Table of contents for IBM Analytics Analyst Summit 2012
IBM has two core Decision Management products, Operational Decision Management and Analytical Decision Management. As noted yesterday CIOs see a need to drive better real time decisions while CEOs see the need to differentiate their organizations by translating insight into actions (by making decisions). The Decision Management solutions are focused on optimizing decisions – not necessarily to fine the theoretical optimum but the best possible in the time and circumstances. These capabilities are delivered as Decision Services to business processes, solutions and applications both from IBM, other vendors and custom development by customers. IBM sees the benefits of Decision Management systems in terms of improved profitability (lower costs, less fraud, better results); increased compliance (with policies and regulations); and more precise and effective risk management.
Business rules and predictive analytics are complementary:
- Business rules establish what we know or are certain off while predictive analytics turn what we are uncertain of into usable probabilities, exposing implicit knowledge
- Business rules determine what to do, what action to take, while predictive analytics enable better choices by showing what is likely to happen
- Business rules are focus on business-friendly natural language while predictive analytics show as mathematical expressions or business rule artifacts like decision trees.
- Business rules are generated as code or executed using an engine while predictive analytics can be deployed in database, as rules, using PMML or using a scoring service.
As an aside, optimization can be applied to arbitrate across a portfolio of decisions as well as to optimize a specific decision (such as the best configuration for a product for instance).
Business rules and predictive analytics can be used to cover a wide range of solutions from identifying situations, eligibility, validation, compliance, calculation, risk, customer opportunity, classification and segmentation and optimization. The first few are very focused on business rules and perhaps event handling while the latter ones are more analytically oriented.
The end game is to be able to pull business rules and predictive analytics together and deploy it as a decision service that can be consumed by business processes, applications and solutions. The rules handle policies, regulations, best practices and know-how. Predictive analytics predict risk, segmentation propensity and associations. An ability to simulate and asses scenarios across decisions is also critical.
Once deployed these decisions are either used in a fully automated process where every decision is handled by the decision service or in a process with an option for manual review of challenging situations, contradictions or exceptions.
Like me, IBM often see decisions locked within processes and the potential for reducing the complexity of processes by building decision services and using those decision services in the process (see this white paper and this presentation for more).
An example from insurance is the claims process that contains decision services to:
- Validate completeness of the claim or first notice of loss (rules)
- Validating customer eligibility (rules)
- Assessing fraud risk and routing high risk/low risk claims differently (rules and analytics)
- Calculating the amount to pay, adjudicating the claim (rules)
These decision services share rulesets and predictive analytic models that implement the sub-decisions on which these larger decisions are dependent (increasingly modeled using dependency networks like the one shown). All of these have to be managed and governed so that they can evolve and improve over time –a decision analysis process for continuous improvement.
A second example would be customer next best action where customer opportunity, risk and cost, compliance and competition must all be traded off to make a best offer. In this case decisions include confirming the application is complete, segmenting customers for propensity, risk, loyalty etc as well as segmentation and ranking of available products and offers. Customer specific recommendations and possibly dynamic /personalized pricing for products with adaptive control (A/B testing) and more. In this case, from a process point of view, there is only a single decision (what action to take) but it is a complex one with a deep decomposition into its component decisions.
In both cases it is important to note that while predictive analytic models are being used to score transactions in real time they do not require the models to be built in real-time. Off line analysis and modeling can be done to figure out what the patterns are and then the resulting model can be injected into a real-time decision service where it is executed.
IBM sees Decision Management being deployed in several ways:
- Tactical deployments focus on a specific business performance issue in a given process and where the process design changes little when the decision is automated.
- Incremental deployments where processes are changed incrementally as decision automation allows them to be simplified and where straight through processing increases
- Strategic deployments where a focus on a new decision-centric business approach, on making processes more customer-centric or on continuous improvement and making processes adaptive
IBM allows customers to begin with a rules-based approach (WebSphere Operational Decision Management) or with a more analytic focus (SPSS Analytical Decision Management). The latest release of ODM (release 8 briefly described here) has integrated business rules and business event processing while focusing on improving collaboration and business user management of business rules. In particular a new social interface showing who has changed what improves collaboration, new editors improve the look and feel and usability of the environment, timelines for better version management etc.
Analytical Decision Management integrates with SPSS Modeler so that all the various analytical modeling approaches are available in Decision Management. ADM allows rules to be defined that leverage these models to define eligibility for a campaign, eligibility for specific actions to take etc. You can also now link to rules managed in ODM, for instance for complex eligibility rules. Simulation of rules and their impact as well as optimization of the decisions across various options and much more is also available.