Risk Assessment & Premium Calculation

Weekly premium is computed per rider from zone history, weather forecast, seasonal patterns, social disruption risk, and individual claim frequency. Strictly weekly — computed every Sunday, covers Monday through Sunday.

Premium Formula

premium = BASE_PREMIUM + min(risk_adjustment, MAX_PREMIUM - BASE_PREMIUM)

risk_adjustment = (risk_from_events + risk_from_forecast + risk_from_social)
                  × seasonal_multiplier
                  × claim_frequency_multiplier

Constants:

Constant	Value
BASE_PREMIUM	₹49
MAX_PREMIUM	₹199
RISK_FACTOR_PER_EVENT	₹12
FORECAST_WEIGHT	₹20
WEEKS_LOOKBACK	4

---

Seasonal Risk Multiplier

India’s weather follows strong seasonal patterns that directly affect delivery disruption frequency. The premium calculator applies a calendar-month seasonal multiplier to the weekly premium (billing stays weekly per use case; this is not monthly pricing).

Months	Multiplier	Reason
Jun – Sep (Monsoon)	1.40×	Peak rain, flooding, waterlogging
Oct	1.15×	Post-monsoon cyclones, residual rain
Nov	1.15×	Cyclone season (Bay of Bengal)
Dec – Feb (Winter)	0.85×	Mild weather, low disruption risk
Mar (Transition)	1.00×	Baseline
Apr – May (Pre-monsoon)	1.25×	Extreme heat waves across north India

const SEASONAL_RISK_MULTIPLIER: Record<number, number> = {
  0: 0.85,  // Jan
  1: 0.85,  // Feb
  2: 1.0,   // Mar
  3: 1.25,  // Apr
  4: 1.25,  // May
  5: 1.4,   // Jun
  6: 1.4,   // Jul
  7: 1.4,   // Aug
  8: 1.4,   // Sep
  9: 1.15,  // Oct
  10: 1.15, // Nov
  11: 0.85, // Dec
};

---

Social Risk Factor

Social disruption (strikes, curfews, lockdowns) is tracked separately from weather. The getSocialRiskFactor function queries live_disruption_events of type social within a rider’s zone over the past 4 weeks:

socialRiskFactor = min(1.0, socialEventCount / 5)

Social events (4 weeks)	Risk factor
0	0.0
1	0.2
3	0.6
5+	1.0 (cap)

This ensures riders in protest-prone zones pay a fair premium reflecting their actual risk.

---

Claim Frequency Multiplier

Riders with frequent recent claims represent higher risk. The premium calculator counts claims from the past 4 weeks:

claimFreqMultiplier = 1.0 + min(0.2, claimCountLast4Weeks × 0.04)

Claims (4 weeks)	Multiplier
0	1.00×
2	1.08×
5	1.20× (cap)
10	1.20× (cap)

The 0.2 cap ensures that even high-claim riders don’t see premiums spike beyond 20% above base due to this factor alone.

---

Historical Event Count

The premium calculator queries live_disruption_events for the past 28 days. For each event, it checks whether the event’s geofence overlaps with the rider’s zone using isWithinCircle():

export async function getHistoricalEventCount(
  supabase,
  zoneLatitude?: number,
  zoneLongitude?: number
): Promise<number> {
  const since = new Date();
  since.setDate(since.getDate() - WEEKS_LOOKBACK * 7);

  const { data } = await supabase
    .from("live_disruption_events")
    .select("id, geofence_polygon")
    .gte("created_at", since.toISOString());

  let count = 0;
  for (const ev of data) {
    const gf = ev.geofence_polygon;
    if (!gf?.lat || !gf?.lng) {
      count++;
      continue;
    }
    if (isWithinCircle(zoneLatitude, zoneLongitude, gf.lat, gf.lng, gf.radius_km ?? 10)) {
      count++;
    }
  }
  return count;
}

Zones without explicit geofence data (e.g., citywide curfews) count toward every rider’s risk score.

---

Forecast Risk Factor

A 0–1 multiplier derived from Tomorrow.io’s 5-day hourly forecast for the rider’s zone. Any forecast hour meeting a trigger threshold (temperature ≥ 43°C or precipitation ≥ 4 mm/h) increments the trigger counter:

export async function getForecastRiskFactor(
  _supabase, lat: number, lng: number
): Promise<number> {
  const hourly = await fetchTomorrowForecast(lat, lng);
  let triggerHours = 0;
  for (const interval of hourly) {
    if (interval.temperature >= 43 || interval.precipitationIntensity >= 4) {
      triggerHours++;
    }
  }
  return Math.min(1, triggerHours / hourly.length);
}

---

Example Calculations

Zone history (4w)	Forecast	Social events	Claims (4w)	Month	Premium
0 events	0.0	0	0	Nov	₹49 (base)
3 events	0.2	1	2	Mar	₹89
5 events	0.5	3	4	Jul (monsoon)	₹175
8+ events	0.8	5	10	Aug (monsoon)	₹199 (cap)

---

Technical reserve load

Premiums apply PREMIUM.RESERVE_LOAD (currently 2%) after core expected-loss math in both the legacy weekly formula and the dynamic engine. This funds technical reserve narrative (IBNR-style lag, tail correlation) and aligns with Policy §10. It is not a separate rider-visible line item — it is baked into the quoted weekly premium before clamping to PREMIUM.BASE–PREMIUM.MAX.

Dynamic engine: raw_premium = expected_loss × (1 + margin + safety_buffer) × (1 + RESERVE_LOAD).

Reinsurance (quota share, catastrophe XL) is described at contract level in Policy §10; it is not modeled in application code in the current release.

---

Plan Tiers

After the premium is calculated, riders choose from three flat-rate plans. The dynamic calculation informs the recommendation, but riders can select any tier:

Plan	Weekly Premium	Payout Per Claim	Max Claims/Week
Basic	₹49	₹300	1
Standard	₹99	₹700	2
Premium	₹199	₹1,500	3

The plan chosen is stored as weekly_policies.plan_id referencing plan_packages.id.

---

Cron: Weekly Premium Recommendations

Every Sunday at 17:30 UTC, /api/cron/weekly-premium runs:

1. Fetches all active policies where week_end_date < today.
2. Sets is_active = false for expired policies.
3. For each rider with an expired policy, recalculates the premium:
   • Fetches historical events, forecast risk, and social risk per zone (cached per zone).
   • Queries the rider’s claim count over the past 4 weeks.
   • Applies seasonal multiplier for the current month.
   • Passes all factors to calculatePremiumWithLlm for final recommendation.
4. Stores the result in premium_recommendations with risk_factors JSONB:

   {
     "historical_events": 3,
     "forecast_risk": 0.25,
     "social_risk": 0.4,
     "claim_count_4w": 2,
     "seasonal_multiplier": 1.4
   }

Riders must manually re-subscribe each week — automatic renewal would require a recurring billing integration (for example Razorpay Subscriptions) on top of the current weekly Checkout flow.

---

Next-Week Prediction (Admin Dashboard)

The admin analytics panel shows a predicted claims range for the coming week, calculated by lib/ml/next-week-risk.ts:

With Tomorrow.io API key (primary path):

1. Discover all active rider zones from weekly_policies + profiles.
2. Deduplicate zones within ~11 km of each other.
3. For each zone in parallel:
   • Fetch 5-day hourly weather forecast (Tomorrow.io) — count hours above heat/rain thresholds.
   • Fetch 5-day AQI forecast (Open-Meteo) — count hours above AQI 150.
4. Aggregate trigger hours and risk types across all zones.
5. Factor in active policy count and severity weight.
6. Return a low–high range, risk level, AQI risk note, and zones checked count.

Historical fallback (no API key):

1. Query parametric_claims for the past 21 days.
2. Calculate the weekly average.
3. Apply a linear trend (week 1 vs. week 3 claim rate).
4. Return avg ± 2 as the range.

interface NextWeekPrediction {
  expectedClaimsRange: string;   // e.g. "8–14"
  riskLevel: "low" | "medium" | "high";
  source: "forecast" | "historical";
  details?: string;
  aqiRisk?: string;              // e.g. "12h of poor AQI across 3 zones"
  zonesChecked?: number;         // number of zones analyzed
}