In a normal master data management (MDM) project, a current state business process flow is built, followed by a future state business process flow that incorporates master data management. The current state is usually ugly as it has been built piecemeal over time and represents something so onerous that the company is finally willing to do something about it and inject master data management into the process. Many obvious improvements to process come out of this exercise and the future state is usually quite streamlined, which is one of the benefits of MDM.
I present today that these future state processes are seldom as optimized as they could be.
Consider the following snippet, supposedly part of an optimized future state.
This leaves in the process four people to manually look at the product, do their (unspecified) thing and (hopefully) pass it along, but possibly send it backwards to an upstream participant based on nothing evident in particular.
The challenge for MDM is to optimize the flow. I suggest that many of the “approval jails” in business process workflow are ripe for reengineering. What criteria is used? It’s probably based on data that will now be in MDM. If training data for machine learning (ML) is available, not only can we recreate past decisions to automate future decisions, we can look at the results of those decisions and take past outcomes and actually create decisions in the process that should have been made and actually do them, speeding up the flow and improving the quality by an order of magnitude.
This concept of thinking ahead and automating decisions extends to other kinds of steps in a business flow that involve data entry, including survivorship determination. As with acceptance & rejection, data entry is also highly predictable, whether it is a selection from a drop-down or free-form entry. Again, with training data and backtesting, probable contributions at that step can be manifested and either automatically entered or provided as default for approval. The latter approach can be used while growing a comfort level.
Manual, human-scale processes, are ripe for the picking and it’s really a dereliction of duty to “do” MDM without significantly streamlining processes, much of which is done by eliminating the manual. As data volumes mount, it is often the only way to not watch process time increase over time. At the least, prioritizing stewardship activities or routing activities to specific stewards based on an ML interpretation of past results (quality, quantity) is required. This approach is paramount to having timely, data-infused processes.
As a modular and scalable trusted analytics foundational element, the IBM Unified Governance & Integration platform incorporates advanced machine learning capabilities into MDM processes, simplifying the user experience and adding cognitive capabilities.
Machine learning can also discover master data by looking at actual usage patterns. ML can source, suggest or utilize external data that would aid in the goal of business processes. Another important part of MDM is data quality (DQ). ML’s ability to recommend and/or apply DQ to data, in or out of MDM, is coming on strong. Name-identity reconciliation is a specific example but generally ML can look downstream of processes to see the chaos created by data lacking full DQ and start applying the rules to the data upstream.
IBM InfoSphere Master Data Management utilizes machine learning to speed the data discovery, mapping, quality and import processes.
In the last post, I postulated that blockchain would impact MDM tremendously. In this post, it’s machine learning affecting MDM. (Don’t get me started on graph technology). Welcome to the new center of the data universe. MDM is about to undergo a revolution. Products will look much different in 5 years. Make sure your vendor is committed to the MDM journey with machine learning.