Decision Engineering is the task of translating, capturing, and replicating people’s expertise and decision-making skills into low code/no code software representations. Enabling this to happen requires a “Decision Intelligence” platform that can support a variety of decision representations depending on the nature of the expertise being automated.
Today, most decisioning tools only support the decision tables format supported by the industry standard Decision Model and Notation (DMN). From our perspective, this approach is limited and won’t scale. True decision engineering needs to incorporate decision tables, decision flows, decision trees, rules, and fuzzy logic. This will allow the capture of the widest range of expertise including assessment, classification, diagnostic, troubleshooting, planning, selection etc. Otherwise, automation of more complex high value decisioning workflows will always remain out of reach.
There are many cases where the Subject Matter Expert (SME) may find it difficult to express their expertise directly as tables, trees, or rules. In such cases, the use of Machine Learning (ML) by rule induction from decision examples can work. These examples are elicited from experts, either as truth tables or via an iterative process of the SME volunteering a small number of examples, generating generalized trees by rule induction, reviewing trees and adding counter examples. The process is repeated until the SME is fully satisfied with the Induced knowledge. This hybrid low code/rules induction methodology is very powerful and can capture complex decision knowledge involving a large number of attributes in just a matter of hours, a task that would normally take many days.
Decision knowledge can also be derived from historic operational data. Data can be historic records of decision making by employees, in which cases ML can be used to reverse engineer the decisions made by people. Alternatively, the data can be a record of business events or performance KPIs in which case ML will generating new knowledge/predictions about the root causes of event and performance KPIs which greatly enhance the understanding these processes.
Where a decision needs to be made based on balancing multiple process performance KPIs to achieve an optimal outcome, an optimization engine is required. From our experience, an evolutionary algorithm is ideal for this type of optimal decisioning, and we have successfully applied this approach in many sectors. One such application is the optimization of manufacturing processes to achieve an optimal balance between efficiency, energy use and product quality.
It is also possible to create Decision Intelligence Frameworks for semi-vertical applications like Risk Assessment, Document Automation, Configure-Price-Quote, Technical Support etc. Such preconfigured frameworks can significantly speed up a no-code decision engineering task.
All these options highlight how we are at a crucial inflection point in how intelligent automation capabilities get delivered and consumed. At the heart of the matter is Decision Engineering. It is the catalyst that drives everything else. The ability to represent, capture, and replicate employee expertise via a low code/no code software is a game changer. Over the next few weeks, I’ll be writing a few more blog posts and sharing where I think this is all going.