Core Decisions (5)
Who sends? Brand / Account Owner / Assigned Agent.
When? D‑30,14,7,3,1 (+ best hour) respecting quiet hours.
Which channel? Email → SMS → WhatsApp → IVR/Agent by score.
Discount? 0/5/10% based on propensity & LTV (No Claim Discount).
Tone & copy? AR default (formal/friendly), EN fallback.
🎯 Real-World Examples: How the System Adapts
See how the AI-powered decision engine optimizes each customer interaction for maximum renewal probability at minimum cost.
Standard Motor Policy
Low-cost path, high engagement
Profile
Ahmed
D-14
Mid LTV
No Claims
AI Decision:
- Send: 18:30 today (best engagement hour)
- Channel: WhatsApp first (best past performance)
- Discount: None (high renewal probability)
- Tone: Friendly formal Arabic
Message (WhatsApp):
«أهلاً أحمد 👋 تذكير بسيط: وثيقة سيارتك تنتهي خلال 14 يومًا. جدّد الآن في دقيقة واحدة من هذا الرابط: {{pay_url}}. أي مساعدة؟ نحن معك.»
✅ Reads & Renews
Cost: Medium | Touches: 1
Low Engagement Customer
Email→SMS failover strategy
Profile
Sara
D-7
Low LTV
Email Bounces
AI Decision:
- Send: 10:00 tomorrow (respects quiet hours)
- Channel: Email trial → SMS fallback in 24h
- Discount: 5% (low renewal score)
- Strategy: Cost-optimized escalation
Multi-Channel Approach:
Email (cheapest):
"Your policy ends soon—renew in 1 minute."
SMS (after 24h no open):
"باقي أيام قليلة على انتهاء وثيقتك! جددّي الآن واستفيدي من خصم 5٪: {{short_url}}"
📱 SMS Success
Cost: Moderate | Touches: 2
VIP Customer
Human touch justified by LTV
Profile
Fahad
D-3
VIP
Recent Claim
AI Decision:
- Send: 11:00 (business hours)
- Channel: Agent call + WhatsApp summary
- Discount: 10% loyalty discount
- Approach: Personalized human touch
Agent Script (AI-generated):
• Acknowledge recent claim, reassure service continuity
• Present loyalty discount and direct payment link
• Follow up with WhatsApp summary
• Present loyalty discount and direct payment link
• Follow up with WhatsApp summary
🎯 Agent Closes
Cost: High | Justified by LTV
IVR AI-Generated Voice
Automated attention capture + SMS action
Profile
Noura
D-3
Mid LTV
No Claims
AI Decision:
- Send: 19:00 (evening — higher answer rate)
- Channel: IVR (AI-generated voice) + SMS confirmation link
- Discount: 5% (medium renewal probability)
- Approach: Automated call to capture attention + follow-up SMS for action
AI Voice Script (Arabic):
"مرحباً نورا، نحن من مدجلف للتأمين. نذكرك أن وثيقتك تنتهي بعد 3 أيام. يمكنك التجديد الآن والحصول على خصم 5٪ عبر الرابط المرسل في رسالة نصية. شكراً لك."
Follow-up SMS:
"وثيقتك قربت تنتهي! جددّي الآن بخصم 5٪: {{short_url}}"
📞 IVR + SMS
Cost: Medium | Touches: 2
IVR Live Agent
Personalized human touch for high-value customers
Profile
Khaled
D-1
High LTV
Multiple Claims
AI Decision:
- Send: 11:30 (business hours)
- Channel: Live Agent call + WhatsApp payment link
- Discount: 10% loyalty discount
- Approach: Personalized reassurance due to multiple past claims
Agent Call Script (AI-generated draft):
• تحية باسم العميل
• الإشارة إلى انتهاء الوثيقة غداً
• التأكيد على استمرار التغطية وخدمات ما بعد البيع
• عرض خصم الولاء 10% مع إرسال رابط الدفع مباشرة على واتساب
• الإشارة إلى انتهاء الوثيقة غداً
• التأكيد على استمرار التغطية وخدمات ما بعد البيع
• عرض خصم الولاء 10% مع إرسال رابط الدفع مباشرة على واتساب
WhatsApp Follow-up:
"أهلاً خالد، كما ناقشنا عبر الهاتف، يمكنك التجديد الآن والاستفادة من خصم 10٪ عبر هذا الرابط: {{pay_url}}"
🎯 Agent Closes
Cost: High | Justified by LTV
Key Insights from These Examples
- Channel Selection: Past engagement data drives optimal channel choice
- Cost Optimization: Email first, escalate only when needed
- Personalization: Claims history and LTV influence approach
- Voice Strategy: IVR for attention, live agents for high-value customers
- Timing Intelligence: Best hours + quiet hours compliance
- Discount Strategy: Propensity-based offers, not blanket discounts
- Multi-Touch: Strategic escalation when single touch fails
- Hybrid Approach: Voice + digital follow-up for maximum conversion
Minimal Flow
- Nightly candidate build (D‑windows) → enrich (profile, claims, consent).
- Scoring Agent computes features (renewal_score, best_hour, channel_scores).
- Decision Agent returns {send_at, channel, sender, discount, tone}.
- Queue & throttle → generate message (copywriter + guardrail) → send.
- Listen to delivery/reply/payment webhooks → adapt follow‑up or stop.
- Daily learner updates channel costs, variant winners, discounts.
Channel Cost Heuristics
| Channel | Typical Cost | Use When |
|---|---|---|
| Very Low | Good deliverability; first touch. | |
| SMS | Low–Medium | Reach is priority; short copy. |
| Medium | High read rates; template costs monitored. | |
| IVR/Agent | High+ | VIP / high LTV / exceptions. |
KPIs
- Renewal rate ↑, Cost per renewal ↓, Touches per renewal ↓
- Time‑to‑renewal ↓
System Diagrams
A) System Overview
Covers webhooks, Scoring Agent inputs, Decision Agent criteria, channel queue/log, and the content pipeline (Copywriter → Guardrail → Voice).
B) Daily Cycle
Nightly collection of D‑windows → create & store score/decision → create content → schedule send → update status & log.
Decision Agent Input & output
A) Input
B) Output
Lean Implementation (6-7 weeks)
Week 1-3
- Connect Policies/Customers/Consent/Payments
- Set up providers & webhooks
- Draft AR/EN templates
- Generic Messaging Service
Week 4-5
- Scoring Decision Agents (rules + small LLM)
- Queues
Week 6-7
- Demo D-14 (10% traffic) then expand
- dashboard