Propose
On a periodic cadence, the platform runs AI-assisted iterations that propose new risk-feature candidates and rebalance weights.
Solutions · EV fleets
Legacy motor insurance reprices a fleet once a year. YAS reads each EV — battery health, position, behaviour — and turns it into a live risk score your insurer can price against, so coverage tracks how the fleet actually drives.
Live in production
The underwriting gap
A policy set at renewal rates on vehicle age, demographics and history, then holds for twelve months while the risk underneath it changes every trip. With no read on how the fleet actually drives, the insurer prices the average — safe fleets subsidise risky ones, and capital sits idle against risk no one can see. YAS scores each trip as it happens, so an insurer can price the risk that's actually there — and free the capital held against the risk that isn't.
Renewal variables — vehicle age, demographics, history — can't follow a fleet whose risk changes trip to trip. The policy is stale the day after it binds.
How a fleet actually drives — speed, braking, exposure, where and when — predicts loss better than any static factor. Tracked live, it becomes a price an insurer can set in real time.
Price the risk that's actually there and capital stops sitting idle against the risk that isn't — tighter combined ratios, and lower premiums for the fleets that earn them.
Data pipeline
A production-grade pipeline already validated at taxi-fleet scale. The same architecture is portable to autonomous vehicles, robots, and drones.
Telemetry from every vehicle, preprocessed at the edge. An in-vehicle compute node handles raw signals — position, motion, timing, sensor state — with local buffering and sanity checks.
Risk dimensions
Every enriched road segment carries dozens of properties — the signals static underwriting ignores, and where a behaviour-and-context risk model earns its edge. We group them into three questions a static rating table can't answer.
Where is the vehicle?
What is it driving on?
How is it moving?
Coverage scope
From battery cells to third-party liability — every layer reads from the same signal YAS scores.
Pack-level State of Health and State of Charge, tracked per vehicle. The policy adjusts as the battery degrades — no manual reassessment, no mid-term re-underwriting.
Premium accrues only while the vehicle is operating — coverage activates on ignition, pauses when parked. The fleet stops paying for idle assets.
Depot chargers, cables, and grid connection points. Equipment damage, downtime, and third-party incidents at charging sites sit inside the policy.
Collisions, property damage, and passenger incidents from EV operations — adjudicated against the same telemetry that scored the trip.
Acceleration, hard braking, cornering, regen patterns. Per-driver scores feed the insurer's dynamic premium adjustment and a 90-day claim-probability flag.
A Fleet Protection Score across every vehicle — risk zones, claims status, and coverage state at a glance, from the same API that scored the policy.
Model evolution
The risk models aren't static. AI proposes improvements; backtesting decides what ships.
On a periodic cadence, the platform runs AI-assisted iterations that propose new risk-feature candidates and rebalance weights.
Every candidate is backtested against historical data and checked for stability, explainability, segment fairness, and distribution constraints.
Only updates that clear the guardrails ship to production. Everything else is logged and parked — no silent changes to live pricing.
Proof in production
JOIE's electric taxi fleet streams telematics into YAS. ARIA has scored more than 10 million kilometres of commercial EV driving — the risk signal behind a Zurich-underwritten motor product in Hong Kong.
Read the JOIE case study
Hong Kong's first behaviour-based taxi fleet — 1,000 EV cabs scored live by YAS, priced on how each driver actually drives.
Questions
Scope a deployment with our team. Most fleets are live in weeks, not quarters.