Email Triage Agent

The Problem

A commercial real estate firm received 400+ emails daily across property inquiries, tenant requests, vendor communications, and internal operations. Their two-person admin team spent most of their day just sorting and forwarding messages—leaving actual responses delayed by 24–48 hours.

Pain points

• Manual triage — Every email read and categorized by hand
• Context switching — Admins bouncing between email, CRM, and property management system
• Missed SLAs — Tenant maintenance requests often delayed beyond 24-hour target
• Knowledge silos — Only certain staff knew which vendor handled which property

The Intervention

We built an email triage agent that automatically classifies, routes, and in some cases responds to incoming messages:

Agent capabilities

1. Classification — Categorize emails into 12 types (inquiry, maintenance, lease, vendor, spam, etc.)
2. Entity extraction — Pull property IDs, tenant names, urgency signals
3. Smart routing — Forward to correct team member based on property assignment and expertise
4. Auto-response — Draft replies for common queries (office hours, application status, parking info)
5. Escalation — Flag urgent items (water leak, security issue) for immediate attention

Technical architecture

Inbox (Gmail) → Classification Agent → Router
                      ↓                   ↓
                Entity Extraction    CRM Lookup
                      ↓                   ↓
                Priority Score       Assignee Match
                      ↓                   ↓
                Draft Response ← → Forward to Agent

Integrations

• Email: Gmail API with OAuth
• CRM: HubSpot for contact/deal lookup
• Property management: Yardi API for tenant and property data
• Vendors: Webhook triggers to maintenance dispatch system
• Notifications: Slack alerts for high-priority items

The Outcome

Before → After:

• Avg. first response: 18 hours → 6 hours
• Maintenance SLA compliance: 62% → 94%
• Admin triage time: 4 hrs/day → 45 min/day
• Mis-routed emails: 23% → 4%

Qualitative wins

• Admin team freed up — Now focused on complex tenant relations instead of inbox management
• Tenants happier — Faster responses, especially for urgent maintenance
• Audit trail — Every classification and routing decision logged for review

Key Learnings

1. Start with classification — Get routing right before adding auto-response; wrong responses are worse than slow ones
2. Confidence gating — Agent only auto-responds when confidence > 0.85; otherwise drafts for human review
3. Feedback loop — Weekly review of mis-classifications continuously improved accuracy (78% → 91% over 8 weeks)
4. Integration depth matters — CRM lookup for sender context boosted classification accuracy 12%

Engagement type: Agent-in-a-Day prototype → 4-week production build
Timeline: 5 weeks total

This case study illustrates our capabilities with a representative scenario. Details have been generalized to protect client confidentiality.