Progress in Decision Modelling

I created this blog in 2009, with a very specific purpose:  to share the initial concepts of Decision Requirements Analysis and get some feedback prior to writing a book on the subject.  It served its purpose;  it forced me to formalize the central ideas, and I had some very useful input from a number of people.  Knowledge Automation was eventually published in 2012.  I now have enormous respect for authors; if I had known how much work was involved in producing even that slim volume I might never have started.  If you are one of the select few who have actually purchased a copy, many thanks.

In Knowledge Automation I introduced a new type of diagram – the Decision Requirements Diagram (DRD) – which models a domain of decision-making by decomposing it into a network of interdependent decisions, supported by areas of business knowledge and data.  I showed that such models can support an efficient methodology for scoping and implementing decision automation projects.  

A simple DRD as proposed in Knowledge Automation

I thought you might like to know where these concepts have gone.

In December last year, the Object Management Group (OMG) published a new standard:  Decision Model and Notation (DMN).  I was part of the team which developed the specification;  I also edited the spec and wrote some parts of it.  DMN adopts the DRD as the graphical notation for the “decision requirements level” of decision modelling, with some minor modifications (including a new Knowledge Source element used to define sources of authority for decisions and business knowledge).

The same DRD in DMN

But DMN is much more than DRDs.  In addition, it provides:

• A “decision logic level” based on a new expression language FEEL, which specifies the logic underlying the decisions and business knowledge models in the DRD
• A standard unambiguous notation for decision tables, as one important form of decision logic
• A semi-graphical "boxed" notation for other forms of decision logic
• A precise execution semantics for both the decision requirements level and the decision logic level of decision models.

As a result, a fully specified decision model in DMN is executable, that is, when evaluated against a set of input data it will return a set of decision results.  DMN is intended to be compatible with BPMN, so it can be used to specify a decision-making activity in a business process, and to automate that decision-making. 

I enjoyed being involved in DMN and have learnt that standards are important:  they create new markets, they enable progress.  A number of tool vendors, including my own company FICO, are now developing products for editing and/or executing DMN decision models.  For the last ten years I have been drawing DRDs on whiteboards, recording the models in Visio and Word, collecting the decision logic in Excel, and handing the results to a development team for implementation in Blaze Advisor.  Very soon I will be able to create a decision model in a single tool, validate it as complete and correct, and with the click of a button create a functional decision service.  And all this in the cloud.

A brief overview of DMN is provided here in an extract from a chapter I co-wrote with other contributors to DMN in Intelligent BPM Systems:  Impact and Opportunity.