Choco Supercharges Food Supply Chains with OpenAI

Choco, a major player in food and beverage distribution, is revolutionizing its operations by integrating OpenAI's powerful APIs. The platform connects over 21,000 distributors and 100,000 buyers globally, but faced a significant hurdle: orders arrived via a chaotic mix of emails, texts, voicemails, and even handwritten notes.

Previously, Choco's teams manually transcribed these varied inputs into structured ERP orders. This process was slow, error-prone, and a major bottleneck for scaling. "The real problem was implicit context: customer-specific SKU mappings, unit preferences, delivery patterns. That knowledge lived in the heads of order desk reps," explains Narbeh Mirzaei, VP Engineering at Choco. The company saw an opportunity with production-ready LLMs to build AI systems capable of direct work execution, making the Choco OpenAI integration a strategic imperative.

AI Agents Tackle Order Chaos

Choco embedded OpenAI APIs at the core of its platform, introducing products like OrderAgent. This AI agent processes multimodal inputs—emails, SMS, images, and documents—converting them into structured, ERP-ready orders. This moves beyond simple transcription to intelligent interpretation.

"The real engineering challenge was building dynamic in-context learning infrastructure, so the system resolves ambiguity against each customer's ordering history and catalog," Mirzaei noted. This capability is key to truly automating order processing with AI agents, a challenge also being explored by IBM Experts on Building With AI Agents.

VoiceAgent, another new product, uses OpenAI's Realtime API for sub-second latency voice orders, allowing customers to order naturally over the phone, even outside business hours. This dual approach addresses the critical need for AI in food supply chain automation.

Tangible Results and Lessons Learned

The implementation was rapid, utilizing OpenAI's SDKs and APIs for seamless integration of speech-to-text, embeddings, and function calling. Choco built a rigorous evaluation framework to ensure accuracy.

The results are striking: over 8.8 million orders processed annually, eliminating millions of manual workflows. Manual order entry has been reduced by up to 50%, freeing teams for higher-value tasks. Sales team productivity has doubled without added headcount, and error rates are maintained below 1–5%.

Key leadership lessons include starting evaluation from day one, investing in AI-native observability for debugging, and setting clear expectations about the probabilistic nature of LLMs to build trust.

Choco plans to deepen its use of OpenAI APIs, enabling non-engineers as "agent orchestrators" to manage these intelligent systems. This marks a significant shift from traditional workflow software to AI-powered execution infrastructure.