An autonomous AI agent built on AWS Strands Agents SDK and deployed to Amazon Bedrock AgentCore for Europe's largest credit management company. The agent reads every inbound email, classifies it, creates a structured case, and routes it to the correct team queue. FCA-compliant. Zero manual triage.
Thousands of inbound customer emails each month, every one read and manually sorted by operations staff before it reached the right team. Skilled people spending hours on clerical sorting instead of resolving customer issues.
Emails needed to reach the correct queue across 13 different teams, each handling distinct case types. Misrouted emails meant delays, re-work, and missed SLAs.
As an FCA-regulated business operating under Consumer Duty, every email classification decision needed to be auditable with full reasoning. Vulnerability indicators had to be flagged automatically.
The operations team used Microsoft 365 and Amazon Connect Cases daily. Any solution had to work with existing systems without requiring agent retraining or process changes.
A Strands Agent deployed to Amazon Bedrock AgentCore Runtime. The agent autonomously reasons about each email and chooses its own execution path. It is not a hardcoded pipeline.
Inbound emails from the Microsoft 365 enquiry mailbox arrive via Amazon SES and are stored in S3. The agent extracts sender, subject, body, and all attachments (PDFs, Word, Excel, images, Outlook items) with full content extraction.
The agent reasons about each email and decides how to handle it. For emails with document attachments, it analyses the attachment content to determine specialist routing. It self-audits every classification before acting on it.
The agent creates a fully populated Amazon Connect Case with typed fields covering category, priority, SLA deadline, team, vulnerability flags, and recommended next actions. Low-confidence classifications are routed to a supervisor queue for human review.
Every classification decision is logged with full reasoning, confidence score, and routing outcome for regulatory retention. Each case is routed to the correct team queue. Human agents open a pre-triaged, pre-populated case and start resolving immediately.
Strands Agent on Bedrock AgentCore Runtime. Serverless, session-isolated, with built-in observability.
Strands Agent deployed to AgentCore Runtime. Serverless execution with session isolation, built-in observability, and consumption-based pricing.
Claude foundation models via Bedrock. The agent uses a domain-tuned system prompt with self-audit verification on every classification.
Secure email intake with long-term archival. Every email stored before processing for FCA retention compliance.
Event-driven classification pipeline. Deduplication, spam detection, and low-confidence flagging built in.
Structured cases with typed fields, routed to the correct team queue. Linked tasks as fallback. Customer profiles auto-created.
Every classification decision logged with reasoning, confidence score, and routing outcome. Long-term retention for regulatory compliance.
Infrastructure as code across three environments. Repeatable, auditable deployments.
The agent correctly classifies the large majority of inbound emails. Low-confidence decisions are automatically escalated to the supervisor queue with full reasoning visible.
Half of all manual email sorting eliminated. Operations staff now spend their time resolving customer issues instead of deciding where emails should go.
The agent calculates and sets SLA deadlines per category, with shorter deadlines for regulated request types and standard timing for routine enquiries. SLA deadline is written as a typed field on every case, enabling real-time SLA monitoring across all team queues.
The agent detects vulnerability indicators at multiple severity levels and flags them on the case automatically. Regulated request types are surfaced for priority handling. Every decision is logged with full reasoning to a long-retention audit trail for FCA Consumer Duty compliance.
Every inbound email processed in seconds. No backlogs. Urgent items surface immediately with the correct SLA and priority. Volume spikes handled without additional headcount.
The agent reads and analyses attachments across common document types (PDFs, Word, spreadsheets, images, Outlook items). The document content drives the routing decision, not just the email text, so specialist cases land with the right team automatically.
We deploy AI email agents in 4 weeks. No platform migration. No agent retraining.
Book a Discovery Call