| MCI uses an IBM SP2 massively parrallel system.
The hw/sw choice was made several years ago, and it seems that MCI
suffered a lot to build a corporate DW from many different sources.
However, they were the first Telco to build a DW of that scale.
I can put here an article recently published about the MCI's DW.
It is particularly interesting to note that they consider the need to
have a corporate warehouse AND Data Marts (We Digital stress this need to
in Telecoms), and also emphasize the importance of Metadata Mgt.
Good reading
Pierre-Yves Visciglio (Telecom Expertise Center - Data Warehousing)
Source:
Communications Week
Communications Week via Individual Inc. : Looking to feed the benefits
of datawarehousing trend analyses to its online lead generation systems,
telecommunications giant MCI is one of the first to add an Operational
Data Store to its data warehousing plan.
Adding an ODS to a data warehouse is a sign that data warehousing has
now entered into a second generation, where trend information from off-line
repositories can be fed back into operational systems to enhance what a
salesman or customer service rep can offer the customer. MCI outlined
the evolution and business case for its ODS at the Data Warehousing
Institute conference late last month.
The company has maintained a 2-terabyte data warehouse for three years.
Warehouse MCI has been crucial toward helping company officials maintain
data repositories, resulting in more knowledgeable representatives and
better customer service, according to Chip Grim, IT director for MCI.
"Customers speak volumes in non-verbal ways," Grim said. "They do it by
their patterns of repeat purchases, the frequency and manner with which they
use a company's products, and by their propensity to switch vendors when
they're offered better prices or more targeted features." The change of
only a few percentage points of customer retention can equate to a matter
of hundreds of millions of dollars to the company, Grim said.
With that in mind, the identification of trends that cause customer
churn can help the carrier quickly repackage services and address customer
support, according to Grim. Because the data warehouse is the only place
in the corporation where information from disparate sources comes together,
it is the only repository from which such trends can be plotted.
MCI calls the usage of the results of the data warehouse analysis
"relationship marketing." To speed up the analysis of these relationships,
the data warehousing team has created 24 data marts, or specialized
subsets of the data warehouse, for the MCI marketing department alone.
Data marts emerged with an aim toward improving system and network
performance.
"Everyone thinks that they need direct access to all the data in the
warehouse.
They simply don't understand how much data is there," said Ryan Sousa,
lead architect on the project.
Unlike the data warehouse, which contains historical, non-volatile
information, the ODS contains current, volatile information. The amounts
listed in accounts payable, for example, may be different in a morning
query than in an afternoon query.
The ODS draws its data from the various operational systems in the
corporation, but it also can add information derived from data keys in a
data mart created by a data warehouse. As such, it can not only add the
value of consolidating a common view of enterprise data, but it can also
add data derived from trend analyses done on the data marts.
Getting Information to Where It's Needed
Unlike the data warehouse, the ODS can feed its information back into
the corporation's Online Transaction Processing systems, so that such
things as bundling deals and cross-selling can be done on the spot by
corporate sales and customer service personnel.
"Our lead generation mechanism needed the actual names and phone
numbers to match the demographics that we were analyzing, and the ODS was
the way to provide that," said Sousa.
Validation of MCI's role in pioneering data warehousing came from a
competitor, a staff engineer at a Baby Bell who asked not to be named.
"MCI is leading the pack," he said. "We may be ahead of them on the Web
side of things, but they are definitely ahead of us on the harvesting and
validation of data," he said.
The biggest problem that the MCI team saw was a constant struggle with
the management of metadata, the data about the data in the warehouse. MCI
has its own master metadata scheme, designed to integrate the definitions
of data across its own operational systems. "What we really need is for the
tools vendors to let go of the metadata and just give us robust metadata
import/export capabilities," said Paula Thornton, information architect
on the WarehouseMCI project.
Grim added that accessing metadata was the key to having his corporate
clients understand the power and usage of the data warehouse. "When we
first put the data warehouse out there, people had tons of questions about
the data.
There was visibility but no legibility, so we had to get the metadata
rushed out there," he said.
Thornton, who spearheads MCI's data mining efforts, said that the most
efficient way to do it is to use a "sandwich paradigm" consisting of
pre-luminary data mining, then data warehousing, then actual data mining.
"We call it a data surveying mechanism. It is very much like real mining,
where you figure out where the rich data veins should be before you do the
actual mining," she said.
|