What’s the advantage of looking at machine learning from the platform perspective? Breadth and depth, says Skytree co-founder and CEO Martin Hack. In contrast to a single point algorithm that learns about one thing really well for a particular need as it goes about analyzing more and more Big Data sets, Skytree takes the tack that there’s a whole universe of algorithms and methods that can be applied to support a variety of use cases.
The company is a startup, closing a Series A funding round this spring to the tune of $18 million. Skytree’s algorithms are available in different packages for application across industries, or even to fit different needs within the same industry or company. The six biggest machine learning use cases across industries, Hack says, are classification, regression, clustering, entity estimation, dimension reduction and multidimensional querying, and those six tasks, individually or in combination, provide the foundation for performing advanced analytics such as outlier detection or value prediction, among others, that are applicable to various scenarios. “Outlier detection could be used in trying to detect fraud or to find terrorists,” he says. “It’s virtually the same platform. The data changes and the output changes, but the computation part is the same.”
Even within the same company, there can be a call to leverage machine learning algorithms for diverse reasons. In financial services, for example, there are marketing and risk constituents that usually operate separately for regulatory reasons, he notes, but their requirements from an analytics point of view are very close. “Literally can use the same platform,” he says. Organizations, he adds, “can just keep adding the services they need to satisfy their data demands. The product scales linearly so they can keep adding another system to have more firepower available.”
In fact, Hack says, it’s increasingly the case for machine learning use cases to be driven by marketing and sales executives. “Marketing is very into advanced analytics,” he says. Skytree supports the CMO role with advanced marketing analytics to apply to the full customer lifecycle, from targeting to conversion to retention. “Those three components have machine learning one way or another. And the nice thing is the algorithm, whether you are scoring a customer for this or for their risk for a financial transaction, it’s the same. The data is different but the algorithm is the same, so it’s very repeatable and universally applicable,” Hack says.
Another advantage Skytree points to is that customers can go from a modeling environment to a production environment in one step. “In the past there were many steps that made it cumbersome, because usually the modeler would build the risk or prediction model on a sub-sample on the desktop, then someone has to rewrite for the mainframe,” he says. “You don’t have to do that with ours. It’s a click and apply model on an enterprise instance.”
As Hack sees it, Big Data is creating a real sea change across business at large, in terms of their desire to understand that data to transform their enterprise. A few years ago, the idea of applying machine learning was still pretty esoteric, he said. But now, “the large Fortune 500s who may not have been sophisticated with advanced analytics, but does know basic BI – we talk to lot of them now, and they want to get to the next stage, from what we do with BI today to those six tasks [noted above],” he says. “The analog is 25 to 27 years ago, when RDBMs arrived on the market and people said why do we need one. You wouldn’t ask that now. We see the same thing happening – companies asking, ‘Can we just buy a system? We just want to be the best analyst on our own data.’”
That said, many executives in those companies who want to leverage machine learning, like those in marketing, don’t necessarily want to have to become experts at it themselves. “What they couldn’t get was how to buy vs. build themselves,” Hack says, and now with Skytree’s off-the-shelf approach they don’t have to.
Among next-steps for Skytree: It’s currently is working on a Hadoop integration.