September 30, 2015 | Written by: Ravi Seshadri
Categorized: Cognitive Computing
Earlier this year our team was working with a couple of large airline clients, pitching some cool analytics software that would predict traffic and weather en-route in airspace or for landing at airports: it would crunch several terabytes of historical data plus real-time feeds and allow airline staff make smarter decisions about flight routes, fuel etc. and thus save millions of dollars.
All went well till the team met up with the CIO: the person who we had believed was our strongest promoter had gone rather cold on the proposal. After the initial emotional meltdown, we sat down with the CIO to understand what was it that really bothered him.
“So your software has this Big Data Analytics capability?” he asked simply. Yes, we said, and rather proudly emphasized that it was actually predictive analytics with advanced data mining algorithms. Of course, he said. And then asked, “So it will need several months historical data?” We nodded, wondering where he was going with this, as we knew his company had archived several years of internal operations data: more than this application would need. What was the problem?
“Can you do predictions with just internal data – of my own operations?” was the next simple question. Obviously not, as predictions would require both real-time and historical externals – which in this case were weather, traffic etc.
“I could give you the real-time feeds but historical… my business has never required me to archive historical external data, so we don’t have that. If this data does not come with the software then we have very little use of a predictive analytics tool.”
Silence in the room was deafening.
With hindsight it looks obvious, but back then we had assumed the client would have all the data needed, and with some cool software we can pour out all the wisdom. Slowly the message dawned on us: the CIO was not buying cool software, he was buying the ability to predict. And since this cannot happen till he had all the data – both for internal operations and external environment, and real-time as well as historical; he would buy only when all these pieces were baked into one offering. A service.
And thus was born a subscription-based data service model enveloping every predictive analytics software we proposed. And a cloud-hosted architecture to deliver all of this as pay-per-use. Simple isn’t it?