AI Quote Automation for Freight & Logistics
The challenge:
A European logistics company handles over 20,000 quote requests per year — almost entirely via email. Each request arrives in a different format: some are plain text, others include Excel attachments, scanned PDFs, or images of shipping documents. The sales team manually reads every email, extracts origin, destination, cargo type, weight, and dimensions, looks up rates in Excel spreadsheets, and enters everything into their logistics ERP to generate a quote.
The process is slow, error-prone, and impossible to scale. High-value requests (sea freight, air cargo, customs) get buried alongside routine road transport quotes. Response times stretch into days, and there's no visibility into what's pending, what's been quoted, or what fell through the cracks.
What we designed:
A cloud-native quoting platform built entirely on Cloudflare Workers, with Claude as the AI extraction engine and a leading logistics ERP as the system of record.
The platform polls the company's Outlook mailbox via the Microsoft Graph API, feeds each email — attachments included — through Claude's multimodal capabilities, and extracts structured data: origins, destinations, cargo specs, incoterms, and special requirements. Claude uses MCP (Model Context Protocol) servers as tools to cross-reference extracted data against the ERP's customer registry and internal rate sheets in real time.
Every request is then routed into one of four operative queues based on confidence and complexity. Fully validated quotes go straight to the ERP with one-click approval. Requests with missing data trigger an automatic follow-up email to the client. High-value shipments (sea, air, customs) are flagged for manual review with all data pre-extracted. Unparseable emails land in a discard queue for human triage.
The sales team works from a web app (with a mobile companion) that gives them a unified view across all four queues — with bulk actions, inline corrections, and a feedback loop that improves extraction accuracy over time through a KV-stored correction memory.
Key capabilities:
Multimodal extraction — Claude processes plain-text emails, Excel attachments, scanned PDFs, and images of shipping documents in a single pipeline, outputting structured data via tool use
Live ERP validation — MCP servers query the logistics ERP's REST API and internal Excel rate sheets during extraction, so quotes arrive pre-validated against real customer and pricing data
Intelligent queue routing — Requests are automatically sorted into four operative queues (auto-approve, incomplete data, high-value, discards) based on confidence scoring and shipment type
Correction memory — Every manual correction the sales team makes is stored in Cloudflare KV, feeding back into future extractions to reduce repeat errors
Automated follow-ups — A cron trigger monitors unanswered requests and sends reminder emails after 48 hours, with full thread management so replies are re-linked to the original request
Client acceptance flow — After quote approval, a Cloudflare Pages link is sent to the client for confirmation, which triggers automatic pre-booking and tracking marker generation in the ERP
The result:
A platform designed to compress the quoting cycle from days to minutes, give the sales team full visibility into their pipeline, and free them from repetitive data entry — while keeping humans in the loop for complex decisions. The architecture scales linearly: more email volume simply means more queue items, not more manual effort.