For a leading global bank with a strong presence across institutional, commercial, and retail banking in more than 160 countries and jurisdictions, every customer call was an opportunity, and also a risk. The bank’s Personal Banking and Wealth Management (PBWM) division needed to fundamentally change how agents handled customer queries during those calls, because the tools they had simply weren’t keeping up. Existing systems offered only limited contextual suggestions and couldn’t scale personalized insight at the pace a bank of this size demanded, every call handled well was a customer retained, and every call handled poorly was a relationship put at risk across one of 160 different markets. The bank saw a real opportunity to close that gap with Generative AI, but at this scale, adopting AI wasn’t as simple as picking a model and turning it on. The bank needed a scalable framework to evaluate and implement multiple GenAI models securely and compliantly, across every jurisdiction it operated in, before a single agent could rely on it during a live call.
Ascendion designed and implemented a GenAI studio to evaluate and compare multiple LLMs in real time, using real call transcripts rather than synthetic data, so the bank could see exactly how each model would perform in the moments that actually mattered. At the core of the studio sat a model-agnostic pipeline that ingests call transcripts, extracts context, and surfaces intelligent prompts to support agents live, during the call itself, not after it.
A prompt engineering framework let the bank manage and tune context-specific prompts across every model it tested, all inside a secure, compliance-friendly sandbox where the bank could safely run experiments and monitor performance across response accuracy, latency, and hallucinations, critical guardrails for an institution operating across 160+ jurisdictions. From that foundation, Ascendion developed more than 20 use cases spanning call summaries, transfer detection, knowledge extraction, and guided response generation, all powered by an Agentic Framework built with guardrails, feedback loops, and continuous improvement through real user testing, so the system kept getting smarter with every call it supported.
Tech Stack:
Delivered up to 3x faster agent response time through real-time GenAI prompts, turning what used to be a scramble for context into an instant assist.
Improved customer satisfaction scores by delivering context-aware, accurate replies that felt personal rather than scripted.
Created a secure GenAI experimentation environment, giving the bank room to safely explore multiple models without risking compliance across its many jurisdictions.
Enabled scalability of agent assist tools across business units and markets, turning a single successful pilot into an enterprise-ready capability.
Established a repeatable GenAI integration model, laying the foundation for enterprise-wide GenAI adoption well beyond this first rollout.
