AI-linked taxi insurance in Hong Kong: Zurich and YAS turn driving behavior into coverage
Zurich Insurance (Hong Kong) and YAS Digital are tying taxi insurance coverage to real driving behavior using AI. The goal is simple: move beyond static proxies like age or vehicle type and reward safer habits in a high-frequency, high-severity segment.
The program analyzes driving data to produce daily performance scores, reviewed quarterly. Drivers are placed into tiers; higher scores unlock broader personal accident coverage and extra benefits, with options that can extend protection to family members.
How it works
The model runs on YAS's TAXY platform, combining in-vehicle systems, smartphone sensors and real-time analytics. Over a six-month pilot ending in September, it processed roughly 39 million data points and delivered a 20% improvement in participant safety scores, according to Zurich.
"Data is analysed to generate daily scores, which will be reviewed quarterly," said Eric Hui, CEO at Zurich Insurance for Greater China. "Drivers can receive corresponding insurance coverage and rewards based on their quarterly performance score."
What feeds the score
- Harsh braking and acceleration patterns
- Speeding relative to legal limits
- Turning stability and lateral control
- Distracted driving signals
- Accident black spots and environmental risks along each journey
- Contextual factors such as weather, traffic patterns and identified danger zones
Data can be captured via government-mandated in-car cameras or the YAS mobile app (smartphone sensors + GPS). Models are refined continuously as new data lands.
Why it matters for insurers
This is usage-based insurance applied to a dense, price-compressed line. Behavior creates a clearer segmentation signal than broad demographics, letting carriers distinguish safer operators and map benefits accordingly.
Done well, these programs can influence frequency, improve loss cost predictability and create an on-ramp for telematics-led product design. Zurich's early read: the approach can modernize underwriting in a taxi segment where risk is elevated and differentiation has been limited.
Guardrails: fairness and transparency
Zurich and YAS emphasize objective indicators over personal traits, with regular audits of outcomes and human oversight before any tier change. Drivers get transparency tools to understand scores and what to improve.
For carriers, this lines up with good practice: document features, monitor drift, test for bias, and give policyholders clear recourse. For local context, see the Hong Kong Insurance Authority's sandbox framework for InsurTech experimentation here and privacy guidance from the Office of the Privacy Commissioner for Personal Data here.
Operational value beyond pricing
The platform surfaces route patterns, downtime and working hours so drivers can cut wasted time and lift daily earnings. Fleet operators can optimize dispatch, coverage and vehicle utilization off the same data, said William Lee, co-founder of YAS.
Product and portfolio implications
- Product design: Tier coverage and benefits to behavior; keep rules simple, visible and achievable. Start with personal accident enhancements before more complex structures.
- Underwriting: Use behavior scoring as a supplemental risk signal; track lift versus existing rating factors. Measure impact on frequency, severity and near-miss indicators.
- Claims: Use trip context to support FNOL accuracy and liability assessment while keeping strict access controls and purpose limits.
- Compliance: Maintain feature explainability, consent flows and data retention policies. Schedule periodic fairness reviews and human checks, mirroring Zurich's approach.
- Change management: Give drivers a feedback loop (what changed, why, and how to improve next quarter). Incentivize progress, not perfection.
- KPIs to watch: Safety score distribution, quarterly migration across tiers, collision frequency per million km, alert rates (harsh events, speeding), claims cost per vehicle, and earned-benefit utilization.
What's next
Both firms see taxis as a proving ground for broader telematics-led underwriting. As usage-based products expand, the data could inform future underwriting considerations across other fleet segments, including those affected by ride-hailing regulation.
Future phases may link to digital commerce platforms and EV charging networks-useful for managing cost-of-operations and driver loyalty in electrified fleets. "Hong Kong's taxis are the heartbeat of the city," Hui said. "Using AI and smart telematics allows us to safeguard livelihoods while contributing to a more sustainable and resilient transport ecosystem."
Practical steps if you plan a pilot
- Start with a single segment (e.g., taxis) and define a minimal feature set that correlates with collision risk.
- Adopt a daily score with quarterly review to reduce noise and gaming.
- Tie benefits to measurable improvement, not demographics; publish a simple scorecard for drivers.
- Run independent audits of model outcomes and add human approval before tier changes.
- Integrate driver feedback loops into your app or portal; coach behaviors with small, specific nudges.
- Instrument a test-and-learn rhythm: weekly model monitoring, quarterly program reviews, and clear success criteria.
"Behavioural data is used to improve safety, not to punish drivers," Hui noted. "Driving behaviour helps us classify risk tiers so we can reward safer habits with better benefits, but we are mindful of keeping the system fair."
Further reading and upskilling
If you're building internal AI capability for pricing, claims or telematics programs, you may find curated training by role useful: AI courses by job.
Your membership also unlocks:
