HOW CAN CABLE COMPANIES DELIGHT THEIR
CUSTOMERS?
Many
customers do not love their cable companies. Advanced analytics and
causal modeling can discover why, and help to figure out
cost-effective ways to boost customer satisfaction, loyalty, and revenues. For
Rogers Communications, Cox Associates developed causal models of customer
satisfaction; identified high-impact interventions for improving customer
satisfaction; and helped to develop achievable targets and strategies for improving
customer experiences in different channels (2011). For Comcast Cable, we
delivered a statistical analysis of the causal drivers of customer satisfaction
to top executives and identified realistic targets and interventions for
improving customer satisfaction (2010-11).
HOW TO FORESTALL BAD
DEBT AMONG ENERGY COMPANY CUSTOMERS?
In 2007,
Cox Associates, in partnership with North Highland, delivered a new statistical
risk model for creating timely "red flags" warning when
energy utility customers are likely to become bad debts. This early
warning system successfully identified the highest-risk customers and showed
where earlier interventions -- targeted at the most actionable customers --
could have the greatest benefits in preventing customer accounts from deteriorating
into substantial losses. The model identified some unexpected pockets of
high-risks, based largely on interactions among payment history variables
that could be identified well in advance of traditional warning signs.
WHAT IS THE VALUE OF IMPROVING
WEB CHANNELS?
In 2006,
Cox Associates, in partnership with North Highland, delivered to a large
telecommunications company a computer simulation model and supporting data
analyses that quantified the economic impact of improving the company's web
channels so that more transactions -- ranging from new service order tracking
to repair requests to billing questions -- could be successfully completed
on-line. By identifying the preferences and behaviors of different
customer clusters using a combination of transactions data and survey data, we
quantified the number of customers and that would shift toward the web channel
if it were upgraded to work better. The model also quantified the value
of sales rep and service rep time saved, and studied the economic values from
each other channels, ranging from retail locations and kiosks to call centers.
WHO IS LIKELY TO BECOME BAD DEBT
FOR YELLOW PAGES COMPANIES?
In June,
2003, Cox Associates, in partnership with North Highland, developed a new
statistical risk analysis model for predicting which yellow pages customers are
most likely to become bad debts in the next 1, 2, 3, or 4 quarters, based on
currently available information about their business types (yellow pages
headings), open invoices (number, ages, amounts, etc.), account age, location,
billing history, and other readily available information. The new model
outperforms all previous approaches
and provides
specific recommendations on which customers are at the greatest risk of
defection (i.e., churn), becoming bad debts, or both. This is the first result
from a new initiative to decide what specific interventions with which
customers when will most increase their values as yellow pages customers. Cox
Associates' causal clustering and customer behavior-prediction technology has
already revealed some striking new insights about which types of customers are
at greatest risk, and programs to reduce churn and bad debt based on these
findings are now being designed.
UNDERSTANDING THE REASONS FOR
CHURN AMONG ISP CUSTOMERS
A
continuing goal for many Internet Service Providers is to reduce the rates of
churn (i.e., unplanned loss) of profitable customers. Cox Associates, in
partnership with Adjoined Consulting, has recently begun a data analysis
initiative to understand how the specific reasons for churn affect observed
behaviors (e.g., connect times, number and duration of sessions per month,
number of attempted sessions, called-in service questions and complaints,
etc.). Based on this improved understanding, we are creating new customer churn
risk management algorithms that use changes in observed behaviors to give early
warnings of high-risk customers - those who are most likely to churn in the
next 6 months - and their most likely specific reasons to churn. These
statistical patterns are then used to drive focused, timely interventions to
reduce churn rates among profitable customers.
COX ASSOCIATES MODELS ACHIEVE 40%
CHURN REDUCTION FOR A MAJOR ISP
In June,
2002, Cox Associates in partnership with Adjoined Consulting delivered to a
major Internet Service Provider a set of recommendations on how to reduce churn
of high-value customers using predictive algorithms and prescriptive decision
models. The recommendations identified specific combinations of changes,
including more accurately targeting initial offers, early detection of usage
patterns and data that predict probable churn, and
timely intervention with the customers who have the greatest predicted loss
potential. In July through August, the recommendations were tried. The client's
evaluation was that, within 60 days of initial implementation, our recommended
churn reduction program achieved a 40% reduction in churn in the most at-risk
group of customers -- a large percentage of the total customer base. Based on this highly successful outcome, the modeling results and
recommendations have been extended and deployed to a wide set of the ISP's
customer base.
WHO INVESTS HOW MUCH IN LIFE
INSURANCE AND RETIREMENT ACCOUNTS, AND WHEN?
In May of
2002, Cox Associates in partnership with Adjoined Consulting completed an
adaptive predictive clustering (APC) system that predicts which customers are
most likely to churn, which are most likely to buy new financial services, and
which are most likely to change their asset investments in existing products.
The system reads in billing records from a major retirement fund, life
insurance, long-term care, and diversified financial services provider and
generates probability scores showing
how likely
each customer is to exhibit each behavior in the next month. It calculates the
expected economic values of different intervention strategies (e.g., focusing
on retention vs. additional sales for each customer) and recommends the most
appropriate action to maximize the expected value for each customer. A key
finding from the model is that the current product portfolio, in conjunction
with a customer's age and a few other factors, provide strong, robust
predictions about probable next behaviors.
EARLY IDENTIFICATION OF BIG
SPENDERS AND LIKELY DEFECTORS AT A CLEC
In
October, 2001, Cox Associates personnel completed a study, in partnership with
another company, to identify ways to predict which competitive local exchange
carrier (CLEC) customers would experience the most revenue growth in the next
quarter and which would be most likely to drop accounts. At a time of chaos in
the telecommunications world, with a huge number of the CLEC's new customers
being acquired in the past year, it appears that our data mining and modeling
technology is still able to make robustly accurate predictions of the customers
who are most and least likely to show near-term growth.
PREDICTIVE MODELING FOR A LARGE
PROPERTY & CASUALTY INSURANCE COMPANY
In
August, 2001, Cox Associates completed an analysis of insurance customer data
showing that combining information from homeowner, auto, and other insurance
lines using classification trees and state transition models had the potential
to dramatically improve accurate identification of cross-sell, up-sell, and
retention opportunities. These insights and the quantitative models that
support them may speed the implementation of customer scores and predictive
modeling at a Fortune 100 insurance company.
WHO WATCHES WHAT ON CABLE TV?
In April,
2001, Cox Associates personnel finished a study of cable
company billing records and service records. This study revealed that
customers who are most and least likely to add new services (e.g., upgrade from
basic to extended basic service, or upgrade to digital service) and those who
are most likely to churn can be predicted with useful accuracy from data
contained in recent billing records. Prediction rules were created and their
performance was evaluated by the cable company, which concluded that they work
well.
WHAT DO QWEST'S LARGE BUSINESS
CUSTOMERS BUY?
In
December, 2000, Cox Associates completed a study of purchasing patterns among
large business customers for Qwest communications. The results show that a few
key products, together with factors such as account age, predict likely
stability or churn of customers. A different small set of products (such
as Frame Relay or business voice messaging), in conjunction with past purchase
histories, can help to predict likely growth potential. This analysis led
to a predictive model of how customer needs and purchase probabilities evolve
over time. This work is continuing in 2001 with larger data sets and more
detailed analyses.
HOW DOES ADVERTISING AFFECT
CUSTOMER PERCEPTIONS AND INTENTS?
In 1Q and
2Q, 2000, Cox Associates delivered to U S WEST (now Qwest) statistical analyses
of the effects of U S WEST and competitor advertising and publicity (including
brand/service commercials, direct mail, and
news stories) on customer ratings of value and loyalty. Applying advanced
statistical and causal modeling methods, Cox Associates identified the ad
campaigns and customer characteristics that most affect outcomes.
WHAT AFFECTS STATED AND ACTUAL
CUSTOMER LOYALTY?
In
January, 2000, Cox Associates, in continuing partnership with the Marketing
Intelligence Decision Support (MIDS) group in U S WEST Communications,
delivered new analyses and causal models of customer survey data that predict
how customer perceptions of value and quality and stated loyalty to U S WEST
will change as different aspects of installation and repair service are
improved. These models are novel in that they explicitly seek to isolate the
causal impacts of U S WEST initiatives on customer survey results. This contrasts
with earlier models, which focused on statistical associations and predictions,
rather than on causal impacts. It also contrasts with previous models developed
by Cox Associates which have emphasized changes in customer behaviors, rather
than changes in how customers will respond to surveys. Also in January, we
completed an analysis quantitatively linking what customers say they will do on
surveys to what they actually do in the months following the survey.
HOW DO ADVERTISING CAMPAIGNS AFFECT DEMAND FOR LOCAL TOLL
SERVICES?
In
September, 1999, Cox Associates, in partnership with the Marketing Intelligence
Decision Support (MIDS) group in U S WEST Communications, delivered the first
results from a data mining and modeling project that quantifies the impacts of
television advertising on local toll usage and revenues. The findings included
some unexpected effects of competitor campaigns and of advertising for other U
S products on customer usage of U S WEST toll services.
WHICH CUSTOMERS WILL BE MOST
VALUABLE?
In June
of 1999, Cox Associates, working in partnership with Price-Waterhouse-Coopers,
delivered to the CRMS group in U S WEST Communications a new data analysis and
predictive model for identifying the likely future purchasing, product-drop,
and account disconnect behaviors of individual customers. In contrast to
previous models, this one incorporates more detailed information about the
products that a customer initially acquired, the time since last transition
(product purchase or drop), and the history of product adds and drops leading
up to the current product portfolio. Testing and validation indicate that the
new model has significantly greater predictive power than previous ones,
achieving lifts of several hundred percent on the task of predicting which 10%
of customers are most likely to buy specific products in the next few months.
Several product-specific lifts exceed 300% and some exceed 500%. The model is
based in part on an underlying discrete-event simulation of customer
transitions, with conditional transition intensities estimated from millions of
historical data points. It has been used to predict the expected change in
products and monthly revenues from individual customers over the next 5 years.
(Client Contact: Dr. Jovan Barac, 303-965-3732)
The
August 30th, 1999 issue of Telephony mentions how these deliverables are now
being applied, in a cover story featuring U S WEST's successful use of
Corporate Data Warehouses, data mining, and modeling methods (http://www.internettelephony.com/archive/8.30.99/cover/cover.htm). In this article, our
client, Dr. Jovan Barac, speaks of the SAS scoring
models developed and delivered by Cox Associates and the value that these
models have brought to U S WEST's database marketing efforts.
IMPROVING CUSTOMER
PERCEPTIONS AND LOYALTY
In July
of 1999, Cox Associates, in partnership with U S WEST's Marketing Intelligence
and Decision Support (MIDS) group, completed a first version of a causal
simulation model that quantifies the likely impacts of different U S WEST
initiatives on customer perceptions of U S WEST quality and value. The
simulator uses causal models based on survey data to predict the sensitivity of
customer perceptions and stated willingness to drop U S WEST to changes in
pricing, service performance levels, service rep training and actions, and
other factors. (Client Contact: Dr. B.J. Deering,
206-345-7932)
PREDICTING CONSUMER
PURCHASES MORE ACCURATELY
In
January, 1999, Cox Associates delivered to U S WEST Communications a
breakthrough predictive model for identifying which specific product(s) each
Premium customer is most likely to purchase next. The new model, implemented in
SAS, is based on data mining algorithms and causal models, including
classification tree analysis and detailed consumer state transition modeling.
Initial tests indicate that the new Next Optimal Bundle model nearly doubles
the expected increase in sales due to targeted marketing compared to older
approaches such as logistic regression.
ANTICIPATING NETWORK IMPACTS OF
MARKETING DECISIONS
On
December 16th and 17th, 1998, Tony Cox and Warren Kuehner
of Cox Associates, in partnership with the new national Business Planning unit
of AT&T Wireless, gave a two-day workshop on the design of an Integrated
Forecasting, Planning, and Optimization system at AT&T Wireless in Redmond,
Washington. The workshop documented the ways in which marketing, finance, RF
planning, network planning, and switch planning currently work together to
forecast network load growth and capacity expansion requirements. It identified
gaps in the current process, data, and algorithms and recommended both quick
fixes and longer-term improvements. A key theme was that causal forecasting
based on predictive models of customer responses, rather than historical
trending, is essential to anticipate the traffic load effects of new price
plans and other marketing initiatives. Approved clients may click here to download the background materials and final
report from the workshop.
PREDICTING CHURN
In
November, 1998, Cox Associates delivered to U S WEST Communications' Small
Business Group (SBG) a new model for predicting which small business customers
are most likely to drop specific U S WEST products and services. The initial
version of the model was based on data mining algorithms applied to behavioral
data (product add and drop histories) only. In December, the model was refined
by including firmographic and event data. It appears
to offer significant power for identifying customers at relatively high risk of
churn several months or quarters ahead of the event.
WHO BUYS DIGITAL CABLE?
In
December, 1998, Cox Associates delivered initial analyses of market survey data
to TCI identifying the types of telephone customers (based on patterns of
residential and long-distance telephone use) who are most likely to subscribe
to digital cable. Our data mining technology revealed some surprisingly strong
relations between the types of telephone services and features that customers purchase and the types of cable services that they
will subscribe to. These data suggest some new ways to bundle telephone and
cable services to stimulate demand for both. Approved TCI and AT&T clients
may send e-mail to us to obtain copies of these results.
A BETTER WAY TO PREDICT CUSTOMER PURCHASES?
Cox
Associates, in partnership with SystemView, has
recently been experimenting with a novel idea: applying advanced algorithms
developed for speech recognition and signal processing to predict future
purchase behaviors for customers based on their observed purchase histories.
Initial results, reported in October at the Fall INFORMS conference in
ECONOMIC VALUE OF DATA ITEMS
In late
June 1998, Cox Associates delivered to U S WEST Communications a method to
quantify the dollar value of specific data elements, such as information on
subscription to online services, ownership of PCs, age of account, household
demographics, and recent customer purchase histories. The method quantifies the
economic values of these data elements for specific business decisions, such as
deciding whom to target for mass mailings as part of new product promotion
campaigns. Working with data specialists from Computer Sciences Corporation
(CSC), Cox Associates applied the new method to the example of targeting mailings
to stimulate sales of Additional Lines.
METHODS:
Our new technique integrates traditional decision-analytic principles for
calculating the Expected Value of Sample Information with the data mining
technique of classification tree analysis for estimating conditional
probabilities of customer responses, given real (possibly incomplete,
imprecise, or incorrect) data about them.
RESULTS:
Our valuation methods shows that the values of
specific data elements are greatest in improving mailing decisions among
customers who have roughly a 20% probability of purchasing. The dollar value
may exceed $2 per customer for some data elements and some customer groups.
CLIENT
CONTACT: Dr. Jovan Barac, U S WEST Communications,
303-965-3732.