Choco automates food distribution with AI agents
Summary (EN)
OpenAI published a case study on Choco, an AI-powered food and beverage distribution platform that says it now serves more than 21,000 distributors and 100,000 buyers across the US, UK, Europe, and the GCC. The update focuses on a concrete operations problem rather than a model launch: order intake still arrived through emails, texts, voicemails, images, and handwritten notes, forcing staff to manually translate unstructured requests into ERP-ready orders. According to the case study, Choco used OpenAI APIs to build OrderAgent, which processes multimodal inputs and converts them into structured orders, and VoiceAgent, which uses the Realtime API to handle phone orders with sub-second latency, including outside business hours. Choco says the harder problem was not just transcription, but resolving implicit context such as customer-specific SKU mappings, unit preferences, and delivery patterns that previously lived in the heads of order-desk employees. OpenAI says Choco selected its models for multimodal performance, structured outputs, and reliability, then paired them with evaluation datasets, monitoring, and A/B testing in production. The company reports operational results that include processing more than 8.8 million orders annually, reducing manual order entry by up to 50%, enabling 2x productivity gains, supporting round-the-clock order intake, and keeping error rates below configurable thresholds. The significance of the release is that it presents an application-layer example of AI automating messy supply-chain execution work, where value comes from embedding models inside operational workflows rather than from a standalone chatbot experience.
Summary (ZH)
OpenAI 发布了关于 Choco 的案例研究。Choco 是一家面向食品与饮料分销行业的 AI 平台，称其服务范围已覆盖美国、英国、欧洲和海湾地区的 21,000 多家分销商与 100,000 家采购方。这次更新并非模型发布，而是聚焦一个非常具体的运营问题，即订单仍然通过邮件、短信、语音留言、图片甚至手写纸条进入系统，导致团队必须把这些非结构化输入人工转换成 ERP 可处理的订单。根据案例，Choco 基于 OpenAI API 构建了 OrderAgent，可处理多模态输入并生成结构化订单；同时推出 VoiceAgent，利用 Realtime API 以亚秒级延迟处理电话下单，并可覆盖非营业时间。Choco 强调，真正困难的不只是转录，而是理解隐含业务上下文，例如客户专属 SKU 映射、计量单位偏好和配送习惯，这些知识过去主要掌握在订单处理人员脑中。OpenAI 表示，Choco 之所以选择其模型，是因为多模态能力、结构化输出和生产环境稳定性，并结合真实标注数据、持续监控和 A/B 测试进行落地。按案例披露，相关系统每年可处理超过 880 万张订单，使人工录单工作量最高下降 50%，团队生产率提升约 2 倍，并支持 24/7 持续接单，同时将错误率控制在设定阈值之下。这一案例的重要性在于，它展示了 AI 如何进入高度琐碎而关键的供应链执行流程，真正价值来自嵌入业务系统完成工作，而不是作为独立聊天界面存在。
Source