≡ Menu

Decision Management – aligning strategy and operations #bas2010


The folks from SPSS and ILOG presented on Decision Management – Aligning Organizational Strategy with Day-to-Day Operations. I have blogged before about SPSS Decision Management and Modeler products. Before acquiring SPSS, IBM talked about the move towards optimizing decisions – replacing sense and respond with predict and act, becoming more fact-driven in real-time and driving analytics to the point of impact to deliver optimized processes. As was discussed yesterday, IBM sees a clear need to link strategy and execution. IBM’s definition of Decision Management is essentially the same as mine:

Decision Management is an approach, combining software and expertise, to automate and improve decision-making

IBM also sees the same spectrum of decisions as Neil and I discussed in Smart (Enough) Systems – from strategic to tactical to operational – ranging from collaborative decisions that can take months or years to operational decisions that must be optimized in real-time. From “how do we differentiate” to “we should have more customer service staff to provide better service” to “we should pay this claim automatically to deliver excellent customer service”.

SPSS began by considering the Observe, Orient, Decide, Act methodology developed by John Boyd. Layered on this is the range of strategic, tactical (managerial control) and operational execution decisions. This means one can Orient by making sense of the big picture, Decide in terms of tactical decisions to change the way we do things, Act operationally by automating decisions and then Observing results so you can learn over time. Different IBM products add value in each area – the collaboration and reporting tools – Cognos – help a lot in Orienting correctly and making strategic decisions. Deciding uses rules and predictive analytics while the WebSphere and SPSS Decision Management delivers value in the Act piece while the event processing and other monitoring tools help with Observe. Of course agility is key in this loop – as John Boyd said of his methodology:

“He who can handle the quickest rate of change wins”

As you would expect, IBM sees business rules and predictive analytics as complementary. Rules handle the explicit knowledge an organization has – regulations, policies,  actions to take etc. These are managed in a way that is accessible to business users and are executable. Predictive analytics meanwhile handle uncertainty, turning uncertainty into usable probability, and are more focused on mathematical ways to develop insight. Used together you can make complex decisions and do rapidly and automatically (though this is not news for anyone who reads this blog regularly).

IBM sees a number of ways to use business rules and predictive analytics together. At design time rules can be used to filter data for predictive modeling, data mining and predictive analytic techniques can generate rules and create explicit rules artifacts like decision trees and scorecards. More interestingly they can be used together at runtime to make automated decisions – combining executable analytic models with rules to take the right action. ILOG and SPSS allow the use of analytics and rules in a single service – bringing PMML definitions of predictive models into the rules environment so that the model can be represented as rules – as well as being able to call SPSS services as a step in a ruleflow. When combined like this sophisticated decisions can be automated and then easily deployed into a business process framework to build smarter processes. IBM has a number of customers already doing this and the products are continuing to evolve to give people more options to use them together. This, of course, is how companies I work with have been using the tools together.

The new SPSS Decision Management 6 product gives companies another option.  The vision was to replace the delivery of predictive analytics as a mathematical model with the delivery of predictive analytics through an automated, personalized interaction for customers (think the Gap store in Minority Report). The product is designed to let people make complex decisions in very high volume. Examples include border protection where the decision is which car to search, insurance where the decision is which claims are fraudulent (like Infinity Insurance) or customer retention where the decision is which customers to save and how. Generally people don’t see Decision Management 6 in operation directly – it drives a different experience like a customized offer in a web site. When you look at these decisions they cover many different situations but they have a common sequence of steps. These were embedded, in a configurable way, into the product:

  1. Connect to Data
    Select data sources
  2. Define Global Selections
    Defining a subset of customers, for instance eliminating customers who are too young or have been marketed to extensively already
  3. Define desired outcomes
    What are the actions that the decision can select from (decisions, remember, must commit to an action and select the action from a list)
  4. Define rules and models
    Simple rules interface allows users to define priorities and experiences and then use existing or new models to give these rules insight to work with. The new interface simplifies the creation of new models while also allowing experts to create and modify models for use in the decision.
  5. Optimize outcomes
    Allows the business user to describe how to trade-off rules and models and simulate their rules and models to see what business results they get – simulation for impact analysis, a key step in bringing business users into the decision definition process.
  6. Deploy
    The product supports one-click deployment as a service so that it can be used by other systems, services or processes.
  7. Report on outcomes
    Integration with various reporting environments so that results can be analyzed over time to feedback the results of these decisions to see what strategic or tactical decisions should be made to improve those results.

The Decision Management 6 product also takes advantage of improvements in SPSS Modeler too, where SPSS has been investing in making the interface for building models easier. Automated technique selection and data conversion has been extended with a new business user interface called Modeler Advantage which builds a Modeler flow under the covers. This allows collaboration between the analytic team and the business users around Modeler flows.

As discussed earlier, you can use Decision Management6 with ILOG if you need more rules capability though open integration relying on service calls between the products. Decision Management 6 ships with 2 versions – one for customer interactions and one for claims processing. IBM says that Decision Management is selling well and quickly. Clearly Decision Management’s time has come at IBM.


Comments on this entry are closed.