Four Essentials for Enabling Pattern-Based Strategies

June 22, 2010
81 Views

Economic turmoil has brought widespread recognition that businesses must improve their ability to identify and act on signs of change. More prescient, quicker action is needed at the macro level, where companies can’t afford to be unprepared again for downturns. It’s also needed at the micro level, where the challenge is to more accurately forecast and treat changing customer behaviors.

Economic turmoil has brought widespread recognition that businesses must improve their ability to identify and act on signs of change. More prescient, quicker action is needed at the macro level, where companies can’t afford to be unprepared again for downturns. It’s also needed at the micro level, where the challenge is to more accurately forecast and treat changing customer behaviors.

But how can organizations consistently and reliably pull signs of change from our information-dense markets? And how can they proactively respond to impending change in a timely manner—that is, before the situation changes again?

Recently published “Pattern-Based Strategy” research by Gartner addresses this issue: “The environment emerging from the recession demands an increased focus on detecting leading indicators of change, and on identifying and quantifying risk emerging from new patterns.”1 We are moving, according to Gartner, “from a world of ‘sense and respond’ to one focused on ‘seek and act.’”

In addition to Gartner’s findings and direction, IT organizations should pursue four essential practices—based on underlying technologies—to enable pattern-based strategies:

1Recognize Business-Significant Patterns of Change Early

Predictive models are statistical analytics that predict the likelihood of a customer behavior or other occurrence. They’re used widely in risk management and in fraud management, but are applicable to a wide range of business requirements where subtle insights need to be pulled from massive quantities of data. They are the best “early-warning” system because they identify the future impact of sudden changes in data.  One reason predictive models have not been used more extensively to date is the time and expense traditionally required to hand-code them into operational systems. But today, models can be rapidly deployed, without recoding, directly into business-rules-driven processes (discussed later), making ROI from model development clearer and more compelling than ever.

Predictive analytics that examine transaction data—models that recognize patterns in consumer transactions such as purchases, bill payments, insurance claims and customer service inquiries—are particularly valuable for pattern-based strategies. Such transactions yield rich detail (what was purchased, when, where, for how much) for analysis. As a result, transactional analytics spot the first signs of changes in risk, and in opportunity. Early indications of difficulty paying bills, for example, can enable a company to apply proactive treatments to avoid delinquency. Or with indications of a new home purchase, a company could offer related goods or services before competitors.

To decode data patterns even faster, businesses today are adopting advanced technologies such as genetic algorithms, which automatically pull out characteristic variables from vast amounts of historical transaction data and test their predictiveness. Unlike traditional data mining, which requires predetermined variables, this technique can detect unknown variables driving emerging patterns.

2Efficiently Adjust Strategies to Initial Signs of Change

When patterns indicating significant change are detected, companies must act swiftly to make corresponding operational changes that will mitigate the negative effects and amplify the positive effects on their business. The only way to do that efficiently and fast enough—before conditions change again—is with business rules management.   Whether rules technology is deployed as a component of a specific application or as a decision service called by multiple applications, it eliminates the need for IT to hand-code decision logic modifications. As a result, making an operational change that would have traditionally required weeks can be accomplished by business users themselves in days or hours. And it’s done without recompiling software, disrupting production operations or compromising quality control processes.

Today’s leading business rules management systems accelerate operational adjustments by enabling predictive models to be imported directly into rules-driven strategies, without recoding. And the most cutting-edge business rules management systems support PMML (Predictive Model Markup Language), a widely used industry standard within the analytic modeling community.

After import, models can continue to be viewed and modified. Business users can make minor adjustments to the weights of a model without having to export and re-import it.

To increase speed and productivity, rules management tools should help business users quickly zero in on just those places needing attention. Business users can also minimize elapsed time and errors through the auto-completion of rule syntax, auto-highlighting of differences between rule services or change versions, rule verification, and validation.

3. Accurately Anticipate the Results of Strategy Changes

One major reason businesses react slowly to change is uncertainty over the possible outcome of modifying their policies. Today, companies can act sooner with greater confidence by employing simulation tools to run historical data through the proposed modified ruleset, including imported models, and analyze the probable business impact. They can also rapidly test alternatives and tweaks to select the strategy that produces the best simulated outcome.  For example, an insurance company could use simulation to make sure that a new underwriting rule wouldn’t inadvertently skew the distribution of tier assignments. Or, it could probe the upside of change by rapidly comparing dozens of slight rule variations.

Strategy optimization offers another approach to confidently launch a new action. Optimization mathematically identifies the best decision strategy for achieving a particular business goal given multiple (even opposing) objectives and constraints. The underpinning for identification and understanding of the optimized strategy is decision modeling. It maps the relationships (very complex patterns!) between numerous inputs, including those from multiple predictive models, possible actions by the business, and likely reactions by customers.

Optimization is critical for dynamic business environments because it enables companies detecting significant patterns of change to start making decisions that are optimal—and more profitable—for the new business situation, before conditions change. Given the pace of change today, conventional trial-and-error testing is too slow and unfocused to drive high performance. Even companies employing experimental design (methodologies that yield more learning from fewer tests), may always be approaching but never arriving at optimal. In addition, time-to-optimal can be further accelerated by deploying optimized strategies via rules management systems.

4. Measure Outcomes, Push Performance…and Prepare for More Change

While simulation and optimization tell managers what to expect from their adjusted strategies, production testing confirms that the changes are meeting these expectations. It also yields data for subsequent, rapid iterations of strategy refinement, pushing performance to higher levels.

Business rules systems should include a strong framework of tools enabling business users to perform systematic champion/challenger testing across multiple channels. This involves applying the adjusted strategy (“the challenger”) to a small, randomly selected population sample, then comparing the results to those of the current business-as-usual strategy (“the champion”).

If the challenger proves superior, business users should be able to promote it—with “push button” ease and speed—to champion status by rolling it out to the larger population. However, it’s important to continue with regular, systematic champion/challenger testing. Where strategies have not been optimized, continued testing enables companies to move toward optimal operating points in an incremental fashion. Even with optimization, continued testing is essential since it reveals when markets and economics are shifting enough to move to a new optimal operating point.

From Reactive to Proactive 

In the post-recession era, top-performers are shifting from reactive to proactive management. To make this transition, companies must be able to detect patterns indicative of change emerging in the overall business environment, and in customer behavior. Companies also need to be able to quickly implement pattern-based strategies that mitigate or amplify a significant change’s effects on their operations.

Many companies are already using, in some areas, the essential enabling technologies for pattern-based strategies: predictive analytics, business rules management, simulation, optimization and a tight feedback loop from operations back to analytics. To support this new level of proactive performance, they must now extend these technologies across their operations.

Don Griest is the Director of Product Management at FICO, where he helps businesses automate, improve and connect decisions across organizational silos and customer lifecycles through effective predictive analytics tools and strategies.