Anti-gaming
Explicit, versioned, per-channel checks: falsified location, passenger-versus-driver, signal-denied, capture completeness. A GPS blackout cannot certify as a clean drive.
For insurers, reinsurers and platforms
A driver-risk signal you can defend to an actuary and a regulator.
Usage-based motor insurance has been permitted in India since 2022. What has been missing is a behaviour signal an underwriter can trust — explainable, anti-gaming, exposure-correct and consent-native. That is what the KYD engine produces, delivered through a consent-scoped partner API.
Four pillars of a priceable signal
Telematics data is easy. Telematics data an underwriter will accept is not.
Most smartphone risk scores fail on at least one of these four. A signal that fails any one of them cannot carry a premium.
Explainability
Six named dimensions with published weights and a per-driver feature breakdown. An actuary can decompose the number; a regulator can be shown how it was formed.
Honest abstention
Where the data will not support a dimension, the engine withholds it and renormalises rather than inventing a value — so a confident score means something.
DPDP-native consent
Scoped, expiring, revocable, audit-logged consent with driver-visible access history. Defensible data provenance, not a retrofitted checkbox.
Partner API · built
Consent-scoped. Raw location never leaves the platform.
A partner JWT scoped to INSURER_DATA_SHARING
serves exactly what underwriting needs and nothing more.
- Driver score and history — current KYD score, confidence interval, percentile band and trend.
- Feature breakdown — per-dimension values against published weights, including any dimension that abstained.
- Trip records — exposure, event rates and vehicle class, without raw location traces.
- Portfolio distribution — score distribution across a linked book, for pricing and monitoring.
- Quote-handoff verification — confirm that the driver who requested the quote is the driver whose score you read.
- Revocation is real — when a driver revokes, the scope closes. A legacy unscoped endpoint was found in our own audit and removed.
// Authorization: Bearer <partner JWT>
// scope: INSURER_DATA_SHARING
{
"uin": "4421••••8807",
"vehicle_class": "TWO_WHEELER",
"score": 732,
"confidence": { "ci95": [78.9, 85.1] },
"percentile_band": "p78-p84",
"exposure": {
"moving_minutes_90d": 6142,
"km_90d": 2841,
"trips_90d": 318
},
"breakdown": {
"speed_compliance": { "w": 0.25, "v": 88 },
"braking": { "w": 0.20, "v": 74 },
"acceleration": { "w": 0.15, "v": 81 },
"cornering": { "w": 0.15, "v": 79 },
"smoothness": { "w": 0.15, "v": 86 },
"context": { "w": 0.10, "v": null,
"status": "NOT_MEASURABLE" }
},
"weights_renormalised": true,
"anti_gaming": {
"falsified_location": "PASS",
"passenger_vs_driver": "PASS",
"signal_denied": "PASS",
"capture_completeness":"PASS"
},
"consent": {
"scope": "SCORE_READ",
"granted": "2026-05-30T11:04:22Z",
"expires": "2026-11-30T00:00:00Z",
"revocable": true
}
}Representative response shape for illustration. Schemas, scopes, rate limits and the sandbox are shared under a mutual NDA.
Where we are, honestly
The score is explainable and anti-gaming. It is not yet claims-validated.
We would rather say this on our own website than have an underwriter discover it in diligence. Score-to-claims correlation is the next proof, and it is the first milestone the current round funds.
Proposed validation study
Not started- Design co-signed with the insurer partner before any data moves.
- Tier 3 fidelity — smartphone plus dashcam plus GPS tag — on the study cohort.
- Correlate the KYD score and its per-dimension breakdown against real claims and incident experience.
- Publish the method to the partner, including where the score fails to separate risk.
Requires partner data and the fidelity layers above. This is what a first insurer proof-of-concept with Sakshya is actually about.
What is available today
- A live, production scoring engine with published weights and per-driver breakdowns.
- A built partner API with scoped JWTs, portfolio distribution and quote-handoff verification.
- DPDP-grade consent, export, erasure and driver-visible access logging — the parts procurement asks about.
- A driver base that can be scored without any hardware deployment on your side.
Frequently asked
What underwriting and procurement ask
Straight answers, including on the things we have not proved yet.
Ask us something elseIs Sakshya an insurer, broker or intermediary?
No. Sakshya is a driver-safety and risk-analytics platform. It does not sell, underwrite, solicit or advise on insurance products. It supplies an explainable driver-risk signal to licensed insurers, who make their own underwriting and pricing decisions under their own regulatory obligations.
What exactly does the partner API return?
A partner JWT scoped to insurer data sharing serves the driver's KYD score, score history, the per-dimension feature breakdown with published weights, trip-level records, portfolio-level score distribution, and quote-handoff verification. Raw location is never served. Every response reflects the driver's live consent state, and consent is scoped, expiring and revocable.
How does consent work when a driver is scored for one fleet and quoted by an insurer?
Consent is per-link, not global. A driver's linkage to a fleet and their linkage to an insurer are separate, individually scoped and independently revocable, each audit-logged. The driver initiates an insurer share themselves from the app when they want a quote, and can see every read of their data afterwards in their access history.
Has the score been correlated with claims experience?
Not yet. Score-to-claims validation is the single proof that converts pilots into contracts, and it is the first milestone the current pre-seed round funds: correlating the score against real claims and incidents using Tier 3 data, with a co-signed study design. Until that study exists we describe the score as explainable, anti-gaming and exposure-correct - never as claims-validated.
Does the regulatory position in India allow usage-based motor insurance?
Yes. India's insurance regulator formally permitted Pay-As-You-Drive and Pay-How-You-Drive motor products in a circular dated 5 July 2022. The constraint since then has not been permission - it has been the absence of a behaviour signal an insurer can trust and defend. That is the gap Sakshya is built for.
What makes a smartphone signal trustworthy enough to price?
Four things together, not one: orientation recovery so the physics is real regardless of where the phone sits; explicit per-channel anti-gaming so a falsified or signal-denied trip cannot certify as clean; honest abstention so the engine withholds a dimension rather than inventing a value; and DPDP-native consent so the data has a defensible provenance. A signal that fails any one of these is not priceable.
Start with one cohort and a co-signed study design.
The fastest path to a defensible answer is a small, well-specified proof of concept: a scoped consent flow, a defined cohort, and an agreed measure of whether the score separates risk in your book.
Talk to us about a POC
Read the technology
No cookies, no trackers, no third-party scripts on this site. Your enquiry reaches the founders directly.