Case Study: Banco do Brazil

Banco do Brasil is one of the largest financial institutions in Latin America, serving retail, corporate, agribusiness, and government clients across a national branch network with the Brazilian government as majority shareholder. Banco do Brasil had been running NiCE CXone in its contact center for years. The harder question was what to do about the rest of the bank, the relationship offices where bankers manage long-term portfolios of high-value clients across asynchronous WhatsApp, chat, and voice conversations, sometimes carrying ten thousand customers per portfolio and juggling dozens of simultaneous interactions. Banco do Brasil’s answer was a deliberate two-phase move: centralize all client communication channels into CXone in 2025, then layer NiCE Copilot for Agents on top in 2026 to give relationship managers real-time summaries, sentiment, and a unified workspace built for the conversations they actually have.

01 Before

A platform that fit one part of the bank, and a question about the rest

Banco do Brasil is one of Latin America’s largest financial institutions. It serves retail, corporate, agribusiness, and government clients across Brazil through a national branch network, with the Brazilian government holding majority ownership. Behind that scale sits a workforce of roughly twelve thousand relationship managers and assistants distributed across the country, every one of them responsible for a portfolio of long-term clients who expect their bank to know them.

Banco do Brasil had been running NiCE CXone in its contact center for years. The platform handled inbound contact center work the way contact center platforms are supposed to. Silvio Sznifer, who leads the operation behind this program, described the fit directly: the platform worked at the contact center like a glove. The harder question was the rest of the bank.

The relationship offices were a different shape of operation. A relationship manager in a high-volume office could carry a portfolio of ten thousand customers. A relationship manager in an exclusive office worked with around eight hundred. The work was almost entirely asynchronous, conducted across WhatsApp, chat, and voice, with clients replying on their own time. A single banker might have ten or fifteen WhatsApp conversations running at the same moment, and some bankers, Márcio Fonseca said, opened as many as eighty tabs at peak just to keep track of who was waiting on what.

That volume came with a different kind of cost. Every time a banker pivoted from one client to another, they had to reconstruct the context themselves: scroll back through the WhatsApp thread, jump into another system to check the account, open the CRM in a third tab to see when this client was last contacted. The platform that made the contact center efficient had not yet been built into the relationship office, and the relationship office was where most of Banco do Brasil’s actual customer relationships lived.

Underneath the workflow friction sat a structural CRM challenge. Manual logging of offer-making activity left room for entries that did not always reflect the precise nature of the contact. The data feeding the bank’s next-best-offer engine deserved a more direct signal: interaction-derived evidence rather than after-the-fact recall. That recognition would shape what Copilot needed to do for Banco do Brasil from the very first feature on.

“The contact center fits like a glove. But we wanted more. We have all these kinds of different branches with a lot of customers, and we needed to do relationship and marketing with them. We wanted to deliver to our customers a real omnichannel experience.”

Silvio Sznifer

Software Engineer Specialist,
Banco do Brasil

02 Desire to change

Channels first. Intelligence second. Both deliberate.

The sequence Banco do Brasil chose was deliberate. Channel integration came first. AI came second. Sznifer described the logic plainly: the first job was to get the channels done. WhatsApp, chat, the bank’s own mobile application, and voice all needed to live inside CXone, integrated with Salesforce as the CRM, before any conversation about layering AI on top would make sense. That phase concluded in 2025 with WhatsApp integrated as the dominant channel for client communication in the Brazilian market, where WhatsApp is not a marketing tool but the primary way a bank and its customers talk.

With the channel foundation in place, Banco do Brasil opened the next question. NiCE Copilot for Agents was the obvious candidate, but it carried two complications. The first was the use case itself. Copilot is most often deployed for contact center agents handling discrete inbound interactions. Banco do Brasil needed something different: a tool built into the workspace of a relationship manager who maintained client portfolios across weeks and months and made active outbound offers in addition to handling inbound contact. The AI requirements were not the same as a call center.

The second complication was governance. Banco do Brasil is a state-controlled bank in a heavily regulated industry, and generative AI sat differently from the natural-language processing and IBM Watson capabilities the bank had used inside its WhatsApp chatbot for years without regulatory friction. Banco do Brasil’s posture was to move more carefully than private-sector peers, in part because the cost of a generative AI mistake in a customer-facing context is a number the bank did not want to find out. The decision was to start with employees, not customers. Copilot would go to bankers first, where the human stayed in the loop and the AI’s job was to make the banker faster, more contextual, and more accurate.

The business championed it. Márcio Fonseca, CRM Manager at Banco do Brasil, articulated three goals for what Copilot needed to deliver, and those three goals shaped what the team built.

Goal one was a summary of past interactions, so a banker re-engaging with a client could see what had already been discussed without scrolling through threads. Goal two was a real-time summary during the current conversation, so a banker pivoting between concurrent asynchronous chats could load the context of any one of them in seconds. Goal three was a closing summary tied to sentiment, written directly back into the CRM timeline so the data feeding next-best-offer was an accurate record of what had actually happened. The three goals were not interchangeable. Each one solved a different part of the same problem: a relationship manager managing too many concurrent conversations to hold every context in their head.

03 NiCE solution

Putting Copilot where the conversation actually lives

The architectural decision underneath the program was simple to state and harder to execute. Copilot would not sit beside the workspace. It would sit inside it. When a relationship manager opened the Banco do Brasil interaction suite, Copilot was activated as a panel directly alongside the conversation, with WhatsApp, chat, and voice all flowing through the same view. The banker no longer had to switch between four systems to handle a single client. One screen, with the conversation, the past, the sentiment, and the next move all visible at once. That architecture mattered most for the work the contact center model was never designed to handle: asynchronous, concurrent, long-tail. Sznifer described the day-to-day. A banker at a high-volume office might have fifteen WhatsApp conversations running in parallel. Customers reply when they can, not when the banker can. The banker fields one, pivots to the next, returns to the first hours later. Without Copilot, every pivot meant rebuilding context manually. With Copilot, the banker opened the summary, absorbed where the conversation had been, and continued without rereading the full thread. As Fonseca put it, “we had reports here of bankers who open eighty tabs just to manage their day. The summary brings the last conversations resumed for them, and even when a banker hands a client off at the end of a six-hour shift, the next person already has a backlog ready to read.”

“We had a problem. My employee has offer goals. From the moment I have the interaction directly, I have no doubts that the contact happened and that what was discussed was real. That changes everything for the CRM and for the next offer we make.”

Márcio Fonseca

CRM Manager,
Banco do Brasil

The CRM integrity story sat alongside the workspace story. With sentiment and conversational content flowing back into the timeline directly from the interaction, the CRM became an evidence-based record of what was actually said. The data underneath next-best-offer became contemporary, not reconstructed. For a bank whose entire relationship model depends on knowing the customer well enough to offer the right product at the right moment, that shift is not a small one.

Sznifer was specific about how Copilot fits alongside the bank’s other AI investments. Banco do Brasil also runs Salesforce Agent Force and other agentic AI capabilities, and Sznifer’s framing was that each tool has a vocation. Agent Force does client 360 and next-best-offer modeling inside the CRM. Copilot lives in real time, embedded inside the conversation itself. The two are not competitors. They are different specialists, each placed where its specialty earns its keep, feeding each other through the data Copilot writes back into the timeline.

The implementation is partner-supported. A5 Solutions worked alongside Banco do Brasil and the NiCE team on the build. The pilot is recent, the rollout is staged, and the early adoption pattern required Banco do Brasil to put a dedicated employee on the floor specifically to drive uptake, because Copilot needed to be activated by the banker rather than appearing automatically alongside the workspace. That activation friction is the kind of detail that surfaces only with real users on the floor, and the kind of feedback loop the team is using to shape what comes next.

04 Results

Honest about the pilot. Honest about what is already visible.

The pilot is genuinely early. Banco do Brasil is monitoring performance indicators continuously, and the headline outcome metrics, including efficiency, conversion lift, response time, and customer satisfaction, are expected but not yet validated. The bank’s own framing of the work is that this is a foundation, not a finished result. That framing is also a discipline. State-controlled banks operating under regulatory scrutiny do not get to overstate, and Banco do Brasil chose not to.

What is already visible is qualitative and consistent. Bankers using Copilot describe the past-conversation summary as the feature that changes the most about their day, because it removes the cognitive cost of context-switching between concurrent asynchronous clients. A banker handing off a client at the end of a six-hour shift now leaves their successor a ready summary instead of an inbox full of unread threads. Bankers who once spent significant minutes per pivot reconstructing context now spend seconds. The work has not gotten faster in a measurable sense yet. It has gotten cleaner.

The CRM data integrity gain is the second visible shift. The bank’s offer-making engine is now fed by interaction-derived sentiment and content, and the team is already seeing the difference in the records flowing through. Whether that translates into measurable next-best-offer conversion lift is one of the metrics being watched, and the answer will come in the next phase of the program.

Underneath those operational shifts is a redefinition Banco do Brasil’s leadership now articulates plainly. Omnichannel, in their telling, is not the act of running ten or twelve customer service solutions in parallel. It is the experience of being treated as one customer by a bank that knows the timeline of the relationship, regardless of which channel a given conversation came through. That redefinition is itself a result. Banco do Brasil is no longer building toward channel coverage. The bank is building toward the experience that channel coverage is supposed to enable, and Copilot is the piece that closes the gap.

“Today I do not have this information in an online report. However, in the near future, I will be able to obtain it: at the end of the day, I will consult this dashboard and verify whether my team is working toward the objective set for the day. If they are not, I will be able to correct course within the same day. Today, the only data available to us is historical—and we are working in the present.”

Márcio Fonseca

CRM Manager,
Banco do Brasil

05 Future

From banker workspace to real-time office visibility

The most interesting forward signal came from a question Banco do Brasil had not originally scoped. With Copilot writing summary and sentiment back into the timeline in real time, the data is now available not just to the banker on the conversation, but potentially to the general manager running the office. That opens a possibility the bank does not currently have: in-the-day visibility into whether a relationship office is working on the right objective at the right moment, instead of waiting until the next morning to read yesterday’s report.

That shift, from retrospective coaching to in-the-moment course correction, is the kind of capability that emerges when the workspace and the data layer collapse into the same surface. Banco do Brasil is not building it yet, but the team is mapping it as a near-term extension of what Copilot already produces.

Beyond the office-visibility extension, Banco do Brasil is continuing to deepen the integration between Copilot and Salesforce Agent Force, letting each tool do what it was made for: Copilot in real time alongside the conversation, Agent Force in the CRM enriching the next-best-offer model with the data Copilot feeds in. Customer-facing generative AI remains a medium-to-long-term consideration, deliberately, because the cost of an error in front of a client at a regulated bank is steeper than the cost of moving carefully. For now, Banco do Brasil is building from the inside out: bankers first, customers later, when the trust has been earned in the regulated context the bank actually operates inside.

“Start small, aim big. We sometimes lose a lot of time trying to start big. We have to start small, with little pieces, get in touch with the technology, get in touch with the solution, and understand how it can help us reach our big picture. Every solution has its vocation. What it was born to do, what it specializes in. Copilot’s vocation is real time, integrated with the conversation. Agent Force’s vocation is client 360 and next-best-offer. Use each one for what it was made for. It is very expensive to use a solution outside its vocation.”

Silvio Sznifer

Software Engineer Specialist,
Banco do Brasil