| |
| Billing World & OSS Today: Features |
The Journey from Analytics to
BI
Susana Schwartz - March, 2006 |
| |
There's an old maxim about four blind people in
a room with an elephant. Each person has a different view of
the elephant, depending on whether he or she is holding the
trunk, the tail, the ear or the foot.
After spending
millions of dollars on enormous ERP, CRM and data warehousing
initiatives, executives are left with the same problem, as
monolithic data stores and departments remain in silos-each
with its own view of customer, product, and network data and
processes.
"These silos grew up around specific
channels, like IVRs, Web portals, transactions and business
processes," says Michael Chavez, vice president of marketing
for ClickFox, which works with companies like BellSouth and
Sprint Nextel to derive intelligence from customer behavior
patterns. "All telcos talk about being more customer-centric,
but they continue to use analytics in a channel-specific
manner-with a customer in IVR looking different than a
customer in the Web channel, or transaction or profitability
systems."
That fact has made customer and product
lifecycle management a nightmare, as churn rates are as high
as 30 percent in some telco markets.
For example, a
customer trying to pay a bill online could be prompted to
enter a PIN code that is listed on the paper bill. If that
person doesn't have that bill, he may abandon that session and
call into a call center after finding the toll-free number on
the Web site. To the carrier, it looks like two separate
channels have been engaged by two different people. To the
customer, it is all part of the same session.
As the
customer is filtered into an IVR, there is no recognition of
the original trigger for the drop from the Web site, so the
IVR asks the same customer for the same PIN. Now the customer
is frustrated, and opts out to a CSR, who then again asks for
the PIN for lack of recognition of the channels through which
the customer has flowed. At $6 per call for servicing the
customer, the ability to predict such "triggers" to session
drop-offs would optimize efficiency and lower cost, as well as
keep the customer happy enough that cross-selling and
upselling would be possible using the same predictive
analysis.
Analytics could be the key to identifying
and eradicating such churn triggers, but carriers must first
clean up and centralize or federate data to foster a
360-degree view. After spending fortunes on ERP and CRM
projects, they have the opportunity to make good on
initiatives to attain that comprehensive view.
"Carriers are sitting on a mountain of gold in terms
of the enormous amounts of customer data that remain
underutilized," says Scott T. Toborg, president of Toborg
Technology Strategies. As a 20-year veteran in telecom who has
been CIO and CTO in charge of data warehousing and data mining
initiatives for various telecoms, Toborg is working on the
cultural struggle as much as the technological one. "Everyone
at the high levels knows they need to integrate silos across
customer care, networks, point-of-sale, marketing and other
departments," he says.
What's needed is dynamic
cross-channel analytics, now possible with advances in
hardware and BI applications. Because systems can now process
much more granular data, it is expected that analytics will
evolve from reactive to proactive.
"The technologies
that are enabling fraud and revenue assurance to become
real-time-like credit card swipes for immediate
authorizations-will enable profiling technology with embedded
behavior algorithms to make assessments on every event in an
organization," says Alison Sullivan, director of product
management for telecommunications fraud solutions at Fair
Isaac & Co.
"As the event is processed," she says,
"advanced neural network technology will recognize when the
behavior is outside normal parameters-such as unusual
location, duration or time of day. Advanced models will then
flag any nuance depicting fraudulent behavior or non-payment
and even suggest what actions to take."
Currently,
sophisticated analytics are prevalent for functions like
dispute management, billing, quality assurance, corporate
credit and fraud reduction.
While churn prediction,
customer retention optimization and win-back prospecting are
still the broadest area of analytics, the returns for
providers can be substantially greater if analytics are turned
into BI for not only revenue assurance, but other applications
within the organization.
"Product management,
regulatory, fraud and risk teams increasingly use common data
and common analytics engines to slice and dice data. As
carriers figure out how to get to disparate sources of data
across the organization's silos, they will expand the common
data and analytics engines to marketing, customer care and
other critical areas," says Greg LeNeveu, vice president of
the Americas for Subex.
Already, the industry is under
increasing pressure to derive intelligence from data in
marketing, customer care and billing. Regardless of the
department, executives need to make more real-time decisions.
Empowering Business Users
Analytics often frustrated business users, as
executive buy-in and queries to specialized IT and financial
groups were time-consuming and complex. Budget approvals and
data feeds to ERP, CRM, OSS and billing systems had to be
established to offer visibility into operations and
transactions.
With the drive toward service-oriented
architecture (SOA), telecom is being pushed away from that
mode of operation, toward developing common business language
that empowers business users to view and even predict trends.
For that reason, the next 18 months will see an
evolution in revenue assurance capabilities, as well as an
extension of revenue assurance-type analytics to marketing,
customer care, operations, SLA management, billing and even
self-service portals. Already, proactive and predictive
dashboards are enabling users to see trends without poring
over complex algorithms, bar charts, tables and graphs.
ERP and CRM companies like IBM, Oracle, SAP and
Microsoft have started to extend into analytics with
integrated suites incorporating components for descriptive and
predictive analytics into business applications.
In
addition to those end-to-end suites, the tried-and-true
specialists in analytics-such as Business Objects, Cognos,
Hyperion, SAS, KXEN, SBSS and Fair Isaac-continue to evolve
their solutions. Currently, BI software continues to
outperform other software segments, making it a top-five
technology project, according to Data Monitor reports.
It's a mixed of solutions, depending on where telcos
might have started and what types of M&A activity they've
gone through. Priorities shift based on management teams and
whether they focus on pure growth or operational efficiency.
As BI tools and packages take advantage of advances in
processing power, there will be more initiatives in segments
of organizations to grasp both the overall goal of the company
as well as their own esoteric goals.
This will happen
through "consumerizing" IT so that business users have the
tools to do complex queries.
As carriers increasingly
migrate from database-driven to event-driven architectures,
intuitive middleware layers and extraction technologies are
emerging to decouple applications from back-end data stores
and expedite decision making.
The growth of data
extract, transformation and load (ETL) processes is evidence
that companies are increasingly extracting data from data
stores and converting it for transmission across EAI and
decision support systems. With ETL, BI tools can access
metadata sitting on top of databases and warehouses so that
business users can make queries using familiar language and
semantics. The fact that ETL is agnostic to formats, data
sources and syntax makes programming and data staging a thing
of the past.
ETL is enabling dynamic access to
consistent information, as well as enrichment of the data
through feedback loops to central repositories or federated
databases via intelligent hierarchies.
As the
technology and process matures, each department will gain a
better understanding of its impact on the company's
overarching data, processes and goals.
There are
different ways to describe that paradise, but Gartner Group
has dubbed it "pervasive BI."
Pervasive and
Predictive BI/Analytics
"Pervasive BI is the
point where analytics enable business users to have a real
impact on diverse work groups-all of which lead, discover,
decide and optimize based on analytics," says Betsy Burton,
distinguished analyst with Gartner Group. "Whether customer
lifecycle management, wholesale partnership management [see
Verizon Business sidebar, below], revenue assurance, new
product development, or network management, valuable data from
each can impact the effectiveness of other facets."
For example, the network can affect the number of
calls coming into the call center, and the level of customer
care can impact the network. Marketing and product development
information can help decide where network build-outs should
occur. The point is that understanding inter-relationships
will enable users to see such things as how the lifecycle of
billing fits into the CRM or sales lifecycles.
Key to
pervasive BI will be the predictive nature of emerging
analytics.
Most of us are familiar with Amazon's
predictive analysis engine. What if the same could be done in
telecom? For example, what if many mobile phone users have
searched for red handsets that only come in black or blue? And
what if the BI tools could tell you that most of those
subscribers were in a certain age group and geography that
also tends to like certain video downloads?
That type
of information could help predict future customer behavior,
such as churn and purchasing habits. Through complex
algorithms, market-based analysis, neural nets, what-if
scenarios and decision trees, predictive analytics finds
associations among not only structured data but unstructured
data, and can find profitable sub-segments. Best of all,
predictive analytics can do so without models or queries based
on known data. In other words, you can find profitable
segments without even knowing what you are looking for.
"How much does the customer buy? How much does the
customer cost me in customer care and reminder calls?" asks
Business Objects' Rani Goel, director of worldwide telecom
marketing. To determine the actual revenue and value of a
churner, profitability data must be integrated with usage,
contact and payment behavior data. "You have to enable that
type of analysis, if business users want to see if customers
buying mid-range products will become more profitable if
high-end products are marketed to them," Goel says. "Or if
they want to measure the impact to accounts receivable if
customers are called five days after they are delinquent,
rather than 15 days after they are delinquent. Perhaps they
want to measure a credit policy change on the company's
ability to collect revenue from customers."
Such
analysis can help a carrier understand that low-cost
customers, such as SMS users, could be more profitable than
higher-end customers who contact call centers multiple times
at $10 each call. "Analytics help you evaluate if a person is
spending $20 outgoing but receiving $200 worth of incoming
calls," says Astrid Bohe, a senior executive in Accenture's
communications practice.
According to IDC, business
tools and packaged applications using predictive analytics
software generate an average ROI of 145 percent. That type of
return cannot be ignored, which is why IDC projected that such
applications would see a compounded annual growth rate of 8
percent from 2004 to 2008.
"While today's BI tools are
well-suited for reporting on reams of charts and tables,
carriers have to seek out those that actually tell business
users what is actually happening now and what will most likely
happen in the future. You want to report, analyze, predict and
respond," says Toborg. That means carriers should seek out BI
tools that possess a level of sophistication for not only
predicting, but taking action on those predictions.
"You can verify data coming off switches and identify
all sorts of revenue leakage, fraud and operational
inefficiencies," says Toborg, "but going the next step of
integrating that information with existing data warehouses,
customer care and billing information is the key."
But
getting information from the network to multifarious systems
is complex. "The approach differs according to the company's
focus. On the network side, you might want to detect
anomalies, so you don't need integration with other data for
that," says Toborg. "But if you want to understand why
customers are churning on the network, or if you want to
identify upsell and cross-sell opportunities, then you want to
integrate data from the call center to see how many times a
customer called in, and then into product data to see how much
the person is spending and on what." After that process is
complete, robust predictive models can be built to identify
the customers likely to churn, and what kind of products,
customer care or network experience carriers can market to
them to retain them or win them back.
For carriers to
reach the type of maturity that enables them to mine data to
solve focused business problems, they must resolve data
integrity issues, and continue to build their data warehouses
or repositories for access as federated databases. For the
most part, carriers have been in the throes of that for about
10 years.
BI Is As Good As Your Data
With BI, carriers must get a handle on which analytics
require aggregate data, and which ones require atomic data.
"The more granular the data, the more processing power
and time you need," cautions Accenture's Bohe. "If you want to
do analysis in a big city where congestion is a problem, you
may want to incorporate network data and data from switches to
boost accuracy so you can predict churn based on unsuccessful
call rate patterns." Bohe says he sees an 8 percent increase
in accuracy when traditional CDR data is incorporated with
network data from switches.
Many traditional core
analytic tools can handle aggregate data when analyzing things
like usage, rate traffic, billing traffic, discounts applied
and invoice details. Such data comprises individual data
items, data groups, arrays, and tables that can be assembled
to form a whole. But in averaging data, generalizations mean
the individual is lost.
"With aggregates, you tend to
lose the relationships among entities such as customer and
product," notes Dan Higgins, director of sales support for
Teradata.
Since customer data is the key to telecom,
carriers are focusing more and more on atomic data and
integration in certain areas where they can support multiple
applications over mixed workloads-such as queries, analysis
and applications-on a single platform.
That means
carriers should seek out solutions that support a variety of
business questions without being constrained by their
platform.
Higgins favors a "share-nothing"
architecture, where parallel processing of data subsets occurs
without sharing memory, disk or computing resources. That
means coordination among processes is not necessary, thus
facilitating true linear scalability without associated
overhead.
The high levels of the organization need to
understand the value of integrating network data as well as
traditional customer and billing data coming off switches,
cell towers and base stations. "Ultimately, every cell phone
can become a performance monitoring point for really
understanding what customers experience and identifying
problems before revenue leakage occurs," says Toborg. "Over
the past few years, more and more telecom companies have found
prodigious amounts of corrupted or dropped CDRs, but there is
so much more that can be done."
He notes that
anomalies at the cell tower level are missed in aggregates
sent to switches and on to mediation. "Those anomalies get
washed away in a flood of data, because there is no predictive
pattern recognition to stop a problem at its source," he says.
Billing and transaction systems are not designed for
predictive analytics, especially if that means pulling info
from multiple systems. If a carrier wants to evaluate the top
100 customers in a certain line of business, they might want
CRM, billing, product, networking and call center data.
For large entities, such a query could put a strain on
processing power.
BI Is As Good As Your
Processing Power
In the United States, the
combined Cingular and Verizon entity will add about 30
customers every minute, and the Sprint-Nextel entity about 40.
As ILECs and mobile companies look to vastly increase data
volumes in data centers for combined triple and quad play
services, architectures will crumble under advanced BI
queries.
The architecture that sufficed for online
transaction processing applications will no longer do it for
real-time, dynamic queries across a large organization. "In
the past, you were only moving single records at a given time,
so you could purchase a large storage unit and connect up to
one or several servers through networking in the data center,
which connected to database management software," says Phil
Francisco, director of product marketing for Netezza. "But
now, with deep querying, you have to move terabytes of data
through the data center networks. That gets very
time-consuming and expensive." He claims Netezza's largest
configuration can process and query at a rate of 160 terabytes
per hour.
The company employs asymmetric massively
parallel processing, where hundreds of processors work with
hundreds of disk drives in parallel.
"Rather than
create aggregates of data, you can push the processing down to
where the disks are, so you don't have to transfer data from
disks to server farms," explains Francisco. "Then you query at
the speed you read the disks. You also save overhead from the
database administration level, because of the simplicity of
the approach."
Teradata's Higgins believes that
carriers should assess the fault tolerance and workload
management functions that are built into systems. "You don't
want to have to deal with the underlying complexity of a
platform; that can become a ball and chain as you grow, as you
end up spending your time trying to make the technology work
instead of getting to the actual business analysis."
Companies should study the underlying database engine
and its ability to support complex data schemas, as there are
many different ways to do analysis and queries. "You want to
avoid dumbing down your data model so you can do the queries,"
Higgins says.
The Bottom Line
While BI can offer a significant competitive advantage
to service providers, using unreliable data and insufficient
processing power could have the opposite effect. To win the
battle for the customer, carriers must focus on enforcing
practices that generate quality data.
That means BI
initiatives must start at the top of an organization and get
buy-in across the enterprise-regardless of egos and cultural
differences.
Embedding BI in company decisions
requires a clear definition of the company's focus, goals,
workflows and hierarchies. Only then can a foundation exist
for building out BI to the business users in the trenches.
|
| | |
|
|
|
