Repost of an article earlier published on the SQL Data Platform Insider blog. 

In an earlier post we talked about a new book titled Predictive Analytics with Microsoft Azure Machine Learning which released in December on Amazon.com where it was doing rather well. The Data Platform Insider blog team at SQL recently had an opportunity to sit down with a couple of Microsoft authors of that book to learn more about their roles as Data Scientists, some of the real-world successes they’ve seen, their perspectives on opportunities in this evolving field as well as about their new book. Click here to hear directly from authors Val and Wee Hyong. ML Blog Team

We continue our series of posts how Microsoft customers are gaining actionable insights on their data by operationalizing ML and advanced analytics – at scale and in the cloud.

As one of the largest independent food delivery service companies in the UK, JJ Food Service provides over 60,000 customers with everything they need for their own food businesses. Their catalog has over 4,500 products ranging from fresh, frozen or dry foods to paper and cleaning supplies and get fulfilled from any of eight warehouses.

Customers can either place orders online or by speaking to call center representatives over the phone. As orders come in each day, logistics teams route and sequence these orders, employees at warehouses then load the appropriate products overnight, and drivers hit their delivery routes the next morning – and the same cycle repeats all over again.

Although the existing processes at JJ Food Service are quite streamlined, as a company that prides itself on staying at the cutting-edge of technology, their ambitions ran much further.

Back in 2004, JJ Food Service implemented Microsoft Dynamics for their ERP and CRM needs. Over the past decade, they refined their operations and Microsoft Dynamics AX now powers their entire operations – right from HR, procurement and sales to warehouse management and order processing.

Recognizing that they had an exceptionally rich vein of customer data, the Chief Operating Officer at JJ Food Service, Mushtaque Ahmed, saw an opportunity to use this data to further boost customer satisfaction. An area where they felt they could save their customers’ some time was by anticipating customer orders, i.e. recommending products to them even before they had entered anything into the system. They had several other ideas for predictive analytics too. At the same time, the company was concerned about the potentially big costs they might incur in staffing up and implementing an advanced analytics project such as this.

That’s when Azure ML entered the picture. 

Predictive Shopping Lists

Customer orders at JJ Food Service, of course, vary widely in terms of what gets purchased and when, order size, type, frequency and many other criteria. In anticipating customers’ future needs, what they needed were tailored insights based on each customer’s past order patterns. For instance, a particular restaurant might order salad greens every day, flour about every two weeks, and cooking oil once a month. “To be successful, we needed to be relevant for that week, that day, that exact point in time,” Ahmed explained.

JJ Food Service was convinced that Azure ML could help them address their needs in a very cost-effective manner. They started working with the Microsoft Azure team, first writing code for their website to capture customer behavior and then using three years of transactional data to train an Azure ML predictive model. Next, they integrated the recommendations from this model into both their call center environment and their website, thus ensuring that their phone-based customers would get the exact same recommendations (via call center representatives) as what online customers would see on their site.

The system took only three months to implement. Today, whether customers call in or log in, the system bubbles up the same predictions using its analysis of past purchases – in both cases, the order pad gets filled out in the same fashion, and automatically.

The net result? More satisfied customers who find a high level of efficiency in their shopping experience.

Recommendations Add a More Personal Touch

In addition to the predictive shopping list, customers also get recommendations for related items that they might want to order. For instance, if a fish and chips shop were to order batter, the system might ask whether they need specific spices that go along with that. Also, prior to checkout, the system reviews the overall order to determine whether the combination of items shopped indicates a need for additional products. For instance, if a fast-food restaurant orders meat, poultry, vegetables and beverages, would it also need cooking oil? Or perhaps paper cups, if their supply might be running low?

JJ Food Service estimates that these recommendations currently make up about 5% of the shopping cart. While that may not seem large – and, in fact, Ahmed expects this number to go down a bit as the system gets smarter at predicting orders even more accurately – when you consider the company’s size, this really adds up. Plus it’s a nice personal touch for customers. As Ahmed says, “The wow factor is huge. Customers are amazed that we can predict so accurately what they need.”

Targeting New Customers More Effectively

JJ Food Service realizes that there’s no better way to capture business from new customers than by making themselves indispensable from the very moment they log on.

By using the Azure ML recommendation system to display products purchased by similar companies, they are now able to show immediate value to brand new customers, shaving valuable time that would otherwise be spent in browsing a new catalog and compiling orders for the first time.

At JJ Food Service this is just the start of a journey. They are looking at additional possibilities beyond increasing customer satisfaction and driving incremental sales. For instance, they plan to stock their warehouses more efficiently by using forecasts of what customers, in aggregate, are likely to buy in the near future. They are also exploring how to use the recommendation system for promotions and to target new product launches at specific types of customers.

As Ahmed concludes. “Microsoft Dynamics AX works hard for us, automating processes. But we also need to make these processes intelligent – and that’s where Microsoft Azure Machine Learning is vital.”

ML Blog Team
Subscribe to this blog. Follow us on twitter

We launched this blog in June 2014 with the intent of sharing important advances and practical knowledge accumulated by Microsoft in the field of ML. After six months of regular posts, many of them authored by world-leading ML researchers and practitioners, we are seeing tens of thousands of readers such as yourself regularly visiting our blog site where, we hope, you are finding articles of value and relevance to your own ML journeys.

As we take one final look back at the year 2014, we figured we would share the top 10 most-read posts of 2014. Here they are, listed below, in increasing order of popularity.

10. Machine Learning, meet Computer Vision
Jamie Shotton, Antonio Criminisi and Sebastian Nowozin explore the challenges of computer vision and touch on the powerful ML technique of decision forests for pixel-wise classification.

9. Python Tools for Visual Studio now integrates with Azure Machine Learning
Shahrokh Mortazavi talks about Python support in Azure ML, including the powerful Python centric Data Science IDE, PTVS – a completely free and open source tool that is helping democratize ML and advanced analytics. 

8. Vowpal Wabbit for Fast Learning
John Langford shares information about the speedy VW open source ML system sponsored by Microsoft.

7. Machine Learning and Text Analytics
Dr. Ashok Chandra talks about how we are now able to take advantage of signals to determine the salient entities being discussed in textual articles.

6. The Joy (and Hard Work) of Machine Learning
Joseph Sirosh discusses how enterprises can tap into the potential of ML to deliver enormous value in diverse applications that can improve customer experience, reduce the risk of systemic failures, grow revenue and bring about significant cost savings.

5. Machine Learning Trends from NIPS 2014
John Platt shares 3 exciting trends he saw at the Neural Information Processing Systems (NIPS) 2014 conference in Montreal this year.

4. What is Machine Learning?
John Platt provides some much-needed context around ML and also shares a taxonomy of ML applications.

3. Twenty Years of Machine Learning at Microsoft
John Platt discusses Microsoft’s 20+ years of experience in creating ML systems and applying them to real world problems, including what it takes to actually deploy ML in production.

2. How Azure ML Partners are Innovating for their Customers
At the Worldwide Partner Conference, Joseph Sirosh talks about how Azure ML – which is changing the game for building ML applications at scale and in the cloud – is being used by Microsoft’s partners to rapidly build novel solutions for our customers.

1. Rapid Progress in Automatic Image Captioning
John Platt talks about the exciting progress researchers have made in creating systems to automatically generate descriptive captions of images.

We wish our readers a very happy and productive 2015!   

ML Blog Team

Our previous post had a video showing Skype's automatic speech translation in action. In this post, we share an infographic from the Skype team about how they perform such automatic speech recognition and translation (including how they translates instant messages in over 40 languages).  

You can register for a preview of the Skype Translator here.

ML Blog Team

Re-posted from an article that recently appeared on 

“… a new Microsoft technology that seems borrowed from the world of Star Trek”

“… a Skype add-on that listens to the English words you speak into Microsoft’s internet phone-calling software and translates them into Spanish, or vice versa.”

“… an amazing technology, and it’s based on work that’s been going on quietly inside Microsoft’s research and development labs for more than a decade.”

Read the original WIRED magazine post here.

ML Blog Team