Partners Say New Azure Machine Learning Service Could Be Microsoft's Secret Weapon In The Cloud

Microsoft earlier this month launched a preview of a new cloud service called Azure Machine Learning, and some partners are already using it to build the sort of apps that used to require a team of expensive data scientists to pull off.

While business intelligence technology slices and dices data to see what has happened in the past, machine learning -- also known as predictive analytics -- crunches historical data to predict what will happen in the future.

Azure Machine Learning is a public cloud-based service that lets developers embed predictive analytics into their applications, Joseph Sirosh, corporate vice president in Microsoft's Cloud and Enterprise group, said in a recent interview.

[Related: Microsoft Preps Preview Of Cloud-Based Machine Learning Service For Developers ]

Machine learning software has been around for years but isn't easy to use or deploy, and it's also expensive, Sirosh said. Packaging up machine-learning-as-a-cloud service solves these problems, and by being first to bring it to market, Microsoft has a head start on the likes of Google, Amazon and IBM, he said.

"I think, on this particular front, that we are the leaders," Sirosh told CRN.

Hiring Sirosh was something of a coup for Microsoft. He joined last July from Amazon, where he spent close to nine years as a vice president in various machine-learning-related roles. Sirosh said former Microsoft CEO Steve Ballmer and current CEO Satya Nadella -- then head of the Cloud and Enterprise group -- recruited him to help Microsoft build a platform that would "democratize" machine learning.

"The reason machine learning needs to be democratized is because until now, it's been the realm of few highly skilled people," Sirosh told CRN.

Microsoft started developing machine learning technology in 1992, and uses it today in Xbox, Bing and its Cortana digital assistant for Windows Phone. The technology also is effective in fraud prevention and for predicting consumer buying behavior. Sirosh said Microsoft has more than 100 partners using the Azure ML preview.

Versium, a Redmond, Wash.-based Microsoft partner, uses Azure ML with its proprietary LifeData predictive analytics service. Combining this with data from its customers' CRM, marketing and other internal systems lets Versium generate "scores" that predict consumer behavior, Versium CEO Chris Matty said in an interview.

A credit score is a simple example, but Versium is currently developing many other types of scores with leading consumer brands, and it's using Microsoft's Azure ML service to build them.

Versium's scores can be used to verify the identities of online shoppers, predict the likelihood that they'll cancel a service to jump to a competitor, and gauge their influence on social media, among other things, Matty said.

"They're all designed to answer questions: Which of my customers are likely to scam me? Who is likely to donate to my cause?" Matty said of the scores. With Azure ML, "We can inject data, build our model, tune it and deploy in a matter of a couple of weeks."

NEXT: How Other Microsoft Partners Are Using The Service

Greg Gomez, vice president of sales at Neal Analytics, a Seattle-based Microsoft partner, said his firm is using Azure ML to help one customer -- an e-commerce company -- get better ad placement for search terms.

Search firms typically sell ad placement via auction, and Neal Analytics is using Azure ML to predict the bids that are required to get a particular ad into a particular position. The service works in conjunction with Neal Analytics' own portfolio of more than 100,000 keywords, each with specific attributes like where the seller and buyers are located, and how often that item is searched for.

"Azure ML allows us to automatically search and find the relationships between those different attributes that are associated with the bid and position," Gomez told CRN.

MAX451, a Seattle-based Microsoft partner, was an early Azure ML adopter and used the service to build an app that automates change management processes. Kristian Kimbro Rickard, CEO of MAX451, told CRN the service was able to handle some tasks that used to require the hiring of expensive data scientists.

Data scientists are probably the most sought-after talent in the tech industry today, so Azure ML could be an attractive option for firms that can't afford to hire them.

When it launches as a paid service, Azure ML will cost $0.38/hour, and Microsoft will also charge $0.18 per 1,000 predictions, plus a $0.75/hour compute charge for deployed web services.

Sirosh said this pricing is about 100 times better than machine learning software currently on the market. That's because Microsoft is charging based on utilization and compute hours, as opposed to big up-front software licensing fees, he said.

Sirosh said the Azure ML preview is fully functional today, but Microsoft is still fine-tuning the service so it can meet its service level agreements. "It needs some baking time for us to feel really comfortable that we have the SLA and can convincingly stand behind it. I'm guessing a few months," said Sirosh.

Microsoft is talking with several vendors about Azure ML partnerships, and Sirosh said a number of big companies will be building predictive analytics apps with the service. These will be revealed in the next few months, he said.

"Machine learning has been hard to do, so people only applied it to the highest value problems. Now that the technology is easier to use, every developer can start using it," Sirosh said.

PUBLISHED JULY 31, 2014