An analytical master data management (MDM) initiative can improve the quality of business information being used for reporting and analysis, and it can also be a good way to get started on the journey to full-fledged enterprise MDM, according to analysts and experienced technology professionals.
But they also have a warning for anyone taking the analytical MDM route: Be sure to architect your MDM systems and processes with long-term objectives in mind. Failure to do so can result in costly problems and delays when plans to expand the MDM program are put into action.
"You need to have a master plan," said Al Moreno, an IT solutions architect with San Francisco-based TopDown Consulting Inc. "You can start with a pilot program and institute [MDM] in various pieces of your enterprise, but you have to have an overall plan."
The master plan should answer several questions, Moreno said, including: How will data governance be handled as the MDM project grows? What tools will be used to enforce metadata and business rules? How will master data be defined across the enterprise?
"Master data management is a process -- and it's not just one process, it's a bundle of processes," Moreno said. "You have to have an overall theory or model that you're following."
The differences between operational and analytical MDM
MDM is a methodology that typically combines middleware tools with data governance and data quality processes in an effort to ensure that data in key business “domains” -- for example, information about customers, products and suppliers -- is accurate, consistent and synchronized throughout an organization’s systems.
Ideally, the master data becomes a "single source of truth" for approved data definitions and classifications. That's important, according to analysts and existing MDM users, because different systems often define data in different ways: "Customer XYZ" in one system might be called "Customer 123" in another. The MDM process makes it clear that those two designations represent the same entity and ensures that updates to information about the customer in one system are transmitted to other systems across the enterprise.
MDM typically comes in two flavors: operational and analytical. Operational MDM focuses on defining, distributing and synchronizing master data that supports transactional operations. Analytical MDM centers on managing master data items used for data aggregation and business intelligence (BI) reporting and analysis, and it typically is limited to the information stored in data warehouses.
"Analytical MDM is often an entry point to MDM," said Rob Karel, a data management analyst at Forrester Research Inc. in Cambridge, Mass. "It's the least invasive way to start an MDM initiative because you're not impacting operational processes."
Despite the differences between analytical and operational MDM, the same technology and processes can be used to support both capabilities, Karel added.
You can buy an MDM hub developed to support operational MDM “and turn it on in phase one as a one-directional system, where the only target it feeds is the [data] warehouse," he said. Later, if your organization decides to move beyond analytical MDM and create a true enterprise master data management process, you can flip the bi-directional switch in the hub and start feeding data back and forth between your operational systems “using the same exact rules," according to Karel.
Making the transition from analytical MDM to operational MDM
Organizations that begin their MDM plans by focusing on the analytical side of the equation may be able to reap an additional benefit in the form of improved reference data, potentially helping to pave the way for operational MDM as well, said John O. Biderman, an information architect at Boston-based medical insurer Harvard Pilgrim Health Care Inc.
Reference data is used to define the characteristics of a company’s data and the relationships between different types of information. At Harvard Pilgrim, Biderman has been one of the leaders of a major data warehousing, MDM and master reference data project that initially involved the insurer’s analytical data.
"Our legacy platform had over 20 different dependent [data] types," Biderman said. "We collapsed all those into seven values that are used for the enterprise, and those are now the ones [that are in the] enterprise data warehouse.” He added that Harvard Pilgrim now plans to use the same seven data values at the operational layer as it expands the MDM deployment into that realm.
"We had to prepare a warehouse that could accept data from multiple systems which support the analytical world through a transition to a new operational platform," Biderman said. "Having done that, we laid a lot of groundwork for master data approaches that we need to take in the operational world."