Many banks are experimenting with AI.
Few are deploying it at scale — with control ⚙️
We’ve documented how NOVACARD transformed customer operations using AI agents — not as chatbots, but as controlled decision systems 🔒
Scaling customer operations in banking is not just a technology challenge — it’s a control challenge.
NOVACARD, a credit card issuer in Mexico, faced a familiar problem: growing digital volumes, rising service costs, and increasing pressure on customer experience. Building AI in-house was considered — but ultimately rejected due to complexity, cost, and time to market. Instead, they took a different path.
What changed?
NOVACARD deployed AI agents using Flametree.ai — not as generic automation, but as controlled decision systems ⚙️
The focus was clear:
→ structured decision logic
→ defined boundaries
→ controlled escalation to humans
Not experimentation — but production from day one 🚀
Key design principles
• AI operates strictly within predefined boundaries 🔒
• Decisions are consistent and policy-aligned 📏
• Clear rules define when to escalate to human agents
• Sensitive and complex cases remain human-led 🤝
• AI attempts resolution before escalation when appropriate
This created a balance between efficiency and control.
Execution
The rollout was both fast and disciplined:
• ~2 months to implement
• ~2 months to recover the investment
The implementation focused not only on technology alone — but also on:
→ translating business logic into AI decisioning
→ embedding risk & compliance into the design
👉 Results
• Hundreds of thousands of conversations automated monthly
• ~50% of customer service handled by AI (and growing) 📈
• 40–60% cost reduction vs. human-only operations 💰
• Improved response consistency and speed
Interestingly, although higher automation was possible, NOVACARD deliberately optimized at 60–70% — recognizing that over-automation degrades customer experience ⚠️
What’s fundamentally different?
This is not traditional chatbot automation. Compared to IVR and rule-based systems:
• interactions are natural and flexible
• use cases expand continuously
• engagement improves significantly
At the same time, full transparency is maintained:
→ every interaction is measurable 📊
→ quality is continuously evaluated
→ both AI and human performance improve over time
Key insight
AI in banking delivers value only when treated as a decision system — not an automation tool.
Too much automation → loss of quality
Too little automation → limited efficiency
The real leverage lies in precision of control 🎯
What’s next
NOVACARD is now expanding AI agents across its operations. AI agents are already implemented for:
• onboarding
• debt collection
The target model is clear:
→ AI handles high-volume interactions
→ humans focus on complex, high-value cases 🤝
Summary
Deployment of Flametree.ai enabled NOVACARD to:
• scale customer service without a proportional cost increase 📈
• achieve rapid ROI 💰
• implement AI in a controlled, risk-aware manner 🔒