Foodservice Distributor Anticipates Restaurants' Needs with Machine Vision

Learn how Synaptiq helped a large foodservice distributor identify opportunities to sell more and different products to existing customers

Problem:

Foodservice distributors often estimate customer demand using a mix of historical sales data, sales representative insights, and customer-provided forecasts. While these methods work reasonably well in stable conditions, they can struggle with variability introduced by menu changes, seasonality, special events, or unexpected shifts in customer behavior. Reliance on past patterns assumes that demand is relatively consistent, which may not hold in fast-changing environment like restaurants. Restaurants may have multiple suppliers, revealing a limited slice of their total need to any one of them. Human judgment, though valuable, can be inconsistent and hard to scale across a large and diverse customer base.

Due to these challenges, a large foodservice distributor asked Synaptiq to build a data-driven approach to demand forecasting that could scale to a large number of customers, capturing their unique consumption patterns and revealing previously hidden information.

What can other industries learn from AI in food services?

Our client's leadership is committed to uncovering new opportunities with their data and better anticipating the needs of restaurants, which improves sales and enhances their customer service. Working on this project also resulted in our Chief Data Scientist, Dr. Tim Oates, further investigating their data. He helped internal teams “think outside the box,” finding new opportunities for their data and leveraging it to drive additional revenues for the company. Synaptiq became a trusted partner through this work, sharing ideas and testing new hypotheses, such as correlations between foot traffic and sales. The initial engagement grew into a partnership, where all players work together to drive sales and optimization.

Solution:

Restaurants are motivated to expose their menus online to entice customers who are looking for a place to eat and to enable revenue from online orders.  Menus convey a tremendous amount of information about the raw ingredients that are used by a restaurant and their relative quantities.  Synaptiq worked with a large foodservice distributor to build an application to download online menus and extract their contents using machine vision, all at scale.

The application used machine learning to sort through all of the pages and the complexity of restaurant websites and determine which ones were menus.  The data was then stored in a database so the system could easily take advantage of improvements in downstream processing.

Synaptiq then developed a machine vision detector for menu items, visual regions in menus that mention the name of a dish, its price, and its description, which is a challenging task given the extreme variation in menu designs, fonts, and colors.  We then used state-of-the-art optical character recognition (OCR) tools to extract the text of each menu item.

The overall system that Synaptiq built allowed the client to target specific restaurants with a fully automated process, from menu extraction to populating a database with ingredients pulled from menu items.  That database served as the foundation for a new class of analytics tools that enabled them to precisely target customers for upsells, at scale.

Outcome:

The client can now extract menu items up to 100x faster when compared to the slow, manual process that was in place before engaging with Synaptiq.

HUMANKIND OF IMPACT

By closely aligning supply with actual demand, foodservice distributors can minimize excess inventory that might otherwise spoil before it's used, thereby reducing the volume of food that 

is discarded unnecessarily. This not only helps combat food waste, but also lowers costs for businesses and consumers. Moreover, improved forecasting supports more reliable food availability, helping ensure that institutions like schools, hospitals, and shelters have consistent access to the food they need.

100,000,000 more lives could be saved using rapid C-IMT results devices.

HUMANKIND OF IMPACT