Predictive Intelligence
Chapter Outline Minimize

Discrete choice analysis, predictive analytics, and Bayesian analytics all leverage observation to quantify relationships and serve as a foundation for predictive model development. The number of observations is critical to the reliability and utility of predictive models developed. Feedback confirming or weakening the strength of the predictive models is also key to keeping the model relevant. This is why much of the decision analytics reference model is focused on managing pre- and post-decision content. From the standpoint of feedback, decisions to accept an offer (positive reinforcement) are just as useful as decisions to decline or ignore (negative reinforcement) the offer.

Predictive intelligence allows IT-centric enterprises of all types (vendors, partners, and end-users) to more readily understand the competitive landscape that they are a part of and make better informed product, service, and strategy decisions that will improve their competitive position. We have been surprised to see the majority of enterprises that maintain they are market/data-driven or argue that innovation is core to their success are unable to point to any material decisioning based on predictive intelligence. This cobbler’s children syndrome is largely driven by a combination of ignorance and neglect. Most enterprises simply aren’t familiar enough with the benefit of decision analytics to know where to start. Those enterprises that do understand the potential of decision analytics may be stymied by the complexity of leveraging advanced analytics or finding a way to demystify the topic enough to gain the support of senior management.

For those enterprises willing to endure the adoption of predictive intelligence capabilities, the payoff can be transformative. Discrete choice modeling and conjoint analysis provide effective techniques to understand market dynamics and direction in a fully unbiased, normalized, and consistent way. This provides the perfect foundation to chart product roadmaps and identify the key messages by which to go to market. Predictive analytics enable an enterprise to compete more effectively and manage risk. A journey down the predictive analytics road can lead to many destinations. One way predictive analytics can be used is to scorecard customers and business partners. This will help an enterprise evaluate how to avoid risk and capitalize on opportunities. This enables the enterprise to reduce cost and increase revenue, which is the best approach to managing profitability. Bayesian analytics permits an enterprise to better assess the likelihood of events based on historical precedent and then monitor how the probability of occurrence changes as new evidence becomes available. Expressing outcomes in terms of probability is immensely useful because of the normalization that is inherent in how probability is expressed and the increased ability it provides to compare and contrast expected outcomes to enterprise governance, risk, and compliance standards.

There are a wide variety of use cases for decision analytics and predictive intelligence. These use cases can be broadly categorized into operational uses cases (internally focused) and go-to-market use cases (externally focused). These use cases can also be grouped either addressing existing capabilities (current needs) or new requirements (future needs). Figure 7 provides a list of selected predictive intelligence use cases.

Figure 7. Selected Predictive Intelligence Use Cases
Figure

Decision analytics for existing capabilities that are operational frequently use correlation and algorithmic techniques to identify clusters that are very effective at identifying and segmenting/categorizing existing customers. Segmentation and categorization are critical prerequisites to facilitate decisioning through conditional logic. Process automation is the use of technology to automate manual activities or integrate process fragments and it primarily leverages conditional and algorithmic decisioning. Process optimization is just that: an optimization activity that enables the enterprise to make sure resources are used as efficiently as possible.

Decision analytics support for new requirements that are operational uses most of the capabilities in the decision analytics continuum. New product development often uses discrete choice analysis to prioritize development activities. Predictive analytics is used to evaluate customer worthiness which helps with cost avoidance, process improvement, and risk management.

Decision analytics for existing go-to-market activities can use discrete choice modeling to understand the elasticity of demand for your products and service and simulate how best to maximize revenue or profit and position against your competition. Bayesian inferencing is very effective at evaluating and helping minimize risk. Predictive analytics is well known for identifying how to better support your customers and prospects (lead generation) by recommending what promotions should be extended to which segments (push marketing).

Decision analytics for new go-to-market activities leverages discrete choice modeling, conjoint analysis, and collaboration to understand new product requirements, pricing, and how effectively your products will compete against the competition. Predictive analytics and collaboration are very well suited to supporting build/buy/partner decisions and precision marketing.

The Decision Analytics Challenge

Currently, one of the vexing issues in decision analytics is the integration of decisioning tools with analytic routines. The origin of this issue dates back many years. Decisioning tools were initially aligned with languages and environments that paired their capabilities with the application development domain. Analytic tools such as SPSS and SAS initially functioned as standalone tools. As these two domains have evolved, effort has been made to bring them closer together. Predictive Model Markup Language (PMML) was a good start and has a following of loyal users. PMML is XML-based and is Java-friendly. Python and R both have fairly comprehensive statistical capabilities, although no real intersection with decisioning tools. The near term solution to this issue is probably to address it through API services management. A rich set of public APIs for each tool and across tools will help significantly with interoperability issues—although true integration will probably come from within vendors that have both decisioning and analytics capabilities. The goal is being able to seamlessly traverse decisioning and analytic components in a stateful way so that context is preserved.