OpenAI and Choco Create AI Food Distribution Platform

Choco has deployed AI agent infrastructure to handle multimodal order intake across its food distribution network. The platform now processes 8.8m orders per year while reducing manual data entry by up to 50% for distributors and buyers in four regions.

The company operates across the US, UK, Europe and the GCC with tools that convert unstructured inputs into executable transactions. Its approach could show how automation may reshape customer engagement workflows in B2B commerce.

How Choco handles customer orders

Choco serves more than 21,000 distributors and 100,000 buyers through a digital ordering system. As transaction volume grew, the business faced fragmentation in how customers submitted requests.

Orders arrived through email, SMS, voicemail, images and handwritten notes. Each format required manual conversion into ERP systems before fulfilment could begin.

This created friction for operations teams. Customer-specific ordering logic sat in employee memory rather than in automated systems.

"Processing those inputs was the first barrier, but not the hardest one," says Narbeh Mirzaei, Vice President Engineering at Choco.

Building inference layers for customers

Narbeh identified a deeper challenge beyond format conversion. Each buyer had implicit preferences around SKU mappings, unit sizes and delivery schedules that were not recorded digitally.

"The real problem was implicit context: customer-specific SKU mappings, unit preferences, delivery patterns," says Narbeh.

"That knowledge lived in the heads of order desk reps and we needed to encode it into inference layers that resolve ambiguity at the point of order capture."

Choco embedded OpenAI APIs into its core platform to address this. The company built two agent systems to handle inbound orders without human intervention.

OrderAgent processes text, images and documents into structured ERP formats. VoiceAgent manages phone orders using the Realtime API with sub-second response times.

Engineering for customer context

According to Choco, the platform handles more than 200 bn AI tokens in production annually. The infrastructure uses OpenAI models for text, vision and audio processing within a single architecture.

Narbeh says transcription and extraction were the starting point. The engineering work focused on building dynamic learning systems that reference each customer's order history and product catalogue.

"The transcription and extraction capabilities gave us a strong foundation," says Narbeh.

"The real engineering challenge was building dynamic in-context learning infrastructure, so the system resolves ambiguity against each customer's ordering history and catalogue."

"That's what separates automation from intelligence."

Results across the network

According to Choco, error rates remain below one to five percent across automated order processing. The platform operates 24 hours per day across all markets without manual intervention.

Sales team productivity doubled without additional headcount according to the company. The shift away from manual workflows has allowed teams to scale operations efficiently.

Choco plans to expand agent-based systems into sales, commerce and supply chain functions. The focus will remain on mid-market food and beverage customers across its global network.