Pervasive Prediction: Waiting is the Hardest Part
A proponent of predictive analytics argues that we're on the cusp of an age of pervasive prediction.
- By Stephen Swoyer
- February 18, 2014
Will predictive analytics (PA) ever be as widely deployed as business intelligence (BI), or is it forever destined to be a kind of niche technology?
A new report from The Data Warehouse Institute (TDWI) suggests that successful uptake and deployment of PA is slowly but steadily increasing.
Moreover, at least one proponent of PA argues that we're on the cusp of an age of pervasive prediction. The traditional barriers to PA -- e.g., poor front-end tools, the complexity of predictive modeling, the inescapably manual process of integrating and transforming data from different sources for analysis -- have been breached, they claim.
Proponents cite a robust lingua franca (the Predictive Model Markup Language, or PMML, now in version 4.1); new delivery paradigms (cloud and/or software-as-a-service); a new user experience (highly visual function- or domain-specific PA apps, typically running on tablets or mobile devices); and, just as important, a new simplicity: PA vendors say they've managed to automate some or most of the esoteric (building, training, and managing predictive models) or time-consuming (data integration) tasks that traditionally complicate PA. Some, such as Predixion Software Inc., are downright bullish on the subject.
Simon Arkell, Predixion's CEO, claims that the problem has to do with established PA players -- namely SAS and IBM; these vendors effectively address a small fraction of the potential PA market, Arkell argues. "SAS and IBM dominate an industry that I think is about 5 percent of the real industry," he argues. "The hidden industry is [i.e., consists of] the net new or greenfield opportunities in predictive analytics. That [segment] doesn't have a leader right now. These are the people within an organization who previously have never had access to predictive analytics. It's the business analyst, the subject matter expert, and the marketing analyst, absolutely -- but also the ultrasound technician, the nurse in the ICU, the maintenance worker in the oilfield."
Predixion's flagship product -- "Insight" -- exposes both a PA design environment (an add-in that drops into Microsoft Excel) and an application development framework (its Machine Learning Semantic Model, or MLSM) for building function- or domain-specific PA apps.
"An MLSM [package] is a kind of end-to-end encapsulated predictive app. It allows for portability. It can be executed in many different environments; you don't have to have it sit on our server or in the cloud," he explains.
In this respect, Predixion's MLSM resembles the "Programmability Extension" feature that the former SPSS Inc. unveiled as part of SPSS 14. It, too, aimed to enable developers to create quasi-portable predictive analytic apps. As Kyle Weeks -- then senior product marketing manager with SPSS -- told BI This Week in 2006: "Our plan is to make it possible ... to embed the entire analytic guts of SPSS in their applications."
The salient point, Arkell argues, is that the MLSM was conceived with portability in mind. It permits Predixion to push PA up, down, and across the business process -- or (in his words) to people "at the point of impact." This gets at what he insists is the biggest difference between Predixion and its established rivals: there are no "users" in Predixion's lexicon. Instead there are "people" -- i.e., workers.
As Arkell sees it, PA was traditionally designed and built for "users;" but this presupposes a certain kind of consumer, usage model, and usage context. Conversely, he argues, designing and building PA for workers entails a very different set of assumptions. The latter scenario is conceived with operationalization in mind, Arkell maintains; the former isn't -- or isn't necessarily.
"Users" live in a business campus, a call center, or in a similar (cubicle-filled) context. Their primary interaction metaphor is point-and-click, in combination with desktop computers.
"Workers," Arkell argues, live outside the business campus. Their primary interaction metaphor is touch-and-swipe, invariably in combination with mobile devices. To equip a nurse in an ICU with an iPad is to explicitly operationalize predictive analytics, Arkell argues: it's to push PA -- and its supporting infrastructure -- into a completely new context.
The traditional mindset of designing PA apps to be deployed on an enterprise desktop, embedded into an existing call center app, and so forth, is a less explicit operationalization of PA, he argues: it's more contingent than necessary.
"At the very least, it's targeting a very different [use case]," he points outs.
"Insight supports that traditional [use case], too, where you take the predictive app or the model that you build [with the MLSM] and push it into any existing software you've been using. You can also [expose] it [in] an entirely new interface -- [as with] a nurse using Predixion on an iPad or iPhone ... being told what needs to be done and going and doing that intervention."