Predixion Aims to Make Predictive Analytics Simple, Accessible
Predixion's goal was to develop a solution that makes data mining and predictive analytics consumable and usable by rank-and-file workers. Has it succeeded?
Predixion Software Inc. made its U.S. splash last year, promising to democratize, or at the very least to demystify, data mining. Predixion's project wasn't just to bring data mining and predictive analytics down from the proverbial mountaintop, but to make them consumable -- to make them intelligible and self-serviceable -- for rank-and-file users, too.
How is the project coming along?
Pretty well, says CEO Simon Arkell, who notes that Predixion shipped a revamped version 2.2 release of its flagship Insight offering in early March. The product delivers support for Microsoft Corp.'s new SQL Server 2012 release. In a slight revision of Predixion's über-democratic, cloud-based model, the revamped Insight 2.2 supports either on-premises or -- via its new partnership with Intel Corp. -- hybrid cloud deployment schemes. (Predixion also has partnerships with Microsoft Corp. -- Insight runs against SQL Server Analysis Services (SSAS) -- and EMC Corp.)
Both vendors resell Insight into customer accounts. Arkell identifies this OEM reach as an important aspect of the democratization of data mining: a company such as Predixion lacks the resources to market and pitch Insight as credibly as an EMC, Microsoft, or Intel, he stresses.
"Our goal all along has been to make predictive analytics and data mining simpler and more accessible for everyone. We're accomplishing that not only through the enhancements we've made to [Predixion] Insight but also through our partnerships with companies [such as] EMC, Microsoft, and Intel," he points out, adding that Predixion's founding brain trust once worked on Microsoft's SQL Server Analysis Services (SSAS) team.
Arkell doesn't say it, but his implication is clear: data mining won't be democratized in a year, and if it's going to be democratized, it's going to need a few evangelists (aka OEMs, in contemporary parlance).
Predixion's pitch has changed, too. It's considerably more focused than a year ago. The democracy and the pervasiveness are still there, but -- these days, at least -- Predixion sounds more pragmatic. Arkell, for example, points to Insight's success among Global 2000 accounts -- it counts a score of such companies (including Chevron, Cisco Systems Inc., and Pilot Flying J) as reference customers -- and says Predixion's had particular success in the health-care vertical, where managed care giant Kaiser Permanente is a customer.
Predixion's messaging has changed in other ways, too. Last year, for example, Arkell evangelized data mining democracy as an end unto itself. Nowadays, he's more willing to talk about specifics: e.g., exposing data mining and predictive analytics to a wider range of users isn't just a good idea -- he explains why it's a good idea. He points to the health-care vertical, where Predixion has developed a predictive analytic technology that addresses the problem of chronic hospital readmissions, which has an estimated $25 billion price tag in the U.S.
Starting in November, Arkell says, Medicare will no longer reimburse health-care providers that readmit patients within 30 days of discharging them. Providers understandably have an incentive to make sure that a patient isn't discharged if readmission within 30 days is likely. The rub, according to Arkell, is that the indicators associated with readmission aren't nationally consistent or even isolated to specific regions.
"We've had to do this on a hospital-by-hospital basis. We found that in New York, far and away the leading indicator for readmission was whether or not the patient takes psychotropic drugs -- particularly anti-depressants. In Texas, however, [the use of] psychotropic drugs was the seventh leading indicator," he explains. "It just shows that if you created a model that every hospital in the U.S. has to plug into, that would be much less helpful."
Because Predixion is predicting for readmission on a hospital-by-hospital basis, Arkell argues, it's relying on feedback and input from rank-and-file users -- in this case, from doctors, nurses, CNAs, and other health-care workers. The contributions of such users will likewise help to drive the evolution of the model, he predicts.
"We're in the process of really integrating [into the predictive model] what happens next. For example, a nurse intervenes and [the model] starts assessing the extra attention they're going to have to give to certain patients based on certain indicators," Arkell says. "This allows you to focus your intervention on the patients that are most likely to readmit. If we can engage and track on the intervention strategy, we can close the loop, and we think that end-to-end loop is where the opportunity is."
Does this vision of predictive analytic nirvana jibe with the one that Arkell touted a little over a year ago, when he suggested that (for example) his grandmother could at some point become both a consumer and a creator of predictive analytic technologies?
That's still the focus, Arkell asserts. "We believe that we can create predictive analysts out of existing business analysts. There are cases where we've actually enhanced existing SAS implementations, and what we've done in Kaiser Permanente is a good example of that," he says, explaining that Kaiser Permanente brought in Predixion to deliver predictive analytic capabilities to employees who weren't using SAS.
"We've created the world's first thin-client visualizations around predictive analytics that are completely interactive -- that are social, that are visual, that are viral by virtue of easily sharing these capabilities [to other potential consumers]."