Score decay for inactive contacts

Without score decay, a contact who was highly engaged 18 months ago — opened five emails, clicked twice, visited the pricing page — retains a high score and continues surfacing as a hot lead. Sales receives contacts who have not interacted in over a year, the pipeline confidence drops, and the scoring model loses calibration. iso-25010:2011 functional-suitability.functional-correctness categorizes this as an accuracy defect: the score no longer reflects the contact's current engagement probability. Score decay is also an anti-sycophancy mechanism in the scoring model itself — it resists inflating lead quality based on historical activity that no longer predicts current intent.

Add a last_activity_at timestamp to contact records and run a periodic decay job (daily via Vercel cron or a scheduled BullMQ repeatable job).

// Runs daily via cron — e.g. src/jobs/score-decay.ts
async function applyScoreDecay() {
  const DECAY_PERCENT = 10         // Reduce score by 10% per inactivity period
  const INACTIVITY_DAYS = 30       // Apply after 30 days of no activity
  const cutoff = new Date(Date.now() - INACTIVITY_DAYS * 24 * 60 * 60 * 1000)

  await db.contact.updateMany({
    where: {
      lastActivityAt: { lt: cutoff },
      leadScore: { gt: 0 }
    },
    data: {
      leadScore: { multiply: (100 - DECAY_PERCENT) / 100 }
    }
  })
}

Both DECAY_PERCENT and INACTIVITY_DAYS should be named constants (not magic numbers) and referenced from your central scoring config. A last_activity_at field with no decay job referencing it does not satisfy this check.