5Qs: IBM’s Colin Shearer

How has predictive analytics evolved as an industry during the past few years? 
The market has matured. Businesses  are now aware of the clear potential predictive analytics holds for them – exemplified in numerous high-ROI cases. This technology has gone from use in innovative projects driven by line-of-business visionaries to being a competitive must-have. This has driven vendors to ensure the tools they provide are focused on delivering business value.

Vendor offerings have evolved in three main ways. First, the ability to analyze more types of data. Predictive analytics can now incorporate a far wider range of data types, including unstructured data such as the text from email communications, call transcripts, contact center notes and case logs; sensor data from processes and instrumentation; attitudinal data from surveys and customer feedback and so on. The real breakthrough isn’t the ability to mine each of these data types independently. It’s being able to combine all available data to give a holistic view of the customer or case being modeled. Second, it makes it accessible to more types of users. Predictive analytics used to be the domain of PhD-level experts who consequently had to be the intermediaries driving the technology for areas of the business that wanted to apply analytics. Now line-of-business users are also able to make direct use of these technologies. Finally, it is deployed more. No matter how good the analysis, no matter how accurate the predictions, they only generate ROI when the results are used to do things differently – when they cause CMOs to make decisions where and when they need to be made. This means the offerings have to go beyond just analysis tools. It has to be possible for the results of analysis – the mathematical outputs of “live” predictive models – to be combined with business knowledge – rules, policies, etc – to create decisions, which are then applied to directly enhance business strategy processes.

Businesses live or die by the quality of decisions they make, and time and again we see how evidence-based decisions — those rooted in the hard, indisputable facts of the situation – outperform decisions that are based on “gut feel”, intuition or hunches. The “seat of the pants” approach to decision making is simply unacceptable in today’s world, where intense competition demands continuous top performance, and an unprecedented focus on accountability means every decision is potentially under scrutiny. Analytics provide the basis for consistent, high-quality, evidence-based decision making, and predictive analytics is at the peak of this: providing robust, actionable, forward-looking information that drives superior decisions, better outcomes and higher returns.

Simple test: is the information that’s provided directly actionable and directly relevant to making business decisions? In any business process, there are points where better decisions will lead to better outcomes. The results of predictive analytics can now be delivered directly into key business processes and the operational systems that support them; whether the improved decisions that result are increasing revenues by making more sales or reducing losses by intercepting attempted fraud, they’re directly contributing to the bottom line.

What will predictive analytics look like in the near future? 
An area which is rapidly evolving is the automation and “industrialization” of predictive analytics. Companies most advanced in their usage are moving from analytical teams developing and maintaining each model to running “analytical factories” where analytical processes that would otherwise take significant human time and effort are automated. This gives huge gains in efficiency, and enables large scale application of predictive modeling that would be impractical or impossible by other means – for example, the creation, management and deployment of hundreds or thousands of models specialized by geography, product or customer segment. The resulting benefit to the business and contribution to the bottom line can be orders of magnitude higher than from individual modeling efforts.