Smart Dispatch & Routing Available · opt-in per tenant

Dispatch, recommended — never replaced.

Google routing on every booking. AI-recommended driver matching with dispatcher approval — opt-in per tenant. Claude Haiku primary, OpenAI fallback, deterministic rule-based fallback.

The two layers

Always-on routing. Opt-in intelligence.

Google routing runs on every booking, every tenant, every time. Smart Dispatch — the AI layer — is opt-in: enable it per company, kill-switch in one boolean flag.

Layer 1 · Always on ✓ Default

Google Routes & predictive ETAs

Every booking gets a Google Routes API ETA, with Haversine fallback if Routes is unavailable. Anticipates traffic, congestion, and pickup-zone pressure before dispatch — not after.

  • Real-time ETAs · Google Routes API
  • Haversine fallback · resilient to outages
  • Working-pattern templates assigned per driver
  • No tenant flag required · runs by default
Layer 2 · Opt-in companies.ai_dispatch_enabled

Smart Dispatch — AI driver recommendations

Available per tenant. Claude Haiku scores driver candidates across 8 weighted criteria, returns a ranked recommendation with reason text. OpenAI is the fallback provider; deterministic rule-based scoring is the fallback to AI itself. Dispatchers always confirm.

  • Claude Haiku · OpenAI · rules-based fallback chain
  • 8-criteria driver scoring (vehicle, distance, ETA, duty, rating, cost …)
  • Every recommendation logged with confidence + reason
  • Kill-switch per tenant via single boolean flag
Auditable by design

Every recommendation logged. Every outcome tracked.

Smart Dispatch isn't a black box. We persist the full decision trail to dispatch_recommendations, and the realised outcome to dispatch_outcomes— so you can audit what was suggested, what the dispatcher chose, and how the trip actually went.

dispatch_recommendations · row decision trail
{ "booking_id": "BKG-9421", "model_name": "claude-haiku-4-5-20251001", "driver_id": "DRV-184", "confidence": 0.91, "reason": "Closest qualifying driver on duty; 5★ corporate rating; cost in band.", "latency_ms": 312, "prompt_tokens": 428, "completion_tokens": 91, "accepted": true, "created_at": "2026-04-14T09:01:48Z" }
dispatch_outcomes · row ML feedback loop
{ "booking_id": "BKG-9421", "recommendation_id": "REC-87421", "completed": true, "arrival_delta_s": -42, "passenger_rating": 5, "no_show": false, "complaint": false, "captured_at": "2026-04-14T09:56:14Z" }
Numbers from the field

Fewer slips. Tighter ETAs. Calmer ops.

Smart Dispatch shipped April 2026. Numbers from the active fleet — direction varies by region, service type, and tenant configuration.

312ms
Avg dispatch latency
across routed bookings · 2026
94%
On-time pickups
across active fleet · 2026
2.1%
No-show rate
across active fleet · 2026
99.9%
Driver utilisation
across active fleet · 2026
Ride along

Sit with our dispatchers for an hour.

The fastest way to see what Smart Dispatch actually does. We'll walk you through dispatch_recommendations on a real day's traffic — including the rejections.

Request a demo Read the architecture