First sip of Context Relevant hints at a winner

Most product vendors can talk for half an hour without even one specific case. But Stephen Purpura, CEO of the predictive analytics startup Context Relevant, actually had a story about something useful — the time his system used scant data to correctly point to a barely known winemaker.

Spotting the wine started with a sip, of course. But as you might imagine, Purpura then fed it to his prediction machine.

Context Relevant’s critical edge is speed, he explained. That comes from good caching; the better you cache, the faster you retrieve, and the faster you can work through questions and arrive at answers you can act on.

“I can almost guarantee the first question is wrong,” said Purpura. “The trick is how fast you adapt.” It’s analogous to arriving at a baseball game: Find the stadium, find the section, find the seat. You aim to do it all before the game begins.

Purpura said that the Context Relevant technology is finally, after two years, fulfilling the vision: to let the user ask a question and get an answer in the time it took to take a sip of coffee. Today, “no one else comes even close to this,” he asserted.

Who is it that actually sits there asking and sipping? For new customers, it’s Context Relevant staff. But he said that with training provided, the customer quickly takes it on. Users on the customer side include anyone from analysts to project managers, quants, traders, data scientists, and IT pros. CR hides the details by default, but statisticians and others can dig in at will.

Context Relevant’s marketing website is lightweight, which I assume is deliberate. Purpura’s trying to reach upper level executives, who are best reached through social contacts. He said the best way to pique their interest is with stories told orally. He goes where they go, and so he serves on boards and attends lots of social events.

Wine is often part of it. At one event produced by the famous Cornell University restaurant school, he said, “this kid had incredible wine.”

The “kid” was Aaron Pott, years later named “winemaker of the year” by several publications. But back then, there was little data on him. He didn’t even have the metric many wine buffs look to for guidance, the Parker score.

Did the sparse data mean game over? Apparently not. Purpura’s machine found plenty in wine forums, including the Parker forum. Comments about wine that did have a Parker score often showed up alongside mentions of Pott’s wines. Mentions by certain individuals also indicated quality. So did Pott’s having worked at wineries that had strong scores.

To a non-technical type like me, sifting through all that sounds easy. But in fact it was, as Purpura put it, “a massive simulation problem embedded within a massive graph problem embedded within a regression problem with millions of parameters and millions of rows of data.” He said that it would be a struggle for most other companies today.

His system inferred great potential for Pott. Years later, in 2012, Food and Wine magazine named Pott winemaker of the year.

To put money on the company, I would need to know more, starting with what the system got wrong. What I do know is that this CEO does a much better job than most telling the story, talking about practical results and not just technology. (Very few industry analysts who like that stuff have any depth there anyway!) What his system did, in fact, makes sense. It’s what I would have tried to do manually.

Another good sign is that on May 20 Context Relevant scored $21 million in new funding. See The Wall Street Journal for details.

I look forward to more from Context Relevant.

Leave a Reply

Your email address will not be published. Required fields are marked *