EVVisorfdi-1.3.0

DADOS DE EXEMPLO · modo de apresentação. Os números são ilustrativos e não vêm de condutores reais.

Back to portal

Methodology

How the Driving Index works

The EVVisor Driving Index is a 0–100 score computed from the car's own motion data. Higher means lower observed risk. It is a behavioural risk score, not a claim probability and not a price: in your tariff it is one candidate rating factor, and its relation to your claim frequency is what a programme measures. The thresholds below are the ones the portal uses. Factor weights and bands are shared under NDA, with the validation plan.

From car signals to one index

  1. 01

    Car signals

    Tesla Fleet Telemetry streams longitudinal and lateral acceleration, speed, gear and time of day.

    about every 500 ms

  2. 02

    Events

    Threshold crossings in those signals become events.

    thresholds below

  3. 03

    Factor rates

    Events per 100 km; speeding as a share of driving time; late-night and short trips (under 5 km) as a share of km.

    six factors

  4. 04

    Normalised 0–100

    Each rate maps linearly between a best band (scores 100) and a worst band (scores 0).

    clamped

  5. 05

    Index

    Weighted sum of the six factor scores, rounded to a whole number.

    0–100

Event thresholds

  • Hard braking below −0.3 g, held for at least 300 ms, with the car's speed falling at 3.0 m/s² or more
  • Harsh acceleration above 0.3 g, with the speed rising at 2.0 m/s² or more
  • Cornering above 0.4 g lateral
  • Late night: 00:00–04:00 Europe/Istanbul
  • Speeding proxy: share of moving time at fine-risk speed (141–150 km/h half, 151+ full), until map-matching

Arithmetic

Factor score = 100 × (1 − clamp((measured − best) ÷ (worst − best), 0, 1))

Index = Σ (factor score × weight)

What the insurer gets, and what stays with the driver

Insurer receives

  • Driving Index

    0–100, per period

  • Factor breakdown

    Six factors, their weights and contributions

  • Data quality

    Km covered, samples per km, largest gap, confidence

  • Model version

    The exact model that produced each score

Stays with the driver

  • Routes

    Never leave the driver's account

  • Location history

    No raw location in any insurer view or API

  • Trip details

    Times, places and individual trips

Individual scores reach the insurer only with that driver's explicit consent. Everyone else counts in fleet totals only.

Governance

Model version

fdi-1.3.0

Every score is stored with the version of the model that produced it, so a historical quote can be re-run exactly.

Append-only scores

A new model version writes new rows. Existing scores are never overwritten.

Confidence levels

  • Provisional

    Fewer than 2 samples per km, or under 50 km covered. Directional only.

  • Standard

    Enough data, but a gap longer than 1 minute inside a trip.

  • High

    At least 50 km, 2 or more samples per km, and no gap longer than 1 minute inside a trip.

Known limitations

  • Seven cars are not actuarial evidence. The pilot shows the method works; it does not prove a link to claims.
  • Speeding is a proxy: share of moving time at a speed that can be fined in Turkey, until map-matching adds posted limits. 141–150 km/h counts half (fined on state motorways), 151 km/h and above in full (legal on no road). It undercounts urban speeding.
  • Normalisation bands and weights are provisional and will be tuned on real data.

This is the model that scores pilot drivers. Its bands and weights are starting points: tuning them on real pilot telemetry ships as a new version, and scores from this version stay on record.

EVVisor computes the index. Mapping it to premium is your actuarial team's decision.