e& Customer Care AI Chatbot

The Challenge

e& UAE serves millions of telecom customers, and operates multiple consumer platforms — the core e& UAE telecom app, Smiles (loyalty and lifestyle), e& Money (fintech), and more. Customer care across these platforms was heavily dependent on human agents, leading to long wait times, high operational costs, and inconsistent service quality.

The challenge was building a unified AI chatbot capability that could be deployed across all these platforms — each with different user bases, use cases, and integration requirements — while maintaining consistent quality and scaling to handle hundreds of thousands of monthly interactions.

My Role

I served as AI Product Owner, responsible for:

• Defining the product roadmap and presenting it to shareholders and senior stakeholders
• Owning the product backlog end-to-end — writing user stories, acceptance criteria, and prioritizing features using RICE and MoSCoW frameworks
• Collaborating with engineering, data science, and CRM teams through sprint cycles
• Coordinating across five internal IT teams (data, integration, CRM, CX, operations) to ship aligned releases
• Ensuring ISO/IEC 42001 compliance for ethical AI practices, fairness, and data privacy

Process

1. Discovery & Stakeholder Alignment

Analyzed customer care data to identify the highest-volume, most repetitive inquiries — billing questions, plan changes, service activation, and troubleshooting. These categories represented the majority of call center volume and were ideal candidates for AI automation.

Presented the business case to shareholders, demonstrating projected cost savings, improved CSAT, and scalability. Secured buy-in and budget for a phased rollout.

2. Architecture & Integration

Defined the conversational AI architecture integrating NLP/NLU engines with e&'s existing CRM and billing systems. Designed conversation flows that felt natural while handling complex telecom-specific queries. Key decisions included:

• Multi-channel deployment strategy (web chat, WhatsApp Business API, IVR)
• Seamless human handoff protocol for edge cases beyond AI capability
• Real-time analytics dashboard for monitoring resolution rates and identifying training gaps
• RAG architecture for dynamic knowledge retrieval from internal documentation

3. Build & Iterate

Worked in dual-track Agile — discovery and delivery running in parallel. Each sprint delivered measurable improvements in intent recognition, resolution accuracy, and conversation completion rates. Prioritized features based on customer impact and feasibility.

4. Scale Across Channels

After proving the model on web chat, expanded to WhatsApp (the dominant messaging platform in the UAE) and IVR. Each channel required adapting the conversational experience — WhatsApp needed concise, mobile-friendly responses while IVR required voice-optimized dialogue trees.

5. Optimize & Govern

Implemented continuous monitoring and optimization loops. Tracked resolution accuracy, CSAT, escalation rates, and average handling time. Applied ISO/IEC 42001 governance to ensure ethical AI practices — bias detection, transparency in automated decisions, and data privacy compliance.

Key Takeaways

• Start with the highest-volume, lowest-complexity queries — quick wins build organizational trust in AI.
• Multi-channel is non-negotiable — customers don't care which channel they're on, they expect the same quality.
• Human handoff is a feature, not a failure — knowing when to escalate is as important as resolving autonomously.
• Governance from day one — retrofitting ethical AI is harder than building it in. ISO 42001 gave us a framework.