based architecture and Web services deployment
provides a perfect framework for facilitating interoperability
between the operational and analytic CRM capabilities within an
organization. When the operational CRM application identifies
a customer on a Web site or at a call center, it packages up an
XML message identifying the customer and interaction. This message
is sent to an application integration server compliant with standards
such as .NET or J2EE. The message is delivered to an analytic
CRM application that retrieves relevant data about the customer,
scores the situation to assess interaction alternatives and returns
a proposed treatment plan (script or Web content identifier).
The XML response message is returned back through the application
integration server to the operational CRM application. The standards-based
application integration server and XML messaging allow seamless
cooperation between analytic and operational CRM components of
an enterprise architecture. Moreover, multiple channel interfaces
can all hook into the analytic CRM applications through the application
integration server in a seamless way. The services-based architecture
described herein has huge benefits in that customers receive consistent
treatment across channels, and organizations are able to avoid
the burden of ODS (operational data store) deployment for each
channel interface to the outside world (Figure 2).
The data warehouse service levels required to
support interactive marketing are extreme. Performance for analytic
CRM queries must often be sub-second to allow immediate response
when cooperating with the operational CRM environment. Moreover,
analytic CRM queries are not simple retrievals; complex analytics
are involved in scoring for offer determination on an individual
customer basis. Availability service levels for the active data
warehouse also become more important than in previous stages of
the analytic CRM evolution. Since the data warehouse has a direct
impact on the quality of customer interaction, it must always
be up and providing decision-making capabilities for personalizing
the relationship.
Conclusions
Applications for analytic CRM that are built on top of traditional
data warehouse implementations are constrained by the limitations
of the service-level capabilities of the underlying information
repository. However, as a data warehouse matures to allow increasing
sophistication in handling ad hoc and data mining workloads, the
organization's target marketing capabilities benefit accordingly.
The transition from traditional data warehousing to active data
warehousing is particularly important in that it provides more
aggressive service levels in the areas of data freshness, query
performance and availability. These improvements enable more sophisticated
analytic CRM for true one-to-one relationships with event-based
and interactive marketing capabilities.
The stages of analytic CRM deploy-ment described
in this article are not mutually exclusive. A comprehensive marketing
program will involve a combination of mass marketing, segment
marketing, target marketing, event-based marketing and interactive
marketing. For different marketing initiatives (e.g. retention
versus acquisition), a single organization might be in different
stages of the evolution, depending on the level of sophistication
required to maximize return on investment. T
1
Brobst, S. and Rarey, J., "Leveraging an Enterprise Data
Warehouse for Tactical Decision Support," The Data Warehousing
Institute Flashpoint, May 25, 2001
2 Brobst, S. and Rarey, J., "The Five Stages
of an Active Data Warehouse Evolution," Teradata Magazine,
Winter, 2001
3 Brobst, S., "Delivering Extreme
Data Freshness with Active Data Warehousing," Journal of
Data Warehousing, The Data Warehousing Institute, Spring, 2002
 |
Stephen Brobst is chief technology officer for Teradata.
He specializes in very large database implementations for
data warehouse and customer relationship management solutions.
Email him at stephen.brobst@teradata-ncr.com. |
 |
Joe Rarey is a lead consultant
at Strategic Technologies & Systems.
He specializes in high-end systems integration for data
warehouse solutions at Fortune 500 companies. E-mail him
at jrarey@strattech.com. |
ILLUSTRATIONS BY JOYCE HESSELBERTH
|
Quick Links
))
))
