🚗 Case Study: How Yango Nepal Handles 22,000+ Calls/Month with AI
Detailed case study of Yango Nepal's AI call center deployment. Real numbers: 22K calls/month, 96.7% answer rate, ~500ms response, multi-language support. Architecture, results, lessons learned.
22,000+ calls/month handled by AI alone. 96.7% answer rate (vs ~60% industry average for human call centers). ~500ms average AI response time. 17 concurrent DID numbers processed simultaneously. 3,700+ auto-generated tickets for unresolved cases. Deployed in 5 days using TalkC.ai.
About Yango
Yango is one of the world's largest ride-hailing platforms, operating in 30+ countries with 100M+ users. Yango Nepal launched in 2025, expanding their global services into the Nepali market alongside competitors like Pathao and inDrive.
The Challenge
Within months of launch, Yango Nepal was struggling with rapidly growing driver support call volume. Drivers needed help with:
- Bonus and incentive inquiries
- Payment issues (cashout, balance, withdrawals)
- App technical problems
- Ride dispute resolution
- Account/registration questions
- Bank account changes
Traditional call center approach would require:
- 20+ Nepali-speaking agents for 24/7 coverage
- $80K-150K/year in salaries
- 3-6 months to staff, train, and stabilize
- Significant management overhead
And even with all that — 60-70% answer rate would be considered good. Drivers who can't reach support quickly become frustrated and may stop driving.
Why AI Was the Right Choice
Driver support calls follow predictable patterns:
- 80% of calls are about known issues with known answers
- Repetitive — "When will my bonus arrive?" asked thousands of times
- Language: Nepali primary, some English, code-switching common
- Time pattern: 24/7 needed (drivers work all hours)
- Volume: Spikes during weekends, peak hours
This is exactly where AI excels: high-volume, repetitive, multilingual, 24/7.
The Solution: TalkC.ai Platform
Architecture
- SIP Trunk: Buel (sip.buel.app) with 17 DID numbers for parallel call handling
- PBX: Asterisk 20 with PJSIP, deployed on DigitalOcean SGP1 (Singapore for low latency to Nepal)
- AI Engine: Gemini 3.1 Flash Live native audio model
- Voice: Leda (female Nepali voice)
- Database: PostgreSQL 16 storing calls, transcripts, sentiment, tickets
- Dashboard: Custom Fastify + Alpine.js admin panel
Knowledge Base
The AI knows:
- Yango bonus structure (peak hour, weekly, monthly)
- Payment cycles and bank transfer processes
- App troubleshooting (cache clear, reinstall, location permissions)
- Ride pricing rules and surge multipliers
- Account verification steps
- WhatsApp support escalation: 9705007877
System Prompt Highlights
- Identity: "तिमी मिगा (Miga) हौ — Yango Nepal को support team"
- Personality: warm, patient, drivers are friends not customers
- Natural Nepali with fillers: हजुर, हस्स, हो नि, अँ
- Tone matching: angry driver → empathize first, then solve
- No corporate language: never say "Your issue has been noted"
- Escalation: detected frustration or "manche chaiyo" → transfer to human
Deployment Timeline
| Day | Milestone |
|---|---|
| 1 | Initial conversation with Yango Nepal team. Identified use cases. |
| 2 | SIP trunk credentials provided. VPS provisioned. |
| 3 | Knowledge base built from existing FAQ documents. |
| 4 | System prompt refined. Initial test calls. |
| 5 | Soft launch — AI handling 30% of calls |
| 10 | AI handling 80% of calls |
| 14 | Full deployment — AI handling all calls except explicit escalations |
The Results (3 Months In)
Call Volume
- Month 1: 12,000 calls handled
- Month 2: 18,000 calls handled
- Month 3: 22,000+ calls handled
Performance Metrics
| Metric | Value | Industry Avg |
|---|---|---|
| Answer rate | 96.7% | ~60% |
| Avg response time | ~500ms | 20-60 seconds (hold) |
| Avg call duration | 62 seconds | 4-6 minutes (human) |
| Cost per call | ~Rs 3 | Rs 30-50 |
| Resolution rate | ~85% (AI alone) | ~70% |
| Sentiment: satisfied | ~50% | ~40% |
| Sentiment: negative | ~10% | ~25% |
Tickets Generated
When AI couldn't resolve a call (complex disputes, app bugs requiring eng team), it auto-created tickets:
- 3,700+ tickets generated in 3 months
- Auto-assigned to relevant team
- Full call transcript attached for context
- Sentiment score helps prioritize
Operational Impact
- Human team: From planned 20 → actual 3 (handles only complex escalations)
- Cost savings: ~$120,000/year vs all-human call center
- Coverage: 24/7/365 vs limited hours
- Scaling: Volume grew 80% but cost stayed flat
- Driver satisfaction: NPS improved 35% (from anecdotal feedback)
Real Conversation Examples
Example 1: Bonus Inquiry (resolved by AI)
Driver: "Mero bonus aaj kati ho hajur?"
AI: "हजुर तपाईंको registered number बाट म check गर्छु। तपाईंको आज को earnings र bonus app मा 'Earnings' section मा देखिन्छ। यदि specific bonus type बारे जान्न चाहनुहुन्छ भने मलाई भन्नुहोस्।"
Driver: "Quest bonus chai?"
AI: "Quest bonus daily देखिन्छ app को 'Quests' tab मा। आज तपाईंले ५ rides पुरा गर्नुभयो भने Rs 300 quest bonus मिल्छ। पुरा गर्नुभयो हजुर?"
Example 2: Payment Issue (escalated)
Driver: "Mero account ma 3 din dekhi paisa aakto chhaina!"
AI: "हजुर त्यो त frustrating हुन्छ नि, म बुझ्छु। ३ दिन भयो भने technical issue हुन सक्छ। म तपाईंको case हाम्रो team लाई priority मा pass गर्छु — within 24 hours response मिल्छ। तपाईंको registered number 98XXXXX1234 हो हजुर?"
[Call escalated to human, full context preserved]
What Yango Nepal Says
"TalkC.ai changed everything for our driver support. We went from worrying about hiring 20 agents to having a system that handles unlimited calls 24/7. The Nepali quality is genuinely good — drivers don't realize they're talking to AI for the first 30 seconds."
— Yango Nepal Operations Team
Lessons Learned
1. Knowledge Base Quality Matters Most
Spent significant time building accurate KB. Garbage in = garbage out for AI.
2. Iterate Continuously
Weekly call reviews led to ~30% improvement in resolution rate over 3 months.
3. Match the Local Tone
Generic AI sounds robotic. Custom prompt with हजुर, हस्स fillers made huge difference.
4. Don't Be Afraid to Escalate
AI knowing its limits builds trust. Always provide human path for complex issues.
5. Monitor Sentiment Daily
Catch problems before they spread. AI sentiment analysis is your QA team.
Want Similar Results for Your Business?
TalkC.ai works for any business with high call volume:
- Ride-hailing / transportation
- E-commerce / order tracking
- Banking / financial services
- Healthcare / appointment scheduling
- Education / admissions
- Government services
Contact team@talkc.ai for a personalized consultation.
Frequently Asked Questions
How long did Yango Nepal's deployment take?
Initial deployment took 5 days from first conversation to soft launch. Full production deployment was complete within 2 weeks.
What was the cost savings vs human call center?
Estimated $120,000+/year savings, plus avoided hiring/training overhead. ROI was achieved within 2 months.
Can other ride-hailing companies use the same setup?
Yes. The architecture is generic. Customizations are in the system prompt and knowledge base for each business. Pathao, inDrive, or any ride-hailing service can deploy similar.
What languages did Yango Nepal need?
Primary: Nepali. Secondary: English (for some drivers and English-speaking customers). Some Hindi handling for cross-border or expat callers.
Did human agents lose jobs?
Yango Nepal kept a smaller team (3 humans for escalations). They didn't have to lay off existing staff — AI scaled capacity without scaling team size.
Ready to see TalkC.ai in action?
Get a personalized demo of TalkC.ai's voice AI platform. See how we handle 22,000+ calls/month for Yango Nepal, OCR Nepal, and government offices — same-day setup, 70+ languages.
Book a Demo →