Architect Real-Time Analytics Into Operational Systems for Competitive Advantage
Just as everything is “in the cloud” these days, it seems that every analytic operation is performed in “real time.” But in truth, the definition of real time can vary widely, depending on the use case.
James Taylor, CEO, Decision Management Solutions, offers a taxonomy of decisions (ibm.co/1EiZi8c) that can be helpful in this discussion. Among them:
- Strategic decisions are the C-suite decisions that guide the overall direction of the enterprise, such as “Should we expand overseas?”
- Tactical decisions manage and control operations. These decisions operationalize strategic decisions, as in “What product should we promote today?”
- Operational decisions put policy into practice, handling each and every customer interaction. For a payment processing system, a typical decision would be, “Is this payment request valid?”
Strategic and tactical decisions are nearly always finalized by humans, who use analysis to inform their choices. Real time, in this context, means getting those reports back within hours or minutes as opposed to days or weeks. The vendors using this “real-time” qualifier are more realistically experiencing “right time” analytics. In actuality, real time is far faster.
The z Systems* platform provides solutions that can produce results when needed—in particular the IBM DB2* Analytics Accelerator for z/OS* and DB2 with BLU Acceleration* for Linux* on z Systems.
Operational decisions are invariably tied to transaction processing systems, which execute hundreds of thousands to millions of transactions daily. Each transaction is the opportunity to put in play a decision that can either improve business results (e.g., suggest-selling) or lower business expense (e.g., detecting and preventing fraud, waste and abuse).
Because a customer is often waiting at the other end of these transactions, service-level agreements (SLAs) are typically measured in subseconds. If you’re looking to inject intelligence into operations with subsecond response time, analytics must more tightly integrate with operational business processes.
Why Predictive Analytics?
Predictive analytics is the go-to technology for operational decisions. Predictive models are designed to detect patterns of behavior and examine how well transactions match up to these patterns. A transaction closely matching a certain buying pattern might trigger an upsell or cross-sell action, while a transaction that closely matches a pattern of fraud might cause a payment request to be canceled. A series of transactions that heavily deviate from known patterns might indicate the emergence of new patterns that can then be figured into the models.
You might try to do this analysis with business rules, but rules aren’t nearly as flexible as predictive models. In a rules-based system, a sequence of, say, 15 rules could be established to determine if a transaction is likely to be fraudulent. Unless all of the rules fire, the transaction is processed. With predictive modeling, several indicators would generate a probabilistic determination of how well a transaction fits known patterns of fraud. A threshold can be set—say 85 percent—that the business deems the point at which a fraud prevention action must be taken. It’s possible that a fraudulent transaction would pass some of the rules but still generate a score of 85 percent.
comments powered by