CurbMind Ops
Mobile command center for dispatchers with agentic exception handling
The story
What I delivered
Last-mile delivery fails on exceptions: access issues, customer not available, traffic spikes, and dwell time delays. Dispatch teams often rely on manual triage across maps, chats, and dashboards.
This matters because last-mile is a major cost center. Multiple industry sources estimate last-mile can account for around 50%+ of total delivery cost.
Opportunity: shift from dashboards to an action-first agent system that detects risk, proposes options, routes approvals, and sends comms.
Audience: Dispatchers and fleet operations teams managing live delivery routes, exceptions, and SLA performance, plus drivers who need fast, clear route updates and instructions.
Detect Risk
Detects late risk early (telemetry + traffic + dwell signals)
Explaining
Explains root cause (why it is happening)
Define Trade-Offs
Proposes actions with trade-offs (stop swaps, reroutes, reassignments)
Gated Actions
Gates high-impact actions behind approvals
Automation
Automates comms to drivers/customers
Agent system
Research signals and numbers that shaped the design
Signal 1: Late deliveries are driven by exceptions and operational variability
Ops teams need systems that detect and handle exceptions quickly to protect on-time performance.
Signal 2: Last-mile is disproportionately expensive
Industry research commonly cites last-mile as ~50%+ of delivery cost, which makes small operational improvements meaningful.
Signal 3: Proactive communication reduces “where is my order” volume
Customer comms and accurate ETAs are repeatedly highlighted as key to reducing support burden.
Design implications
“Decision queue” over “monitoring dashboard”
Action gating (autopilot vs approval)
Explainability and trace view for trust
Comms studio baked into the workflow
Users
Dispatcher: triage, approve, resolve exceptions fast
Driver: receive route changes and clear instructions
Customer Support: proactive messaging, accurate ETAs