Mike Ferguson pointed me to this article he wrote on Building Intelligent Agents using Business Activity Monitoring for DM Review and this made me think about EDM and BAM. Firstly let me say that this is a well written article, a nice summary of BAM and how it builds on and adds value to both a business process (BPMS) infrastructure and a business intelligence (BI) one. The essence of BAM, as Mike points out, is to "continually and automatically monitor events in ... operational business processes". He also notes that a failure to link BI systems to everyday operations has resulted in BI becoming associated with off line/passive analysis not operational improvement.
Much of what I write about EDM on this blog is about attacking these two problems - making BI relevant to operational systems by embedding insight into those systems and making processes and systems respond more automatically. Mike highlights both key EDM technologies in his proposed solutions - predictive analytics/data mining to deliver insight and predictions of future problems or opportunities and business rules to automate action taking. In the list of required technology, for instance, he talks about being able to "Automatically analyze the selected integrated set of data to product intelligence" and then"rules to automatically whether to take action on the intelligence and, if necessary, actually take those actions". These two steps, taken together, are the heart of EDM.
As Mike drills into the detail he tends to identify automated analysis or data mining separately from the rules engine. While this is somewhat true I believe the separation is more between off-line development of models and rules and in-line execution of same. If you look over in the EDM FAQ there is a post on "How do the basic EDM building blocks work" where the following diagram is used.
If you compare this to Figure 2 in Mike's article there is a clear mapping - the data required to develop rules and models (on the left of my diagram) comes out of the even-driven data integration step while the resulting rules and models are deployed as a service that consumes similar data and makes decisions about what to do. The system(s) shown on the right of my diagram could include the BAM infrastructure.
I do think there is a difference though in that I believe that better models will come out of doing off-line analysis and then running the models in the event stream. This clear separation of development (of rules and/or models) and execution will be more likely to deliver the performance needed. This allows sophisticated data mining and analytic tools to be used to mine data to find the right rules and thresholds (such as customer segmentation rules derived from historical data) and to build new predictive models that essentially enhance the available data by adding predicted characteristics of customers, orders, partners etc. The resulting rules and models can be combined with judgmental rules and regulatory requirements and executed as decision services using the rules engine.
One additional attraction of deploying decision services is that they can be used by BAM infrastructure, by BPMS infrastructure or by legacy systems. Thus I might reuse a service that makes a decision in the standard process and in the BAM-based event processing. Even if I don't reuse the whole service I can reuse rules and models between decision services by taking advantage of rules management capabilities.
BTW I also wrote something on how EDM and EII work together based on work Sun is doing with Fair Isaac and Ipedo here.