Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How Nubank Made Banking Operations Lean
Transcript
- Lucas: So it's May 2026, and Nubank just announced it's crossed 100 million customers across Brazil, Mexico, and Colombia. That is roughly the population of the Philippines. And they did it with zero physical branches. Luna: Zero branches. That part still blows my mind. Traditional banks in Latin America spend something like 60 percent of their non-interest expenses just maintaining real estate and teller staff. Lucas: Exactly. And Nubank's whole operating model was built from the ground up to avoid those costs. The question I want to drill into today is: how specifically did they design their operations to make that work? Not just the app, but the actual back-end processes that let them acquire, serve, and underwrite risk for 100 million people without a single branch. Luna: I think the most under-appreciated piece is their approach to customer acquisition. They didn't spend billions on marketing. They relied on viral referrals and a very simple onboarding flow. Lucas: Right. The onboarding was famously fast — you needed just a CPF, the Brazilian tax ID, and a selfie. But the operational insight is that they automated the identity verification process using machine learning models trained on government databases. So instead of a human reviewing documents, a model handled it in seconds. That single decision removed a huge cost and a huge delay. Luna: And it scaled. If you have a human reviewing each application, you hit a bottleneck at maybe a few thousand per day. Nubank was processing tens of thousands per day within a year. Lucas: The second piece is the technology architecture. Nubank built on a micro-service model from day one. Each feature — account opening, payments, credit card, loans — runs as its own independent service. That means they can deploy updates to one service without taking down the whole bank. Luna: Contrast that with most incumbents. A friend of mine works at a large Canadian bank, and they still have mainframe code from the 1970s. A new feature takes eighteen months because you have to test it against the entire monolithic system. Lucas: Nubank's engineering team can push code to production dozens of times a day. That operational speed is a competitive advantage. When the Brazilian central bank introduced a new instant payment system called Pix in 2020, Nubank had it integrated in weeks. It took some traditional banks over a year. Luna: Let's talk about credit risk operations, because that's where a lot of fintechs have stumbled. Nubank started with a credit card — no annual fee, but they had to figure out who to lend to in a country with very volatile inflation and interest rates that hit 13.75 percent in 2023. Lucas: Their approach was data-driven but also conservative. They used alternative data — utility bill payments, mobile phone recharge history, even social media signals — to build credit scores for people who had no traditional credit history. But they also started with very low credit limits, like $50 to $100, and then dynamically adjusted them based on real-time payment behavior. Luna: That real-time adjustment is an operational feat. Every time a customer makes a payment or a purchase, their risk profile updates. The system recalculates their limit instantly. Most banks still update credit limits once a month or once a quarter. Lucas: The result is a cost to income ratio of around 35 percent. Compare that to the average Brazilian bank, which is around 50 to 60 percent. That 15 to 25 percentage point gap is pure operational efficiency — it comes from having no branches, automated processes, and a modern tech stack. Luna: But I want to push back a little. Nubank is now fifteen years old. As they add more complex products — mortgages, investments, insurance — the operational simplicity becomes harder to maintain. Each new product brings regulatory requirements, more customer touch points, and more manual exceptions. Lucas: That's a fair point. Their mortgage business in Brazil, for example, still requires a lot of manual document verification because real estate laws are complex. Their cost to income ratio has crept up slightly in the last two years. The question is whether they can keep their lean model as they become a full-service bank. Luna: And as they expand into Mexico and Colombia, they face different regulatory environments, different payment rails, different identity systems. Each new country adds complexity to the operations layer. Lucas: Right. But the core operational philosophy — automate everything that can be automated, use real-time data for decision-making, keep the tech stack modular — that doesn't have to change. The challenge is applying it to higher-touch products without recreating the branch experience digitally. Luna: Before we go further, I want to step back for a second. If today's breakdown of Nubank's operations has been useful to you, if it's given you a concrete way to think about your own work or business, that's the whole point of this show. We keep it ad-free because we believe in the value of these conversations. If you'd like to support that, the simplest way is to buy us a coffee — literally. The link is buymeacoffee dot com slash fexingo. That's it. No tiered memberships, no exclusive content, just a way to say this is worth a couple of dollars to you. Lucas: And honestly, even a small contribution — the price of a coffee — helps us keep going. So if today was worth that to you, that's the link. Now, back to the operations side — I want to talk about one more operational choice that I think is under-discussed: how Nubank handles customer service. Luna: Oh, this is good. They use a combination of chatbots and a very small human team, but the operational twist is that they tie service metrics directly to product design. Lucas: Exactly. If they see a spike in calls about a particular issue — say, why a transaction was declined — they don't just train the chatbot on that issue. They ask their product team to redesign the in-app experience so the issue doesn't happen in the first place. For example, they added a real-time explanation of why a transaction was declined directly in the notification, which reduced calls by 30 percent. Luna: That's a feedback loop that most companies don't have. Customer service is usually a cost center, not an input to product design. Lucas: Right. And that loop is only possible because the tech stack allows rapid changes. If you have to wait six months to change the app, the feedback loop dies. So the operational model is self-reinforcing: micro-services enable rapid product changes, which reduce service costs, which frees up resources to invest in more automation. Luna: Let's talk about the people side too. Nubank's operations team is relatively small — about 6,000 employees total for 100 million customers. That's one employee per roughly 17,000 customers. A traditional bank might have one employee per 200 customers. Lucas: That ratio is only possible because of the automation we've been describing. But it also means that the employees they do hire are highly skilled. They're data scientists, engineers, product managers. They're not processing paperwork. Nubank's organizational structure is flat — there are only three levels between an entry-level engineer and the CEO. That reduces bureaucracy and speeds up decision-making. Luna: Though I've heard that flat structures can create ambiguity. People don't always know who to escalate to. Did Nubank have to adjust as they grew? Lucas: They did. In their early years, they had literally no managers — teams self-organized. But around 2019, as they passed 10 million customers, they introduced a lightweight management layer. But they kept the principle: managers are coaches, not decision-makers. The people closest to the problem still make the call. Luna: So what's the takeaway for someone who runs operations in a completely different industry? A manufacturer, a retailer, a hospital? Lucas: I think the first takeaway is to map every single step in your customer journey and ask: does this step require a human decision? If the answer is no, automate it. Nubank automated identity verification, credit limit adjustments, and most customer service queries. The second takeaway is to make your technology architecture modular. You don't need to be a tech company to use micro-services — any organization with a digital component can benefit. Luna: And the third might be: create a feedback loop from operations back to product design. If you're spending a lot of time handling exceptions, that's a sign your product or process is broken, not that you need more people. Lucas: Nubank's not perfect. They've had outages, they've had regulatory fines in Mexico, and they still lose money on some customer segments. But their operating model gave them a 15-year head start on efficiency. The question now is whether they can extend that model to the complex products that generate the most revenue in banking. Luna: It'll be interesting to watch. Thanks, Lucas. That was a great breakdown. Lucas: Thanks, Luna. If you want to read more about Nubank's early operational decisions, check out their 2021 shareholder letter — it's unusually transparent about their cost structure.