One of the sessions I am most looking forward to is next – Ryan Schmiedl and SAS’ view of Decision Management. Ryan kicks it off with examples of operational decision making – credit card approvals, car loan approval, claims payment. Decisions about transactions made in real-time or near real-time. These systems, Decision Management Systems, embed SAS analytics into operational systems to drive analytical, transactional decisions. This focus is growing as more of SAS’ customers push analytics from their back office into agile, analytic and adaptive operational systems.
Historically there is a high degree of dysfunction resulting from silos as you move from from discovery to modeling, deployment and modeling. Instead need an integrated lifecycle and a common platform as well as strong collaboration between business, IT and analytic teams. And this needs to have a continuous process focus, bringing continuous improvement to the decisions being made, and efficient. Customers need to avoid what can be 9 months to go from start to finish thanks to re-coding models, dealing with hand offs, IT issues etc.
SAS therefore is building out a Decision Management Platform. A consistent user experience and management experience for analytics, IT and business people across preparing data, modeling, optimization, decision flow, monitoring and case management. All of this running on the SAS LASR analytic server across deployment environments like batch, in-memory and streaming.
Ryan interviewed an early adopter customer (under NDA) and then walked through a demo of the forthcoming Decision Management capabilities from SAS. His scenario was how to identify that a second hand car is a good deal for a dealer or not. Wayne Thompson came on to help with the demo and began by showing Visual Data Builder to pull together the data needed for a decision. After building a visual query he writes out the historical data to an analytic server. As in the marketing demo he uses Visual Analytics to explore the data. This data includes a flag for those cars that were a bad deal (kickers) so he can investigate what kinds of attributes correlate with the kicker flag. Enterprise Miner (review here) or Rapid Predictive Modeler (review here) could be used on the same data to build predictive models.
Once the models are available they need to be brought together to make a decision and that uses the new SAS Decision Manager product. Here models built in SAS or models brought in through PMML can be integrated with other elements. Each model has a signature showing the attributes it needs so the data it will need to score a transaction is clear. Mike Ames came up to demonstrate a decision flow is then laid out in Decision Manager to integrate the models that are available with business rules to make them actionable. The Decision Flow allows sub-flows, A/B splits as well as nodes representing rules and models.
From the Decision Flow you can navigate to a Decision Table editor where you can specify the rules against the same properties used by the model, the model outputs or other data. These rules can represented policies like maximum values allowed, brands we don’t want to buy or whatever. The combination gives us good decisions (thanks to the model) that follow our policies,are compliant with regulations etc. Within the environment a set of data can be run through the decision flow and the results analyzed and used to refine the rules. Rule log information available too, important for regulatory compliance and for analyzing decision behavior for continuous improvement. The decision flow can be used to walk through test cases, seeing what is done at each step. Finally it can all be deployed rapidly to support, say, a web or mobile app.
The various projects, changes made and deployed, the lineage of the decision and all its components and much more can be investigated using other SAS tools. Data governance and more can then be applied because all the output of these tools is fully integrated with the SAS platform. Decision Manager and the Business Rules Manager both up for release in Q2 with follow-up releases in Q4. Model Manager of course is already available (review here).
Very cool, looking forward to seeing it in release.