The challenge
The operator needed a faster way to understand why shippers reduced volume or declined renewals. Collecting comments was easy; reading open replies, grouping the real causes and writing a summary the commercial team could act on was the bottleneck. A fixed survey was not enough: “transit is slow” might mean a missed vessel, a congested warehouse or a pricing dispute, and account managers needed a relevant follow-up before the answer went stale.
What we built
A LangChain workflow that runs each campaign from brief to report. The team writes a brief (target accounts, lanes or warehouses in scope, questions to explore); the agent turns it into a short questionnaire, opens the conversation on WhatsApp and asks one follow-up per reply that builds on what the shipper just said. When collection closes, GPT-4.1 analyses the threads and returns recurring themes, a sentiment split and the quotes behind each theme.
The result
Two engineers shipped the system to production in five weeks, through four conversation-design iterations that tightened how the agent follows a reply and holds the brief. Campaign setup, WhatsApp collection and reporting now sit in one workflow, and every finding links back to the source quotes.