DMN Tool Update

Since my last post on the current state of the DMN tools market, I've learned a bit more from a few vendors...

Sapiens Decision

The Decision Model, on which the Sapiens offering is based, can be viewed as a progenitor of DMN, but I was unaware of Sapiens' intent to support the DMN standard in their offerings. That is significant. Where most DMN tools start with barebones decision modeling, Sapiens enters the market with all the bells and whistles needed to support production use today in decision-intensive industries such as global banks: business glossary, intensive role-based governance and audit trail, and of course the "principles and guardrails" that are the foundation of TDM methodology. At present, I believe, the DMN support is limited to a toggle between a TDM decision model view and a DMN-compliant DRD for what we would call a decision service. What I learned, however, is Sapiens is working on a standalone DMN modeler, possibly supporting FEEL and boxed expressions. That would be a real stimulus for DMN as a standard! DMN models created in this tool can be transformed into TDM views to make use of the governance and deployment features of the Sapiens Decision suite.


Trisotech's DMN editor now includes full S-FEEL support, including an expression editor. Work on full FEEL support, including validation and execution, is underway. Also, on a side track, Trisotech is assisting me personally by integrating my DMN validation/execution code based on XSLT with their tools and hosting it in the Trisotech cloud. I've got the Lending decision example from the spec working now; next steps are model validation, then the UServ and Mortgage Recommender examples from my book.

DMN Reference Implementation

At bpmNEXT, Keith Swenson of Fujitsu and WfMC proposed the idea of an open source reference implementation for DMN, and led by Keith, a small team has now begun weekly meetings. The goal is running code, probably Java, that can accept a DMN model in XML, plus input data instance values, and generate the correct output. Not necessarily as fast as a production engine might be able to do it, but enough for design-time testing and analysis. We'll see how it goes. Full FEEL implementation is not easy. If successful, the code should be available to tool providers.