Tonik Bank

As the first digital neobank in the Philippines, Tonik Bank built its entire customer relationship through a single mobile app — making responsive, always-on support a core product requirement rather than a back-office function. Rapid customer growth brought a surge in frequently asked questions around account features, transactions, and onboarding, creating queue pressure on human agents and lengthening wait times. In a mobile-first emerging market where customer trust in digital banking is still developing, slow or inconsistent support responses directly threatened retention and brand credibility. Scaling the human support team proportionally was neither cost-effective nor fast enough to match digital growth rates.

Tonik Bank partnered with Gupshup — a Conversation Cloud provider serving over 45,000 brands globally — to build and deploy a Generative AI chatbot embedded directly into its mobile app. The solution draws on a multi-model architecture combining Meta's Llama 2, OpenAI GPT-3.5 Turbo, Mosaic MPT, and Flan T-5, each selected to balance response accuracy, latency, and cost across different query types. The models were tuned specifically for banking functions — account inquiries, product FAQs, transaction guidance — and the implementation included enterprise-grade security measures appropriate for a regulated financial institution. By integrating the chatbot at the app layer rather than as a standalone channel, Tonik preserved a smooth user experience while enabling real-time, personalised responses without human intervention.

The deployed chatbot now handles 75% of customer queries autonomously, freeing human agents to concentrate on complex, high-judgment interactions that benefit from personal attention. Response times dropped significantly as the AI handles routine inquiries instantly at any hour, directly addressing the wait-time problem that motivated the project. On the financial side, Tonik Bank projects the streamlined operations will generate over USD 20 million in cost savings over three years — a figure that reflects both reduced agent workload and avoided headcount expansion.

• 75% of customer queries resolved without human escalation
• $20M+ projected three-year cost savings
• Reduced wait times across the mobile app customer service channel
• Human agents redeployed toward higher-complexity customer issues

• Multi-model architectures let neobanks tune for cost, speed, and accuracy simultaneously — no single foundation model needs to do everything.
• Integrating the chatbot directly into the primary customer channel (the mobile app) avoids friction and drives higher containment rates than bolt-on deployments.
• In emerging markets where mobile is the dominant banking interface, AI-driven support availability around the clock is a tangible competitive differentiator.
• Achieving 75%+ autonomous query resolution requires domain-specific fine-tuning, not just a general-purpose LLM deployment.
• Projecting multi-year ROI at the outset of deployment helps secure internal alignment and sets a clear benchmark for ongoing performance measurement.