CIO Matters

Prescriptive analytics is coming to a future near you

Linda TucciLinda Tucci

Atanu Basu has seen the future, and initially he didn't know what to call it. "We used to call it predictive decision management, and then we realized prescriptive analytics was a better phrase," he said. Basu is CEO of Ayata, a 10-year-old company that makes software designed to analyze the present in order to predict the future and -- here's its claim to fame -- prescribe how to benefit from what's ahead. Think of it as the next evolution of analytics. Rather than describe the future and leave it up to us humans to decide what to do, prescriptive analytics just tells us. "It's like a cooking recipe: Add oil, wait for five minutes, let the color turn brown; when the color turns brown, now you sprinkle the spice," he said. Voila! Bon appetit!

Well, not quite. The do-this-then-that format of the prescription (action pathway in Ayata parlance) might sound simple, but like a jet plane on autopilot, the mechanics behind it are not. To predict the future, the software must analyze the present and, as CIOs know, the present world never stops generating data, 80% of which does not fit easily into traditional databases. Images, videos, audio, texts -- a digital gusher of data is being pumped out by technologies that were nonexistent not so long ago. Indeed, 90% of the world's data was created in the last two years, as the data mavens are fond of quoting. "Drones! Each drone captures video that would take a human being 80 years to analyze. I mean, I may not live 80 years," Basu said.

To analyze all these images and video requires lots of technology -- image processing, computer vision and machine learning technology. If the video has sound, add speech recognition and natural language processing to the list. "So you're combining scientific disciplines that have never been combined before with functionalities that were not possible before," he said.

Think of it as the next evolution of analytics. Rather than describe the future and leave it up to us humans to decide what to do, prescriptive analytics just tells us.

How these technologies come up with a plan was voodoo to my ears. Basu explained the software extracts variables, concepts, indices, themes and so on from all this unstructured data and combines these variables with numbers in a time-dependent fashion to predict and prescribe. The time-dependent part is crucial because things change. "We are not fixing the problem you have; we are going to preempt the problem you are going to have, without compromising other priorities," he said. The prescription in other words also takes into account any possible side effects of the recommended action. (See Lorraine Baines' near-incestuous encounter with Marty McFly.) Impossible? Not apparently if you possess an enormous amount of algorithmic processing power.

Dynamic recalibration

In prescriptive analytics, algorithms basically run wild. "We use rules, but minimally," Basu said. That's because rules don't scale. They are dependent upon human beings, and in a big data world, human beings cannot keep up. "You end up being as good as the guy making the rules." Instead, the algorithms are allowed to take over but are programmed to adapt automatically based on changes in established parameters. Algorithmic Darwinism? "It's called dynamic calibration, and some of that is guided by rules. A rule could be, 'If you see this data change by more than 10%, trigger recalibration,'" he explained. "We're not the perfect solution, but we are miles ahead of everybody else."

One of the companies willing to take a flyer is Apache Corp., a $16 billion oil and gas company based in Houston. I asked their CTO, Mike Bahorich, how the software was working out for them. "Overall it's going well," he said. Apache uses the software to optimize the equipment and materials required to pump oil out of the ground -- electrical submersible pumps, power generating turbines, the drill rigs and the drilling lubricant or "mud" pumped in when breaking through rock. And there are many other potential applications, he said. I noted Apache had a wastewater toxic spill in June. Did he believe prescriptive analytics could have prevented that? "I would say that a combination of sensors and software could have made a difference in something like that potentially, although we're quite a ways away from having that level of technology for a case like that," Bahorich said. "So I would say in theory it's possible; in practice, we're a long way away. I'd love to see us move in that direction."

Just do it

So Paradise regained is not around the corner. As for whether Ayata is "miles ahead," well, everything is relative. IBM offers prescriptive analytics capability through a combination of software products, notably SPSS, notes Ovum analyst Warren Wilson in a recent report. What seems more certain is that the marketplace is not yet teeming with prescriptive analytics vendors, and very few companies are using the software -- just 3% compared with the 30% of companies using predictive analytics, according to Gartner. But anyone can see how all that is destined to change.

Recent CIO Matters columns

Bad analytics

Social apps need to grow up and go to work

Consumerization hits the e-discovery market

This morning I got an automated call from CVS. After verifying my identity, I was informed that allergy season was approaching and prescriptions can take days, even weeks to fill. Their records show that I am a user of Fluticasone Propionate (a steroidal nasal spray). Would I care to refill my prescription now? A trivial matter to be sure. I did not press 1, my option to load up now, or 2 or 3, because as I listened to the automated offer I was wondering how much Flonase is too much Flonase. Imagine if the autogenerated voice on the other end could predict the future effects of my nasal spray use and tell me with confidence what to do? "Add oil, wait for five minutes …" The dilemmas go on and on, big and small. There are days when I would be happy to let Ayata (Sanskrit for future, by the way) tell me what to do. As for the next step in automated human decision-making? That's a future when humans stand aside and the machines just do it. 

Let us know what you think about the story; email Linda Tucci, executive editor.

This was first published in August 2013