Machine Learning Helps Food Distributor Reach Its Customers

For B2B organizations, the ongoing shift from call center-based services to ecommerce portals has been a huge boon, decreasing costs and increasing efficiency. But it has also resulted in a loss of interaction with customers, and in many cases that lost interaction has cost them cross-sell and upsell opportunities. The U.K.’s JJ Food Service has responded by leveraging its data and machine learning.

JJ Food Service is one of the largest independent food delivery companies in the U.K., with 60,000 customers in the form of cafes, restaurants, schools, office canteens and the like. The family-owned company has been in business for 26 years and has nationwide operations with eight distribution centers across the U.K. It offers all sorts of food products — fresh, frozen and packaged — as well as related products, like cleaning supplies. In total, the company’s catalog lists more than 4,500 products.

“Our products are bigger pack sizes and predominantly focused to restaurants and catering organizations,” says Mushtaque Ahmed, COO of JJ Food Service. “We try to be a one-stop shop and serve all of their needs.”

The company is no slouch at what it does. Last year, The Grocer, a weekly U.K. magazine focused on the fast-moving consumer goods (FMCG) industry, named JJ Food Service the Wholesaler of the Year.

Until about five or six years ago, Ahmed says, JJ Food Service was almost entirely dependent on call center agents for orders. It accepted orders from other channels like Fax and email (and still does), but the vast majority of orders came through the call center.

“A lot of our customers are not quite tech-savvy,” Ahmed notes.

Then the company built its ecommerce portal. It now takes about 60 percent of its 5,000 daily orders via ecommerce and about 40 percent via call center. The other channels are negligible. On the whole, the move has been very good for JJ Food Service Ahmed says, but the loss of personal contact with customers has hurt.

“Before we had the opportunity to talk to customers and do some upselling, cross-selling, soliciting of new products and telling them about the market segment they were in,” he says. “When we moved to the ecommerce portal, we lost that capability of talking to the customer. There was nobody to tell them about anything new that we were doing or to suggest some of the products they weren’t buying. We were missing out on those opportunities.”

But what the company did have was data. In 2004, JJ Food Service implemented Microsoft Dynamics for ERP and CRM. For the past 10 years it has been refining its operations, and Microsoft Dynamics AX now powers the company’s entire operations, from HR, procurement and sales to warehouse management and order processing.

Ahmed felt that rich vein of data could be the answer to capitalizing on those lost opportunities by using predictive analytics to anticipate customer orders and recommend relevant products to them at the point of sale. To do it right, the company needed to tailor insights based on each customer’s past order patterns; for instance, the fact that a particular customer orders salad greens every day, flour about every two weeks and cooking oil once a month.

Machine Learning in the Cloud

The costs associated with staffing and implementing an advanced analytics project like this seemed daunting, but because JJ Food Service is a Microsoft shop, Ahmed was aware of a new Azure service in preview: the machine learning service Azure ML.

“Machine learning was quite new to us,” Ahmed says. “We didn’t have any clue what was happening out in the market. We thought it must require a big budget with big data scientists involved and we thought we weren’t prepared for that investment. We also didn’t have any internal resources to do a small proof of concept or prototype. But we were already on the Microsoft stack. Azure ML took away a lot of the fear factors. That’s why we went down the Azure ML path.”

Azure ML is a fully managed cloud service for building predictive analytics solutions. It features visual workflows and startup templates for common machine learning tasks, many of them based on the algorithms Microsoft developed for its own products like Xbox and Bing.

The company started working with the Microsoft Azure team, first writing code for the website to capture customer behavior and then leveraging three years of transactional data to train an Azure ML predictive model. They then integrated the recommendations from the model into both their call center environment and their website (both powered by Dynamics) so that phone-based customers get the exact same recommendations (via JJ Food Service’s call center representatives) as online customers.

In all, Ahmed says the system took three months to implement. The end result? When customers call in or log in, they system uses its analysis of past purchases to automatically pre-fill customer shopping carts with the items they intended to purchase. Ahmed says nearly 80 percent of the items customers intend to order are already pre-filled in carts when customers call or log in.

“We can predict what the customers are likely to buy today,” Ahmed says. “Rather than leaving it open to the customer and letting them search product by product and it to the shopping cart, we wanted to make the shopping experience very focused and very quick. Our intention was to keep customers on our website as little as possible.”

Knowing What the Customers Want Before They Do

Making the ordering process more efficient has been a big hit with customers, Ahmed says. And the new predictive analytics capabilities allow JJ Food Service to provide customers with tailored recommendations for related items they may want to order. For instance, if a fish and chips shop orders batter, the system might recommend specific spices to go with it. And just prior to checkout, the system analyzes the cart to determine if the combination of products suggests that something might be missing.

So far, about five percent of items recommended this way get added to customer carts, though Ahmed believes that percentage may go down as the system gets better at anticipating customer needs in the first place. The important thing, Ahmed says, is that many of the products added in this way are items that customers didn’t know JJ Food Service carried.

“The wow factor is huge,” he says. “Customers are amazed that we can predict so accurately what they need.”

JJ Food Service is only at the beginning of its machine learning journey. Based on its initial success, Ahmed says the company is looking for other ways to leverage the technology beyond increasing customer satisfaction and incremental sales.

“We’re now thinking of expanding into other areas,” he says. “Dynamic campaign management, a price optimization service for catalogs so that we can price products based on the time of the year, season and demand — if we can sell at the most optimized price point we won’t overprice or underprice.”

The company may also use the technology to optimize their warehouse stock by using forecasts to determine what customers, in aggregate, are likely to buy in the near future.

Follow Thor on Google+