The Problem
A Berlin-based logistics tech firm specialising in EU-China trade lanes was processing ~3,200 shipments monthly. Each shipment required documentation: commercial invoice, packing list, bill of lading, certificate of origin, customs declarations for both origin and destination jurisdictions. Documentation was prepared by a 7-person ops team that worked across time zones to keep pace. Errors in customs declarations triggered hold-ups (sometimes days at a port) and penalty fees.
The CEO had a hard target: drop cost-per-shipment by 50% in 12 months without compromising compliance.
What I Built
1. Document extraction + cross-validation. Inbound shipping data lands in many formats. Excel from manufacturers, PDFs from freight forwarders, narrative emails from partners. AWS Textract + Claude extract structured data; cross-validation logic catches mismatches (e.g., quantities on the packing list don't reconcile to the commercial invoice).
2. Automated customs declaration generation. The system generates HS-coded customs declarations for both origin and destination jurisdictions. For German customs (ATLAS), filings are submitted directly via API. For other jurisdictions, the system produces submission-ready packets.
3. Exception triage. When a shipment hits an exception (port hold, missing document, classification dispute, payment issue), Claude classifies the exception type, drafts the resolution path, and either auto-resolves (92% of common exception types) or routes to the right human with full context.
4. Customer-facing status automation. Shippers receive proactive milestone updates: documents ready, customs cleared, shipment dispatched, ETA changes. No more "where's my shipment?" emails.
Outcome
The CEO's 50% cost-per-shipment target was beaten. final figure was 64%. Customs hold incidents dropped from a daily occurrence to a weekly one. Penalty fees fell by €340K annualised. The firm took on 2 new lane partnerships in 2025 because they could now offer pricing competitors couldn't match. and they did it without growing the ops team.