Case studiesAI deliveryOrtigas Center, Metro Manila
AI Chatbot Deployment with Human-in-the-Loop for a Southeast Asian Fintech
78%
AI Autonomous Resolution
94%
Human Escalation CSAT
-34%
Customer Acquisition Cost
8 min
Average Resolution Time
The challenge
- Customer support cost scaling linearly with rapid user growth, threatening unit economics
- Tier-1 support handling 70% of the same repeatable queries, indicating high automation potential
- Previous chatbot deployment attempts hitting under 50% resolution, damaging user trust
- Escalation quality inconsistent across chat, email and in-app channels
- Regional Tagalog-English code-mixing patterns breaking generic multilingual LLM performance
The Corpshore approach
- Built a dedicated AI training team labelling and refining Tagalog conversation data for model fine-tuning
- Deployed a human-in-the-loop workflow with a sub-60-second escalation SLA to protect user experience
- Established a RLHF loop feeding customer feedback signals directly back into weekly model training cycles
- Implemented language-aware routing distinguishing Tagalog code-mixing, pure English and Filipino formal register
- Instituted monthly model performance reviews with the client's machine learning engineering team
Results and impact
| Metric | Before | After | Change |
|---|---|---|---|
| AI Autonomous Resolution | 48% | 78% | +30 points |
| Human Escalation CSAT | 72% | 94% | +22 points |
| Customer Acquisition Cost | Baseline | -34% | 34% reduction |
| Average Resolution Time | 4.0 hrs | 8 min | -97% |
| Code-Switching Model Accuracy | 62% | 88% | +26 points |
| Cost per Ticket | Baseline | -58% | 58% reduction |
The RLHF loop combined with Filipino-native trainers cracked a Tagalog code-mixing problem generic chatbot platforms could not solve.