Wherever there is data, Azure Machine Learning can make business – and life – better
With data, businesses can predict trends, such as employee attrition and energy demands in buildings. With data, gyms can figure out the best time to send renewal promotions to retain members. With data, a shoe company can decide which products to stock – and which to boot.
“Using data intelligently makes all the difference in the world,” says Joseph Sirosh, corporate vice president of Information Management and Machine Learning at Microsoft. “That’s really where machine learning comes in. Machine learning is really about looking at historical data patterns and being able to predict ahead. It allows you to take the past and peer into the future. So instead of looking in the rearview mirror, you’re looking forward.”
Microsoft Azure Machine Learning helps organizations and businesses make sense of huge amounts of data, and within the Azure Marketplace, it’s possible for them to shop for pre-packaged apps tailored to their problem that can take the information they’ve got and turn it into a strategy they can apply for a better outcome, such as forecasting APIs.
In fall 2014, the Practice of Machine Learning Conference held at Microsoft crowned Dr. Pig the winner of its first-ever Azure Machine Learning Cloud App Contest. Developed by the Microsoft Lab 1711 – the subsidiary team of the Cloud & Enterprise China Engineering team that included an intern whose father raises pigs – this cross-platform mobile app helps small-scale independent pig breeders in China predict market conditions up to six months, which helps them decide on the types and quantity of pigs that will maximize their profits with the most minimal risk so they can run their farm more efficiently. These farmers lack the tools necessary to predict market volatility, so Dr. Pig is able to take the past two years of market data – such as feed and prices of piglets – and project at least six months’ worth of expected profit and loss.
Apps like Dr. Pig break down the barriers that used to prevent small-and-midsized companies from being able to take advantage of the wealth of data they had on-hand.
“We addressed a real problem,” says Jian Zhou, the senior software architect of Microsoft Lab 1711, who has been leading this project for 18 months and making the idea into reality. “Dr. Pig is not just an application, it’s a template.”
“We can easily apply this template to other industries,” adds Bin Zhou, the head of Microsoft Lab 1711, who is passionate to apply Azure Machine Learning technologies to address real problems. “We’ve gone through the data collection and processing, then we leverage the machine learning mechanism so we can easily do predictions. We can predict corn prices, for instance, using a similar approach.” She also mentions other predictions: stock prices, airfare, air quality, peak traffic hours.
The next version of the app – which was developed for Windows Phone and Windows 8.1 using C# — will be developed with the browser-based HMTL5. (Even in the provinces, Bin Zhou says, farmers have smartphones.) It is currently going through the approval process for the Windows Phone Store.
“As long as we can collect the data, we can apply the mechanism and adjust it easily,” says Bin Zhou. “That’s the beauty of Azure Machine Learning.”
The tools in Azure Machine Learning include an easy-to-use visual interface, starter templates and drag and drop workflows. It’s also simple to load data without any programming.
“Two or three decades ago, software and PCs started becoming available in every household, every business. And that changed the game in a big way, with applications from software having a huge impact on people,” Sirosh says. “A new trend is the collection of large amount of data. Data is available in so many forms, being collected by so many sensors – the world is awash in data. And that trend is making it possible to use that data in so many intelligent ways, to make life better, to make businesses better and to do all sorts of things much more efficiently than ever before.”
For instance, ThyssenKrupp Elevator uses machine learning to drive preventive and predictive maintenance of its elevators, ensuring greater reliability and uptime.
At Microsoft, machine learning is used widely throughout the company, Sirosh says. Within Bing, all the ad targeting and showing users what’s most relevant is based on machine learning. It’s the same technology behind an Xbox user who gets suggestions on what to watch and listen to next. Cortana, the personal digital assistant on Windows Phone 8.1, also owes its existence to machine learning, particularly in how it delivers recommendations.
And as software maps well to the rise of apps and the various stores that sell them, data can be packaged the same way using machine learning and made available in something like the Azure Marketplace – an example of the new data science economy.
“Imagine if there’s an app that helps you predict the demand for an item, or predict when something is going to fail, so on and so forth. This is a new market where people who know how to create these kind of machine learning applications can make it available on the cloud,” Sirosh says. And then the business that needs forecasting can find an app that best fits their needs.
“Originally, we took orders in our call center. Over time, about half of our customers switched to ordering online — and in the process, we lost the ability to cross-sell. We needed a way to capture the impulse to buy online, and customers who weren’t talking to the call center weren’t being asked key questions about products they might need to purchase,” says Mushtaque Ahmed, chief operating officer for JJ Food Service Limited. “But by using Azure Machine Learning, we can now make recommendations to customers ordering a particular item. This feature is vital to promote new products or bring customers’ attention to products that they currently go elsewhere to purchase.”
The company is one of the largest independent food delivery service companies in the United Kingdom, and provides more than 60,000 customers with food, paper and cleaning supplies. Now, the wholesale retailer can anticipate customers’ orders in advance by pre-populating their cart using Azure Machine Learning – cutting down the time it takes them to re-order supplies.
“Machine learning has made the biggest impact online,” Ahmed says, “where almost 50 percent of our customers trade.” They were able to create these recommendations based on three years’ worth of sales transaction data from Dynamics AX (ERP) plus the click data from the e-commerce portal — roughly 6GB with approximately 25 million sales transactions lines.
“There is just such a huge gap between the needs of these companies, the wealth of data they’ve collected, and the people who know how to collect that data and build systems that help them optimize their business,” says Sirosh, referring to the shortage of data scientists worldwide. “That’s really where Azure Machine Learning comes in. It really democratizes machine learning and makes it accessible to a much broader audience than before. It puts a simple, integrated tool in their hands that they can deploy as an application.”
Lead image caption: A panoramic presentation of the Dr. Pig app at Microsoft China Center One.