AI food distribution platform Choco is using OpenAI's API to bring AI agents into the supply chain. The company says its system now handles over 8.8 million orders per year, has processed over 200 billion tokens in production, and has cut manual order entry by 50% while doubling sales team productivity.
Reconstructing food distribution in the AI era
Choco is an AI platform for the food and beverage distribution industry, serving over 21,000 distributors and 100,000 buyers in the US, UK, Europe, and the Gulf region. It connects restaurants, suppliers, and distributors to handle ordering, sales, and customer management in one place.
As order volumes grew, the company hit a bottleneck: a huge number of orders still came in through email, text, voicemail, images, and even handwritten notes. Teams had to manually convert those into ERP orders — slow and hard to scale.
Narbeh Mirzaei, Choco's VP of Engineering, says the challenge isn't just processing input — it's the hidden context behind each order. Customer-specific SKU mappings, measurement preferences, delivery patterns — that knowledge used to live inside people's heads. Now it has to be encoded into a reasoning system that handles ambiguity.
As large language models matured, Choco began shifting from workflow software to AI systems that can directly execute tasks. The OpenAI API became the foundation for that transformation.
Product rollout path
Choco embedded the OpenAI API into the core of its platform, launching a new AI-native product called OrderAgent. It handles multimodal input — email, text, images, documents — and automatically converts them into structured ERP orders.
Mirzaei says transcription and information extraction provide the basic capabilities, but the real engineering challenge is building a dynamic context-learning infrastructure. The system has to resolve ambiguity by combining each customer's order history with their product catalog — moving from automation to actual intelligence.
On top of that, Choco launched VoiceAgent, built on OpenAI's Realtime API. Customers can place orders naturally over the phone with sub-second response times, even outside business hours.
The company says it chose OpenAI for model performance, multimodal capabilities, structured output, and production-grade reliability. Integrating text, vision, and voice into a unified ecosystem also lets them consolidate previously siloed workflows into a single intelligent system.
On the deployment side, Choco used the OpenAI SDK and API to quickly integrate speech-to-text, embeddings, and function calling. It built an evaluation system with real labeled data, continuous monitoring, and A/B testing to ensure accuracy in production.
For users, the product uses a frictionless integration strategy: customers don't have to change how they order — by phone, text, or email. The system adapts to them, not the other way around.
Implementation lessons
Choco shared several key takeaways:
First, set up an evaluation mechanism from day one. Even 10 to 20 samples can help validate iteration results.
Second, build observability for AI systems. Tracking model inputs, outputs, and reasoning traces becomes critical, compared to traditional logging.
Third, set expectations early. Since large models are probabilistic, not deterministic software, helping teams understand that difference builds trust and reduces deployment friction.
Next steps
Choco says it's continuing to expand AI capabilities across the food distribution ecosystem, letting agents handle more complex operational tasks.
The company is also exploring ways for non-engineers to become "agent orchestrators" — designing and managing intelligent systems that drive business outcomes.
Going forward, Choco plans to expand its use of the OpenAI API across sales, commercial, and supply chain operations, deploying more autonomous and context-aware systems. The goal is to keep pushing the company's transformation from workflow software to an AI-driven execution infrastructure.