Intelligent Email Classification and Routing for Financial Services

The organization received a large volume of emails across customer service and operational inboxes, requiring employees to manually review, classify, and forward messages to the appropriate teams.

The main challenges included saturated shared inboxes, manual email triage, incorrect or delayed routing, difficulty identifying customer intent, repeated review of ongoing conversations, limited visibility into pending requests, and SLA breaches caused by slow initial classification.

The organization needed an automated solution capable of understanding each incoming email in the context of the broader conversation rather than treating every message as an independent request.

Rootstack implemented an end-to-end AI-powered email processing workflow using Claude, n8n, PostgreSQL, Python, REST APIs, webhooks, Docker, and OpenTelemetry, integrating with the organization's service desk, CRM, and enterprise systems.

The workflow followed the sequence: Email Intake → Thread Identification → Context Retrieval → Intent Classification → Data Extraction → Priority/SLA Determination → Ticket Creation or Update → Routing.

A key component was a custom conversation-threading layer built with PostgreSQL. Instead of treating every incoming message independently, the system identified whether an email belonged to an existing conversation and connected previous messages, replies, forwarded emails, customer requests, and existing tickets. This allowed Claude to evaluate new messages using relevant conversation history and distinguish between new requests, follow-ups, clarifications, complaints, status inquiries, and responses to existing cases.

The solution also extracted customer identifiers, account information, reference numbers, transaction details, dates, request descriptions, and urgency indicators. This information was normalized and used to automatically create or update tickets, assign queues, set priorities, and trigger follow-up workflows.

Intelligent routing applied business rules based on intent, customer type, product or service, request category, priority, SLA requirements, and existing ticket ownership. n8n orchestrated the complete workflow, while SLA rules helped prioritize requests approaching or exceeding response targets. The modular architecture also enabled new email categories and routing rules to be added incrementally.

The implementation was completed in 5 months and achieved an estimated 70% reduction in email triage time, with 85–90% of emails automatically classified and an estimated 75% reduction in manual routing. Duplicate ticket creation was reduced by an estimated 60%, SLA compliance improved by approximately 25%, and the overall operational workload was reduced by an estimated 50%. The solution operates 24/7 and was deployed incrementally by email category.