Our industry is confused [in thinking] that predictive and statistical models are the only way to predict the future. For example, when you do payment collection, whether it is for a loan origination company or a credit card company or even tax payment collection, the best analytic models associated with doing a better job at that are called Markov decision processes. The best way to model a computer chip production line and the future yield of chip production is called constrained logic programming. The best way to do things like real-time decisioning for cross-selling or up-selling through the call center is through a policy engine and a rules engine.
You won't be surprised to hear that IBM has bundled up all these and other "best way" techniques into prepackaged solutions (including 14 listed in the interview). These solutions will be attractive to many, I'm sure, particularly those who have yet to invest in analytics expertise and development environments. But the gurus already on staff may be skeptical about any claim to offer "the best" approach for everyone. These professionals - whether employed by banks, insurance companies, auto manufacturer, retailers or health care concerns - are used to testing multiple models, using their own data to prove which approach is best for their company-specific scenario. In fact, these people may pride themselves on developing a unique competitive advantage, so they may scoff at the idea of a "best way" for everyone.
Everyone should be receptive to Goyal's point that you have to look beyond just analytics when considering best-possible techniques and approaches. He cites an example of a firm that took an elaborate, and highly expensive, predictive modeling approach to tackle a problem that could have been solved much more simply and inexpensively. In these times in particular, it's a good idea to consider whether, say, a rules engine or another alternative might offer a better, faster, cheaper way of doing things. An analytics champion, after all, may not be inclined to think outside his or her own bag of tricks.
Could it be that IBM is attempting to tarnish the halo that predictive analytics seem to have earned in recent years? Mainstream interest was certainly stoked by Tom Davenport's book Competing on Analytics. Yet despite its $5 billion investment in Cognos, IBM doesn't offer anything akin to the all-purpose analytics development environments offered by SAS and SPSS. For now, Big Blue is expanding on its portfolio of industry- and domain-specific solutions. But if you peek at the "customers" tabs on the SAS and SPSS Web sites, it's easy to understand why IBM would be only too eager to add a hammer to its ever-expanding tool belt.The gist of Ambuj Goyal's message in this week's Q&A interview is that predictive and statistical modeling - key offerings for the likes of SAS and SPSS - are overrated... I'm fine with challenging conventional wisdom and seeking simplest-possible solutions... but I won't be surprised if and when IBM acquires an analytics vendor.