Q&A: Using Decision Management to Maximize Analytics
What is decision management and how is it related to advanced analytics? What kind of decisions can be managed (and thus improved), and how can you identify them?
- By James E. Powell
- June 17, 2014
Enterprises need data to make decisions, but how can you improve those business decisions by making them more accurate? James Taylor , CEO of Decision Management Solutions. Mr. Taylor is conducting a session at the TDWI World Conference in Boston (July 20-25, 2014) entitled Evolving Your Requirements Approach to Advanced Analytics with Decision Management. In this Q&A, we explore what decision management is and what decisions can profit from its approach.
TDWI: What, exactly, is decision management and what does it have to do with advanced analytics?
James Taylor: Decision management is a systematic approach to automating and improving operational business decisions. It increases the accuracy, consistency, and agility of these decisions while reducing decision latency and the cost of decision-making.
Decision management treats operational decision-making as a business issue and considers decisions a corporate asset. Decision management involves discovering and modeling decisions, building and deploying decision-making IT components that combine advanced analytics with business rules, and putting in place monitoring and learning infrastructure to ensure the long-term quality of decision-making.
Decision management relies on and exploits advanced analytics. It maximizes the ROI of investments in advanced analytics by focusing them on the critical, high-volume, operational decisions that drive business results. At the same time, decision management relies on advanced analytics to embed accurate, data-driven assessments of risk, fraud, opportunity, and demand in decision-making.
What's the history of decision management? Where did it come from and does it work?
Decision management has been applied to the development of systems for decades, although the phrase was first used about 11 years ago. Decision management began in credit risk where the need to make individual decisions about credit -- how to price a loan or who to give a credit card to -- required accurate decisions to be made rapidly in large numbers. Taking advantage of some of the earliest production predictive analytics models, credit scores, the approach allowed banks and credit card issuers to profitably innovate in the consumer credit market.
Detecting and preventing fraud soon followed; decision management was used to build systems that did more than just "pay and chase" fraudulent transactions but actually prevented them from entering the banking system at all. Since then, other industries such as insurance and telecommunications, as well as areas such as arketing and operations, have successfully applied the approach.
What kind of decisions are we talking about here?
Decision management is focused on decisions that:
- Are repeatable, made over and over again by an organization
- Are non-trivial, often involving the need to effectively apply data to improve the accuracy of those decisions
- Can be automated, where all or some of the decision-making can be handled by a computer
- Are impactful -- they make a difference to the organization's critical metrics and KPIs
It is useful to break down the decisions an organization makes into one-off strategic decisions, somewhat repeatable tactical decisions related to management and control, and highly repeatable operational decisions made at the front line of the organization -- decisions about a single customer or transaction. Decision management focuses especially on these operational decisions, though it has strong use cases also for repeatable tactical decisions.
These are not the decisions people usually focus on with analytics, so how can you identify suitable decisions?
Many organizations begin by trying to apply analytics to strategic decisions, yet this is fraught with challenges: executives have considerable experience and so resist analytics, strategic decisions by their nature don't generate patterns of prior success that can be analyzed, and operational decisions can seem too localized to make a difference. Yet applying advanced analytics to these decisions can have a disproportionate effect thanks to the number of times the decision is made -- the repeatability of these decisions acts as a multiplier for the value of the analytics.
Identifying these decisions is a critical element in succeeding. Generally, several techniques are applied in combination:
- Business processes are analyzed for worklists and escalations that bring human decision-makers into the process
- KPIs and metrics are analyzed to see what decisions are being influenced by those metrics and determine who is being motivated to decide differently because that metric is observed
- Business intelligence assets often trigger decision-making or are used to support it and so can be a rich source of decisions
- Brainstorming and the search for analytics opportunity can identify existing and potential decisions
Why do you need a new way to describe the analytics requirements of these decisions?
Existing requirements-gathering techniques largely ignore decision-making. At best, the need for a decision might be identified in a use case or requirements document, but once it is identified, the focus shifts to data design or UI design and the "design" of the decision is neglected. Effective decision design and effective decision requirements need four elements:
- Information design to show what information is needed in each piece of the decision-making
- Knowledge design to show what policies, regulations, experience, best practices, and analytics insight should be applied to make an appropriate, accurate, profitable decision
- Precision in that the decision is broken down --decomposed -- to provide a precise description of how the decision should be made each time; the information and knowledge required for each piece can be identified as the decision is decomposed.
- The context of the decision in terms of the business processes, business events, information systems, organizational units, and roles involved
Decision modeling brings these four elements together to accurately specify decision requirements.
How do these decision models help you develop and deploy advanced analytics?
Decision models can help with any kind of analytics by making it clear what decision-making is being supported by the analytic. When it comes to advanced analytics, it is particularly important to understand the decision-making the analytics is being designed for:
- The decision model ensures that the role of the advanced analytics is known before it is developed, ensuring that the right model is developed -- one that will be usable in a business decision-making context
- Potential consumers of the advanced analytics can see exactly where in the decision-making the analytics is required ensuring the analytics will be used to actually improve the decision
- The other elements of the decision-making (the policies and regulations that constrain it, for instance) are clearly modeled so that rules and policies can be effectively integrated with the analytics
- The implementation and deployment context of the analytics is known in advance, speeding time to value by ensuring that potential challenges in deployment can be planned for
- Decision models are a great tool for bringing together business, IT, and analytics teams so that there is a shared understanding of what the analytics are for and how they are going to help