I Know What You Bought Last Summer:
Predicting Customer Behavior
BY JEROME NADEL, SLP INFOWARE
Customer relationship management (CRM) has become a fundamental
operating requirement for any service provider or merchant. CRM is about
maximizing customer relationships: addressing the needs and opportunities
associated with each customer, increasing revenue per customer, customer
loyalty and overall customer base.
While CRM is essential to profitability, its benefit is limited without
the ability to predict customer behavior before it occurs. To illustrate,
let's look at a few scenarios that typify the kind of lost revenue
opportunities new economy businesses face every day.
Ed is a high-tech road warrior who doesn't like his cell phone service.
Making calls home to his family while he's on the road is too expensive,
and the phones that are compatible with his service don't have the
features he needs. Not surprisingly, he's become increasingly attentive to
offers from other cell phone service providers.
Molly is a reader and a music lover. She bought books and CDs on the
Internet three times last year, each time from a well-known e-commerce
site. Over the same period, she's visited the site a dozen times and
abandoned several half-full shopping carts.
Philippe loves gadgets and has a new WAP-enabled phone that can surf
the Web. It has a sharp, 2 x 2-inch screen and a tiny stylus you can use
to touch hyperlinks. But after a few attempts to look up the information
he needed while traveling and checking his e-mail, he mainly uses it as a
phone.
For the cellular service provider, the e-commerce site and the wireless
Internet company, Ed, Molly and Philippe are lost opportunities. Ed is a
high-value customer about to churn, or jump to a competing provider. Molly
would be happy to buy all her books and CDs on the Web, if only she could
find what she wants. Philippe is a classic early adopter. If the wireless
Internet startup can't convince him to use their new browser, good luck
convincing the rest of us.
Predicting customer behavior is the key to maximizing the value of the
customer relationship in each of these situations -- maximizing value both
to the company, in terms of revenue and profitability, and to the
customer, in terms of convenience and satisfaction.
At the core of any CRM system is customer information -- collecting it,
managing it, distributing it. As CRM systems become ubiquitous, they both
create an opportunity and highlight a problem. How can companies that want
to understand and effectively meet the needs of their customers put all
these new data to work? The volume of CRM-generated data overwhelms
traditional methods of analysis. As a result, the intrinsic value of the
data is largely ignored.
Potentially, that information has value throughout the enterprise. In
sales, it helps companies make the offers that are most likely to be
accepted and focus sales efforts on the highest lifetime-value customers.
In after-market service, it reduces customer hassles, lowers costs and
streamlines repair and return processes. In marketing, it is a key to
planning, executing and evaluating campaigns. In manufacturing and
distribution, it helps forecast optimal inventory. In design, it guides
development of product features and styles. In finance, it helps manage
credit risks and opportunities.
In reality, though, only a fraction of these multiple values is
realized. The two methods companies traditionally use to make sense of
customer information -- rules-based systems and statistical analysis --
are both proving inadequate.
Rules-based systems are usually either too crude to model real customer
behavior or too difficult to implement. Most often, experienced marketing
experts use them informally to plan campaigns based on rules they have
derived from previous campaigns. In today's environment, this approach is
too slow: it can't catch emerging behavior patterns and, with its reliance
on skilled marketing people, it doesn't scale. Automated rules-based
systems are similarly inadequate. The complex, shifting set of
relationships between a company and its customers is more like an ecology
than a simple mechanism.
Statistical analysis is not crude and can identify new behavior
patterns, but even using the most sophisticated statistical software, the
process is iterative, time-consuming and requires intervention by experts.
It takes three to four weeks to build a statistical model of a particular
customer segment's behavior toward a particular product. Because customer
behavior tends to shift over a three-month period, the value of that
expensive model building effort is slashed by a third.
What if you have customers who fall into 15 meaningful segments? What
if you have 15 different services or products you want to sell to each
segment? What if you have 15 different questions you want to ask for each
product/customer relationship, such as questions about pricing,
cross-selling, upselling or credit risk?
Neither rules-based approaches nor statistical analysis tools can help
companies maximize the value of their CRM investments. Companies need an
automated analysis to complement their CRM data tools. They need an
industrial-strength, fully integrated prediction factory.
Enter predictive customer relationship management. This is the art and
science of putting customer data to work. Predictive CRM holds the promise
of enabling companies of any size and scope to treat each customer as he
or she wants to be treated. It is the capstone of the customer focus
revolution.
For predictive CRM to meet this promise, system providers must offer a
compelling business case that goes beyond effective data analysis. The
finish line is a system that enables companies to meet the needs of
customers like Ed, Molly and Philippe. Getting to the finish line requires
eight key components:
Seamless data uptake. Predictive CRM systems must integrate with
the data acquisition APIs of the company's CRM system so that customer
data can be analyzed rapidly for real-time applications such as call
center screen pops, as well as for offline applications such as modeling
marketing campaigns.
Effective modeling. The algorithms used to identify and model
patterns in customer behavior must be highly accurate and must begin to
show customer-level results based on relatively few data points. This
requirement is particularly pressing for e-commerce and other
product-based applications, which tend to generate less customer data than
service-based applications such as ISPs and wireless service providers.
Rapid modeling. Offline model building must be fast, so users
can implement campaigns quickly and build multiple models to cover the
wide variety of company/customer/product situations.
Simple modeling. Modeling must be simple and intuitive enough so
that stakeholders throughout the enterprise can create and use models on
an ad hoc basis. Systems that require statisticians to operate are no good
to organizations that do not employ statisticians. Even large companies
that have deep statistical resources are better off with systems that
marketers, financial personnel and others can operate for themselves
rather than relying on a centralized resource.
Integrated campaign design and management. The key to
transforming a system that models customer behavior into one that
influences customer behavior is campaign support. Designing behavior
models is the first step. The next step is integrating those models into a
system that helps users choose the appropriate model for every
application. That system should guide users through the whole process of
designing, optimizing, deploying and benchmarking marketing campaigns.
Just-in-time/just-in-place results distribution. The results of
predictive CRM behavior analysis may include tailored offers for customers
presently online or on a call center queue, a credit risk list, on
personalized Web pages served in real-time or product documentation picked
from a database. Distribution platforms may include screen pops at a call
center, automated e-mail, personalized Web pages, automated faxes or IRC
file transfers. For maximum effectiveness, predictive CRM systems must be
integrated throughout the enterprise so whatever the content type, and
whatever the appropriate distribution platform, the customer information
arrives at the right place at the right time.
Efficient process flow. The goal of a predictive CRM system is
to allow stakeholders from anywhere in the company to create appropriate
predictive knowledge and distribute it to the right place at the right
time. Realizing this vision involves not only analysis and modeling, but a
supportive, user-friendly process flow that uses wizards and other
heuristic devices to guide users through the entire prediction loop
process.
Superior ROI. Maximizing customer relationships is a concrete
value with tangible impact on both profitability and revenue. Predictive
CRM systems compete for software budget share in a wide arena that
includes a variety of enterprise applications. However, wherever the
business environment includes high marketing and customer acquisition
costs, predictive CRM systems have an advantage. Predictive CRM systems
are extraordinarily attractive on an ROI basis because they can enable
companies to dramatically reduce customer churn, target their marketing
campaigns more effectively and tailor individual product offers so they
are more likely to be accepted.
As the Internet continues to create paradigm shifts, new markets mature
and e-commerce migrates toward mobile computing, saturation will
increasingly become an issue. Marketers can no longer be concerned merely
with acquiring new customers, but must also focus their efforts on
establishing meaningful, long-term relationships to maximize customer
value and loyalty. This means it is imperative for operational CRM to
include the complexities of customer behaviors and be able to use
predictive technologies to address them in a proactive manner.
Jerome Nadel is responsible for the strategic positioning and
marketing communications for SLP
InfoWare, a provider of customer relationship management software
solutions for the telecommunications, e-commerce and mobile e-commerce
industries.
[ Return
To The September 2000 Table Of Contents ]
|