How AI powers farm-to-table supply chains

Automation helps restaurants and their suppliers operate more efficiently, making fresher food a reality.

In a society where software has truly eaten the world, you might be surprised to find that many everyday businesses around you still rely on manual, human-driven processes.

Have you ever visited a restaurant where the vegetables were so fresh, they tasted like they came straight from the garden? There's a lot of operations magic that goes into helping farm-to-table restaurants serve you the best dishes, and this starts with the suppliers who provide your favorite chefs with the freshest vegetables every single day.

In this process, restaurants order their ingredients for the following day by 9pm, with the expectation that their orders will be fulfilled by 5am the following day with zero room for error. For some suppliers, this means providing restaurants with a real point of contact to whom they can text instructions, send photographs, and provide additional clarifying instructions. Software ordering portals with escalation tickets and next-day responses? No, Chef!

An example text message order from a real restaurant.

How does this system work in practice? Restaurant suppliers hire supply-chain account managers who end up spending hours every day parsing text messages and other multimedia to extract and save data on customer purchases. Not only that, but these account managers also must save customer-specific metadata corresponding to considerations such as "What brand of cabbage does restaurant X prefer?" or "Does restaurant Y prefer small, medium, or large strawberries?" For each of these scenarios, the specific product SKU variants that correspond to a customer's specific preferences are saved into the supplier's database.

A sample invoice that would correspond to a fulfilled order.

After a number of restaurant supply companies reached out to us with this identical problem, we added support for this workflow on our platform. We've added a new function that uses state-of-the-art OCR and AI technology to extract information from images and structure them into JSON and CSV format for easy exporting. As always, we also provide human-powered QA to ensure that the resulting structured data accurately maps to the original raw image information. If a customer has an existing database of SKUs that they want to pre-train our models to recognize, they can integrate this database into our workflow with a few clicks and use it as an additional source of truth for QA.

Instead of spending hours every day transcribing text messages, account managers can now just upload the relevant images to Callback and get their answers instantly. Yes, Chef!