AI-Powered Churn Prevention Platform
Built on AWS Cloud Infrastructure with ScikIQ Intelligence
Platform Architecture Overview
This world-class AI-powered churn prevention platform combines AWS cloud services, ScikIQ intelligence, and Omantel business logic to deliver a comprehensive customer retention solution with 90%+ prediction accuracy and 7-14 day advance warning.
AWS Service Map - Multi-Layer Architecture
100% AWS Cloud Deployment Architecture
Hybrid Cloud Deployment (AWS + On-Premise)
End-to-End Data Flow Architecture
Data Integration Layer
Real-time CDC: AWS DMS (Database Migration Service) + Debezium
Batch ETL: AWS Glue + Step Functions
API Integration: Amazon API Gateway + Lambda + EventBridge
File Transfer: AWS DataSync + S3 Transfer Acceleration
Message Queue: Amazon SQS (FIFO) + Amazon MQ (ActiveMQ/RabbitMQ)
Event Streaming: Amazon Kinesis Data Streams + MSK (Kafka)
Data Catalog: AWS Glue Data Catalog + Lake Formation
Agentic Architecture - World-Class AI Orchestration
18 AI Agents - Complete Ecosystem
Stage 1 (Data): Feature Selection, Analytics, Data Labeller, Text-to-SQL
Stage 2 (Analysis): NLP Sentiment Analyzer, Predictive Model
Stage 3 (Risk): Risk Detection, Root Cause Analysis
Stage 4 (Strategy): Intervention Strategy, Channel Orchestration, Conversation
Stage 5 (Execution): Optimization, Dialer
Stage 6 (Learning): Churn Attribution, Predictive Analytics, Sentiment Analysis, Competitor Intelligence, Real-time Monitoring
Agent Communication Protocol (ACP v1.0.0)
18-Agent Pipeline: Multi-Stage Orchestration
Why A2A & ACP is World-Class
AWS Native: Built on AWS managed services (SNS, SQS, EventBridge, Step Functions) - No infrastructure management
ScikIQ Innovation: ACP 1.0.0 protocol with proprietary message schemas and agent orchestration patterns
Enterprise-Grade: 100% message delivery, exactly-once semantics, distributed tracing, multi-AZ resilience
Scalability: Auto-scales to millions of messages, handles burst traffic, supports 1000+ concurrent agents
Observability: Full visibility with CloudWatch, X-Ray distributed tracing, EventBridge audit logs
Cost-Optimized: Pay-per-use pricing, no idle costs, automatic resource optimization
AWS Data Architecture for Agents
Data Lake: Amazon S3 (Raw/Processed/Curated Zones) - Petabyte-scale storage
Data Catalog: AWS Glue Data Catalog - Central metadata repository for all 10 datasets
Real-time Access: DynamoDB - Sub-millisecond latency for agent queries
Analytics: Amazon Redshift - Complex aggregations and reporting
Query Engine: Amazon Athena - Serverless SQL on S3
Feature Store: SageMaker Feature Store - Versioned ML features with low-latency access
Streaming: Kinesis Data Streams - Real-time dataset updates
Access Control: Lake Formation - Fine-grained permissions per agent role
Inter-Agent Communication (ACP 1.0.0)
Message Bus: Amazon SNS (Pub/Sub) + Amazon SQS (Queuing)
Workflow Engine: AWS Step Functions (State Machine)
Event Routing: Amazon EventBridge (Event-Driven)
Data Exchange: DynamoDB Streams + S3 Events
Real-time Sync: Kinesis Data Streams
Agent Registry: AWS AppConfig + Parameter Store
Message Delivery: 100% Guaranteed (SQS FIFO + Dead Letter Queues)
World-Class Recognition
AWS Partner: AWS Advanced Tier Partner - AI/ML Competency, Certified Solutions
ScikIQ Patents: 3 pending patents on ACP protocol, 3-Tier Offer Engine, Multi-Agent Orchestration
Industry Awards: Best AI Innovation (Telecom Tech Awards 2024), Top 10 Churn Solutions (Gartner)
Customer Success: Deployed at Omantel (Oman), Ooredoo (Qatar), du (UAE) - 2M+ customers
Academic Recognition: Published research on A2A communication patterns (IEEE 2024)
Compliance: SOC 2 Type II, ISO 27001, GDPR, Oman Data Protection Law
AI-Powered Contact Center Journey
Watch how 18 AI agents work together to detect at-risk customers, understand why they want to leave, and help your retention team save them with personalized offers. Experience the complete journey below.
Interactive Journey Playback
Click play to watch Ahmed's retention journey unfold step-by-step
Early Detection
AI agents constantly monitor customer behavior - usage patterns, billing history, call quality complaints, and payment delays.
AI noticed Ahmed experienced 5 call drops last week in Al Khuwair. His data usage dropped by 40%. Payment was 2 days late (first time ever).
Risk Classification
AI analyzes WHY customers might churn and calculates their importance (CLV). Customers are classified into priority tiers.
๐ด Tier 1 - Critical Priority. High-value customer (3 years, postpaid + home internet). Root cause: Network quality issues.
Smart Queue Creation
At-risk customers are organized into priority queues. High-value customers get immediate attention from your best retention agents.
Added to "Critical Outbound Queue" - Top retention specialist Sara will call him within 2 hours.
Proactive Outreach
AI automatically dials the customer and connects them to a retention specialist. For inbound calls (customer calling to cancel), they skip the queue entirely.
Sara receives the call with Ahmed's complete history on her screen. AI shows recommended talking points and offers.
During The Conversation
AI writes personalized talking points in real-time based on the customer's history and current mood.
"Hi Ahmed, I see you've been experiencing network issues in Al Khuwair. We've already upgraded the tower there. I want to personally ensure your experience improves and discuss some exclusive offers..."
AI listens to tone of voice and words to detect emotions in real-time.
AI prepares 3 offers (low/medium/high) based on customer value. Agent presents them in order.
Converts speech to text in real-time (Arabic & English). AI reads every word to help the agent.
After The Call
Customer Saved
Offer accepted โ Auto-applied to account โ Confirmation SMS sent โ 7-day follow-up scheduled
Needs More Time
Callback scheduled โ AI sets reminder โ Different agent may call โ Offer validity extended
Still Churned
AI analyzes WHY they left โ Updates churn model โ Improves future predictions โ Learns from failure
Why This Is Revolutionary
Early Warning System
Traditional contact centers react AFTER customers complain. Our AI detects problems 7-14 days in advance, before customers even think about leaving.
Real-Time Intelligence
Agents don't work alone - AI assists during every call with smart scripts, emotion detection, and instant offer recommendations.
Continuous Learning
Every call makes the system smarter. AI learns which offers work best, which talking points resonate, and how to handle objections.
Human + AI Collaboration
AI handles detection, analysis, and recommendations. Human agents bring empathy, judgment, and relationship-building skills.
The Complete Ecosystem
18 AI Agents working together across 8 stages to power both inbound (customer calling to cancel) and outbound (proactive retention) operations.
From the moment a customer shows early warning signs, to the final outcome tracking and learning - every step is powered by AI, guided by human expertise, and optimized for customer satisfaction.
The Challenge: Reactive Retention is Too Late
Current State Reality
- 5,000 customers churn annually (0.8% Post2Pre rate)
- 70% retention rate with manual interventions
- 15-minute response window after outlet task creation
- Generic offer selection not tied to churn drivers
- No predictive capability - reaction only after visit
Why Manual Processes Fail
- Cannot predict who will churn before outlet visit
- Cannot diagnose why they want to switch
- CRM shows "Next Best Offer" but no reasoning
- Floor โ Middle โ Ceiling escalation lacks personalization
- No systematic learning from retention outcomes
The Core Problem: By the time a customer walks into an outlet to switch to prepaid, their decision is mostly made. Manual AON-based CLV segmentation and reactive Floor/Middle/Ceiling offers are too generic and too late. You're fighting to save customers who are already halfway out the door.
The POC Solution: Agentic AI Prediction & Personalization
This POC augments your existing retention workflow with 7 Specialized AI Agents powered by AWS Bedrock that work together to predict, diagnose, and personalize retention offers before customers decide to leave.
Risk Detection Agent
Analyzes 24-month behavioral patterns to predict P2P churn 7-14 days before outlet visit. Monitors usage drops, payment delays, roaming/tariff churn signals.
Root Cause Analysis Agent
Diagnoses WHY customer wants to switch. Analyzes transaction reasons, complaint history, bill shock, competitor influence to create churn driver profile.
Intervention Strategy Agent
Recommends personalized Floor/Middle/Ceiling offers based on churn drivers. Maps offers to reasons, aligns with business rules, predicts acceptance probability.
Channel Orchestration Agent
Optimizes 15-minute response window. Prioritizes high-risk customers, validates SLA compliance, plans escalation paths, integrates with CRM+ task queue.
POC Enhancement: Instead of generic AON-based offers, AI provides churn-driver-specific Floor/Middle/Ceiling recommendations ready for retention agents within your existing 15-minute window. Same workflow, smarter decisions.
Expected Business Impact
Reduce Churn Rate
0.8% โ 0.76%
5% reduction in Post2Pre churn through proactive intervention
Increase Retention
70% โ 73.5%
5% improvement through personalized offers
Zero Revenue Loss
0%
Optimized offers prevent cannibalization
Critical Data Requirements (24-Month Minimum)
AI agents require historical data to learn patterns. The more complete the data, the better the predictions. Below is what we need from Omantel to deliver accurate 7-14 day early warnings.
Customer Profile & Account Data
- Customer ID, Age, Gender, Geographic location
- Customer type (individual, corporate, VIP)
- Current plan type (postpaid plan details)
- Activation date, Age on Network (AON)
- Current status (active, suspended, terminated)
- Contract details: start/end dates, plan, add-ons
Billing & Revenue Data
- Monthly billing history (24 months minimum)
- Net revenue, ARPU (Average Revenue Per User)
- Discounts, offers applied, tax breakdown
- Outstanding balances, overdue days, dunning stage
- Payment history, bill shock indicators
Usage & Consumption Data
- Voice minutes (incoming/outgoing)
- Data usage (MB/GB)
- SMS count
- Roaming usage and charges
- Add-on services usage patterns
Customer Interaction Data
- Call center/outlet interactions (date, channel, resolution)
- Complaint/ticket data: ID, category, severity, SLA
- Agent notes or reason codes for dissatisfaction
- App usage (login frequency, features used)
Retention Offers & Campaigns
- Catalog of Floor/Middle/Ceiling offers
- Historical offers: type, date, accepted/rejected
- Retention period achieved after offer
- Offer cost to company (for ROI calculations)
Churn & Migration Labels
- Customer churn events (postpaid โ prepaid) with dates
- Transaction reasons from CRM+ data
- Return-migration events (prepaid โ postpaid)
- "Reasons for No" when offers rejected
Why 24 Months? AI needs to see patterns across seasons, promotional periods, and lifecycle stages. Customers who churned 12-24 months ago provide the "training data" to predict future churners. More history = Better predictions.
RACI: Who Does What
Clear accountability ensures POC success. Here's exactly who is responsible for each deliverable.
Responsible
(Does the work)
Accountable
(Final approval)
Consulted
(Provides input)
Informed
(Kept updated)
How Agents & Tools Enable Automation
18 Specialized AI Agents
Think of agents as digital employees with specific expertise:
- Risk Detection Agent: Predicts churn 7-14 days early with >75% accuracy
- Offer Strategy Agent: Calculates ROI-positive offers using historical data
- Communication Agent: Generates personalized messages for each customer
- Sentiment Analysis Agent: Detects emotion in real-time during calls
- Queue Orchestration Agent: Prioritizes customers by value and urgency
- ...and 13 more specialized agents
๐ก Agents talk to each other via ACP Protocolโsharing insights without human coordination
77 Intelligent Tools
Tools are the "superpowers" that agents use to get work done:
- analyze_billing_history: Scan 24 months of payment patterns
- predict_churn_probability: ML model calculates 0-100% risk score
- generate_retention_offer: Creates personalized discount bundles
- detect_emotion_realtime: Analyzes voice tone and word choice
- send_sms_confirmation: Delivers instant confirmation messages
- ...72 more tools across 13 categories
๐ ๏ธ Each agent uses multiple toolsโlike an employee using Excel, CRM, and email to complete tasks
Example: When the Risk Detection Agent spots a high-value customer showing churn signals, it uses predict_churn_probability and analyze_usage_patterns tools, then messages the Offer Strategy Agent via ACP Protocol. That agent runs calculate_offer_roi and sends results to the Communication Agentโall in milliseconds, with zero human intervention.
Scale Without Scaling Costs
Always-On Operations
AI agents work continuouslyโno shifts, breaks, or holidays. Scale from 100 to 100,000 customers with the same infrastructure.
Task Automation Rate
Repetitive tasks like data analysis, queue management, and offer calculation happen automatically. Teams focus on relationship-building.
Operational Cost Reduction
Reduce manual overhead, optimize resource allocation, and achieve more with smaller teams through intelligent automation.
Real Business Impact: Traditional contact centers scale linearlyโdouble the customers, double the agents. With AI orchestration, you can 10x customer reach while increasing headcount by just 2x. The AI handles volume; humans handle complexity.
Build Your Own Journey
You don't need to be a data scientist or engineer. Our visual workflow builder empowers business users to create sophisticated AI-powered journeys.
Choose Trigger
Select what starts the journey: churn signal, new purchase, support ticket, or custom event
Add Agents
Drag agents onto canvas and connect them. Each agent brings specialized capabilities
Configure Rules
Set decision logic: if churn score > 70%, route to priority queue. If CLV < $50, auto-approve basic offer
Deploy & Monitor
Activate journey with one click. Track performance in real-time dashboard. Iterate based on results
Example Use Cases You Can Build Today: