Predictive Shopping Lists and Personalized Recommendations for JJ Food Service Using Azure Machine Learning

Vinod Anantharaman, Business Strategy Lead of Microsoft Information Management and Machine Learning (IMML), presents a case study translated by Xie Shendi.

JJ Food Service, one of the UK’s largest independent food delivery companies, serves over 60,000 customers with more than 4,500 products stored across eight large warehouses, covering fresh, frozen, dry foods, paper, and cleaning supplies.

Customers place orders online or by calling a service centre; the logistics team processes orders daily, packs them overnight, and the freight team delivers them the next morning.

Although the current workflow operates like an assembly line, JJ aims to stay at the forefront of technology.

Since 2004, JJ has used Microsoft Dynamics for ERP and CRM needs, continuously refining its operations over the past decade. Today, Dynamics AX supports HR, procurement, sales, inventory management, and order processing.

Chief Operating Officer Mushtaque Ahmed recognized the valuable customer data they possess as a huge opportunity to increase satisfaction, such as predicting orders before customers type any keywords, but building advanced analytics in‑house would be costly.

Azure Machine Learning was identified as the solution.

Predicting Your Shopping List

Because orders vary by product, timing, quantity, type, frequency, and many other criteria, accurate predictions require personalized analysis of past purchasing behavior, down to the specific week, day, or even time slot.

JJ partnered closely with the Azure team: they wrote code to capture customer‑behavior data, trained a predictive model on three years of transaction history, and integrated the model’s recommendations into both the website and phone ordering experience.

The system was debugged in three months; now, whether a customer orders via phone or online, the same AI‑driven predictions appear and orders are automatically completed.

The result is a more efficient shopping experience that raises customer satisfaction.

Human‑Touch Product Recommendations

Beyond predicting the shopping list, the system suggests related items—for example, offering sauces to a fish‑and‑chips shop that bought batter, or prompting a fast‑food outlet that ordered meat, poultry, vegetables, and drinks to consider oil or paper cups.

JJ estimates that recommendation‑driven purchases account for about 5% of the cart; while modest, the scale makes it significant, and the company aims to increase this share with smarter predictions.

Ahmed notes that the “human‑like recommendation feels personal, and customers are delighted when we accurately anticipate their needs.”

More Efficient Acquisition of New Customers

By using Azure Machine Learning’s recommendation engine, new customers receive product suggestions similar to those purchased by existing clients, saving them time and creating an immediate sense of value.

JJ continues to explore ways to boost satisfaction and sales, such as forecasting near‑future orders to improve inventory turnover, running promotions through the recommendation system, and targeting new products to specific customers.

As Ahmed concludes, “Dynamics AX has done the heavy lifting for automation, but we now need to make those processes smarter—this is where Azure Machine Learning will continue to shine.”