mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-19 07:49:34 +00:00
123 lines
3.8 KiB
Python
123 lines
3.8 KiB
Python
"""
|
|
Churn prediction — flags customers likely to churn within 60 days.
|
|
|
|
Inputs are signals from the Revenue Memory + Customer Success layer:
|
|
- days_since_last_login
|
|
- drop in monthly engagement
|
|
- support ticket spike
|
|
- billing issues
|
|
- low NPS
|
|
- drop in pipeline added by Dealix
|
|
|
|
Each signal gets a weight; the composite score is mapped to a band.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass, field
|
|
from typing import Any
|
|
|
|
|
|
@dataclass
|
|
class ChurnPrediction:
|
|
customer_id: str
|
|
score: float # 0..1 — higher = more likely to churn
|
|
band: str # safe / watch / at_risk / critical
|
|
drivers: list[str] = field(default_factory=list)
|
|
recommended_action_ar: str = ""
|
|
confidence: float = 0.7
|
|
|
|
|
|
def predict_churn(
|
|
*,
|
|
customer_id: str,
|
|
days_since_last_login: int = 0,
|
|
monthly_engagement_drop_pct: float = 0, # 0..1 (drop vs prior month)
|
|
support_tickets_open: int = 0,
|
|
billing_failures_last_90d: int = 0,
|
|
nps: int | None = None,
|
|
pipeline_added_drop_pct: float = 0, # 0..1 (drop vs prior month)
|
|
months_as_customer: int = 6,
|
|
) -> ChurnPrediction:
|
|
"""
|
|
Compute churn probability + drivers + recommendation.
|
|
|
|
Weights tuned to the early-pilot cohort. As more data flows in,
|
|
these can be re-fit by the AI Quality module.
|
|
"""
|
|
score = 0.0
|
|
drivers: list[str] = []
|
|
|
|
# Engagement: huge weight
|
|
if days_since_last_login > 30:
|
|
score += 0.35
|
|
drivers.append(f"لم يدخل المنتج منذ {days_since_last_login} يوم")
|
|
elif days_since_last_login > 14:
|
|
score += 0.20
|
|
drivers.append(f"دخوله متباعد ({days_since_last_login} يوم)")
|
|
|
|
if monthly_engagement_drop_pct > 0.5:
|
|
score += 0.20
|
|
drivers.append(f"انخفاض الاستخدام {monthly_engagement_drop_pct*100:.0f}%")
|
|
elif monthly_engagement_drop_pct > 0.3:
|
|
score += 0.10
|
|
|
|
# Support tickets
|
|
if support_tickets_open >= 3:
|
|
score += 0.15
|
|
drivers.append(f"{support_tickets_open} تذاكر دعم مفتوحة")
|
|
elif support_tickets_open >= 1:
|
|
score += 0.05
|
|
|
|
# Billing
|
|
if billing_failures_last_90d >= 2:
|
|
score += 0.15
|
|
drivers.append("فشل في الدفع متكرر")
|
|
elif billing_failures_last_90d >= 1:
|
|
score += 0.07
|
|
|
|
# NPS
|
|
if nps is not None:
|
|
if nps <= 6:
|
|
score += 0.20
|
|
drivers.append(f"NPS منخفض ({nps})")
|
|
elif nps == 7:
|
|
score += 0.05
|
|
|
|
# Outcome — Dealix's job
|
|
if pipeline_added_drop_pct > 0.5:
|
|
score += 0.15
|
|
drivers.append("Dealix لا يجلب pipeline كما كان")
|
|
elif pipeline_added_drop_pct > 0.3:
|
|
score += 0.07
|
|
|
|
# New customers cushion: <3 months gets a -0.1 because honeymoon
|
|
if months_as_customer < 3:
|
|
score = max(0, score - 0.10)
|
|
|
|
score = min(1.0, score)
|
|
|
|
if score >= 0.65:
|
|
band = "critical"
|
|
action = "اتصل بالعميل اليوم. أعرض QBR + offer to extend pilot. هذا يكلف $0 إذا أنقذناه."
|
|
elif score >= 0.45:
|
|
band = "at_risk"
|
|
action = "حدد call مع decision-maker. أرسل Proof Pack + roadmap للنصف القادم."
|
|
elif score >= 0.25:
|
|
band = "watch"
|
|
action = "راقب أسبوعياً + أرسل insights مخصصة. لا تدخل عاجل."
|
|
else:
|
|
band = "safe"
|
|
action = "صحي — فكّر في expansion / upsell."
|
|
|
|
confidence = 0.6 if len(drivers) <= 1 else min(0.95, 0.6 + len(drivers) * 0.07)
|
|
|
|
return ChurnPrediction(
|
|
customer_id=customer_id,
|
|
score=round(score, 3),
|
|
band=band,
|
|
drivers=drivers,
|
|
recommended_action_ar=action,
|
|
confidence=round(confidence, 3),
|
|
)
|