Initial Challenges and Customer Context
The challenge is to meet growing demand without increasing operational capacity or compromising service quality.
Almost a million calls a month, and a clear starting point: 0% automated resolution.
Visa Prisma Medios de Pago is Argentina's largest payment processor, connecting banks, merchants, and consumers and processing billions of transactions every year. Its operation has key particularities: a single phone channel serves customers from 44 banks, among them BBVA, Santander, Galicia, ICBC, Macro, Provincia, Nación, and Ciudad, adding scale and complexity to service.
Phone support for partners ran entirely on traditional channels. Every call came in through an IVR, and when it couldn't resolve the request, it was routed to a human agent. There was no digital voice channel and no production bot: 0% of calls were automated, resolution depended entirely on people.
The challenge wasn't simply adding a bot. Prisma needed to meet growing demand without scaling its operation at the same rate, integrating a new way of serving customers into its existing infrastructure without compromising the continuity of a critical operation.
The goal was ambitious: autonomously resolve at least 70% of the inquiries handled, complying with current regulations and ethical interaction principles, delivering a safe, clear, and empathetic experience. Only when necessary is the inquiry routed to a human agent, sharing the conversation history with whoever picks up the case.
Solution Development
Technology matters. How you integrate it into a critical operation matters even more. A successful PoC does not mean it is production-ready. Integrating, scaling, and sustaining it within a critical operation requires far more than technology alone.
The solution wasn't simply adding a bot to the phone channel. Together with Prisma's IT and Customer Services teams, EDSA integrated conversational AI into an existing financial operation, connecting it to the telephony, systems, APIs, and processes Prisma already used. To carry out this integration, using the Kore.ai conversational platform, a VoiceBot integrated with Avaya was deployed, able to identify the customer, understand the reason for the call, and resolve the flow autonomously. When a person's involvement is needed, the VoiceBot hands off the call with the full context of the interaction — who's calling, why, and what happened up to that point — so the agent can pick up from there without starting over.
For that experience to work inside a real operation, the solution had to bring together several layers: conversational AI to understand and manage the interaction; Avaya as the telephony platform; ASR and TTS for speech recognition and synthesis; Prisma's APIs to connect with business processes and execute operations; and a data lake to store sessions and recordings for later analysis. The architecture, development, and implementation were coordinated by a dedicated EDSA team, working alongside Prisma's IT and Customer Services teams. The result was a new capability integrated into Prisma's technology ecosystem, designed from the start to onboard new use cases on a common foundation.
In an operation of this volume, automating isn't enough — it has to be done without compromising continuity. That's why, together with Prisma, a big-bang rollout was ruled out in favor of a progressive rollout. Each new flow starts with controlled traffic, in limited time windows and under close monitoring by the team. As the automation's behavior is validated, volume is increased until reaching full 7x24 operation. Every production release is also paired with a Hypercare stage to catch deviations, adjust the flow, or even stop traffic if needed. The same logic applies to every rollout: test, measure, adjust, and only then scale.
The journey began with a proof of concept on Purchase Notification and Fraud Report, which validated the conversational capabilities and built the project's business case. From that experience, Prisma and EDSA moved toward full production, reusing development work and training to speed up implementation. Purchase Notification was the first flow taken to full 7x24 operation, followed by Credit Disputes. With the capability already in place, scope began expanding to new cases, such as complaint classification and Debit Disputes, while exploring the evolution toward new agentic capabilities. What started as a test on two use cases became a conversational AI capability that Prisma continues extending across its operation.
Impact and Achieved Benefits
More automatic, more scalable, and always available, with people focused where they add the most value.
The starting point had been 0% automation. With Purchase Notification already in full operation, the VoiceBot hit the target set out for it: 7 out of every 10 calls were resolved autonomously, without human intervention. Results from this first flow in full operation:
The impact wasn't limited to automated calls. As of the measurement cutoff, Purchase Notification had generated savings equivalent to 23,010 person-hours. That operational capacity was freed from handling repetitive inquiries and could be redirected to interactions and tasks where human involvement adds the most value.
Purchase Notification was the first test at scale. From there, Prisma began extending the same capability to more complex cases. Credit Disputes handled more than 58,000 calls since the start of its pilot and reached 96% comprehension, while continuing its maturation process toward the automation target. New flows began joining the same capability, progressively expanding the reach of conversational AI within the operation.
Prisma went from an operation where all resolution depended on people to having a conversational capability integrated into its ecosystem, proven at scale, and ready to keep growing. The impact can be measured in calls and hours saved. But the more important change is different: Prisma no longer needs to start from zero every time it wants to automate a new contact reason. The technological, operational, and learning foundation already exists. This wasn't about automating a task. It was about building a new way of serving customers that Prisma continues to expand, flow by flow.
This wasn't just another bot. It was a new capability within the operation.