Confidential / Tony Fernandes, AirAsia
Bo is broken.
Here is why, and how to fix it.
Here is why, and how to fix it.
Yewwai Kong
Founder, ourteam AI
Page 1 of 2
~1.9M
Annual Bo interactions (est.)
63M passengers at a 3% industry contact rate. Every one routed through a decision tree that cannot reason.
15M
MOVE monthly active users
AirAsia MOVE is targeting an IPO. Bo is the single biggest drag on app NPS and the investor metrics that determine valuation.
RM283
Revenue at risk per churned passenger
RM231 average fare plus RM52 ancillary per passenger. Every passenger Bo loses is not a lost ticket. It is a lost customer relationship.
37%
37% of passengers switch airline loyalty after one poorly handled disruption (OAG 2023). At 63M passengers, this is a structural revenue risk, not a service metric.
01 / The root cause
Bo was architected as a decision tree in 2019 and has never been rebuilt. It cannot reason, hold context, or access live data. Every real problem falls through the cracks.
Bo today
Rule-based · Script-driven · No reasoning
Matches keywords, misses context entirely
No memory. Asks same questions repeatedly
No live data access. Unaware of actual booking
Informs only. Cannot take any action
No human escalation path when it fails
Never learns from failed resolutions
Regulatory liability. Air Canada was held legally responsible for incorrect chatbot information in a 2024 court ruling. AirAsia carries the same exposure under MACPC.
"I've been going in circles for 2 weeks. Bo just keeps sending me back to the start." — verified passenger review, ComplaintsBoard 2025
Bo rebuilt
LLM agent · Context-aware · Action-capable
Reads full natural language, understands intent
Remembers full conversation. No repeating
Pulls live booking, refund, and flight data
Executes changes in-chat, confirms instantly
Smart escalation with full context pre-loaded
Retrains weekly. Measurably improves
Page 2 of 2 / Architecture and Roadmap
Architecture Brief
How to rebuild Bo
the right way.
the right way.
Yewwai Kong
Founder, ourteam AI
Page 2 of 2
02 / The right architecture
Five layers. One conversation. Zero dead ends.
Qatar Airways, Singapore Airlines, Emirates, and Delta, all ranked in the Skytrax global top 25, have deployed or are actively deploying this architecture. The gap between Bo and the world standard is not a feature gap. It is a fundamental architectural difference.
01 Understand
Frontier LLM parses full natural language: intent, urgency, emotion, booking context. RAG grounds responses in live AirAsia policies and MACPC rules. Conversation memory persists across turns. Supports BM, EN, ID, TH, ZH.
02 Access
Real-time API read access: booking engine, refund status, seat inventory, disruption data, BIG loyalty balances. Bo knows the passenger's full situation before they finish typing.
03 Act
Reschedules flights, processes refunds, selects seats, applies vouchers, sends confirmations. All in one conversation, under two minutes. Actions beyond authority route to a human Allstar with full context.
04 Escalate
Sentiment and confidence scoring on every turn. Complex or high-frustration cases transfer to a human Allstar with a structured briefing. The agent starts informed. The passenger never repeats themselves.
05 Learn
Bo gets measurably smarter every week. Every escalation and failure is logged, reviewed, and corrected. Policy changes update the knowledge base within 24 hours. Resolution rate, deflection, and customer satisfaction are tracked on a live dashboard visible to your team.
02B / The strategic opportunity
Bo is not just a support tool. It is the most underleveraged revenue asset in the MOVE ecosystem.
Every passenger who opens Bo is warm, identified, and at peak travel intent. They have a booking. They are days or hours from flying. No marketing spend was required to get them here. Right now that moment generates nothing but frustration. A rebuilt Bo converts it into a revenue channel that compounds across 1.9M annual interactions.
Phase 1 · Months 1 to 6
Fix the core
Rebuild Bo as a reasoning agent. Resolve disruptions. Process refunds. Handle rebookings. Eliminate the loop. This phase restores trust, the prerequisite for everything that follows.
Revenue impact
RM67M
Direct cost savings from 1.5M deflected cases at RM45/case
Phase 2 · Months 7 to 12
Activate ancillary revenue
Bo recognises the context and makes relevant offers. Passenger rebooks a flight, Bo offers the seat upgrade. Disrupted passenger, Bo offers travel insurance and hotel. Connecting flight, Bo offers airport transfer via MOVE Ride. Personalised, not generic. Timed to the moment of highest intent.
Revenue impact
RM94M
10% ancillary uplift on 1.9M interactions at RM52 baseline ancillary per pax (Microsoft AI benchmarks, 2024)
Phase 3 · Year 2 onwards
Power the MOVE flywheel
Bo becomes the conversational interface for the entire MOVE ecosystem. Hotels, SNAP bundles, Ride, duty-free, MOVETIX, all surfaced contextually within the same conversation. A passenger asking about a delayed flight can land on a hotel booking within two messages. Tony's stated target of 50% non-flight revenue runs through this interface.
Revenue impact
RM200M+
Addressable cross-sell conversion across MOVE portfolio at 32% YoY growth trajectory (MOVE Q3 2025 operating results)
Combined 3-year revenue opportunity
Cost savings + ancillary uplift + MOVE cross-sell, conservative model
RM361M+
Cumulative over 36 months
Why this matters for MOVE's IPO trajectory: AirAsia MOVE generated RM641M revenue in 2025 with NPS at 57. A rebuilt Bo that converts support interactions into hotel, ride, and bundle transactions adds a measurable, recurring revenue layer to MOVE's financials, the kind of engagement metric that directly supports a higher valuation multiple at IPO. In Q4 2025 alone, MOVE revenue hit RM300M, nearly doubling year-on-year. The IPO trajectory is real, and Bo is the brand liability attached to that valuation story.
03 / Industry benchmarks
The world's top-ranked airlines are all rebuilding their AI layer. AirAsia, ranked 28th by Skytrax 2025, is falling further behind.
Qatar Airways · Sama
Skytrax #1
Agentic AI booking agent launched 2025. Emotion-aware, voice and chat. Curates full itineraries. Built with OpenAI.
Singapore Airlines · Kris
Skytrax #2
85,000 questions weekly. Partnered with OpenAI in 2025 for full agentic support. 250+ AI use cases across operations.
Emirates · OpenAI
Skytrax #4
Enterprise-wide ChatGPT deployment signed Nov 2025. AI Centre of Excellence established.
Delta Air Lines · Concierge
Skytrax #22
Generative AI concierge launched late 2025. Tracks bags, answers real-time travel queries, proactive alerts. CEO Ed Bastian's flagship AI investment.
KLM · BlueBot
Skytrax #21
50% of all customer interactions handled autonomously. Booking, check-in, travel enquiries. Web and WhatsApp.
AirAsia · Bo
Skytrax #28
Rule-based decision tree. Dead loops. 48-hour case response for same-day disruptions. No live data. No action capability. 1.3★ across 1,653 reviews (ComplaintsBoard · App Store, 2024 to 2025).
Every airline ranked above AirAsia in the Skytrax 2025 rankings has deployed or is actively deploying a reasoning AI agent. Qatar Airways, Singapore Airlines, Emirates, Delta, and KLM are all investing at the platform level, not patching a chatbot. The gap between Bo and the industry standard is architectural, not incremental.
04 / Financial impact
At ~1.9M annual interactions, the real cost is in the 37% who never come back.
At a 3% industry contact rate across 63M annual passengers, Bo handles ~1.9M interactions per year. The financial damage is in the passengers who switch airlines after a single poorly handled disruption.
Agent fallback and case handling
~1.9M interactions · 20% requiring human fallback · RM180/case
RM68M
Regulatory penalties and exposure
RM4.85M cumulative CAAM/MAVCOM fines · repeat-offense escalation
RM1M+
Passenger churn
63M pax · 5% frustrated · 10% defection rate (conservative)
RM200M+
Conservative addressable loss per year
~RM205M
Upside: 80% autonomous resolution deflects ~1.5M cases. At RM45/deflected contact = RM67M direct savings. Each 1% churn recovered = RM18M retained revenue.
05 / Delivery roadmap
MVP in 90 days. Measurable improvement from week one. 80%+ autonomous resolution by month six.
Phase 1 · Weeks 1 to 6
Foundation
LLM fine-tuned on AirAsia policies and historical cases
RAG knowledge base covering FAQs, procedures, and rules
Booking API read integration
Intent accuracy evaluation and benchmarking
Human review panel validates Bo responses in first 30 days. No silent failures reach passengers
MVP: Understand and answer
Phase 2 · Weeks 7 to 14
Action Layer
Booking API write access: change, cancel, refund
Smart escalation and agent handoff with context
Multilingual support (BM, EN, ID, TH)
Live testing and iteration with real traffic
Launch: Understand and act
Phase 3 · Months 4 to 6
Intelligence
Weekly retraining pipeline goes live
Proactive passenger notifications on disruptions
Ancillary upsell recommendations
KPI dashboard: resolution rate, CSAT, deflection
Target: 80%+ autonomous · RM67M direct savings · NPS improvement measurable
ourteam AI builds and deploys production AI agents for enterprise across Southeast Asia.
Production-deployed across telco, aviation, and financial services clients in Southeast Asia. Available to scope and commence immediately.
AI Agents
Agentic Workflows
Voice AI
RAG Pipelines
LLM Fine-tuning
Multilingual NLP