Logistics · 2025
Agentic document classification, GDPR-compliant by design
A multi-agent pipeline that classifies high volumes of operational documents for a UK logistics operator, with personal data protected at every step.
- Client
- Confidential (UK logistics operator)
- Services
- Applied AI · Multi-agent systems · Compliance engineering
- Status
- In production
The challenge
Our client handled thousands of operational documents every week: delivery records, customer correspondence, claims and exception reports. Nearly all of it contained personal data, and nearly all of it was sorted by hand. Automation was the obvious answer, but sending customer data to third-party AI services was a non-starter under GDPR.
Our approach
We built the system as a set of cooperating agents rather than a single model. An ingestion agent normalises incoming documents from email, scans and portal uploads. A redaction layer strips personal identifiers before anything reaches a language model, so classification decisions are made on minimised data. A routing agent assigns each document a type, priority and destination, and an audit agent records every decision with the evidence behind it.
Anything the system is not confident about lands in a human review queue instead of being guessed at. All processing stays within UK infrastructure, and the audit trail is designed to be handed to a data protection officer, not just a developer.
The outcome
Documents that took hours to reach the right team now arrive in seconds. The review team works only on the genuinely ambiguous cases, and the business has a complete, inspectable record of how every document was handled. The architecture is now the template for further automation across the client’s operations.