Nubank, setting up the data infrastructure to enable big data decisions
From the start Nubank made sure to set up its engineering infrastructure and product development to enable decisions with big data. Even the programming language used in its backend, Clojure, was chosen thinking ahead of the massive growth the company experienced, both in transactions and in clients. Now Nubank stands as a true disruptor of financial services in Brazil and Latin America, having acquired 35 million users in Brazil alone (a country with a 200 million population), with products such as credit cards, digital accounts, personal loans, life insurance, and many others in the works. As a former employee at Nubank, I’d like to share how data was used and how it continues to be used across levels and functions in the organization.
Use case 1) customer experience
Instead of using readily available tools for customer support, Nubank built its very own customer support software from scratch. The interesting part of it was that the software was set up in a way to allow agents to label contact reasons with extreme ease, populating a large database of customer support tickets, some key information exchanged, and the final contact reason. Using machine learning and natural language processing, Nubank became very assertive on predicting a contact reason with just a few lines of text from a customer, enabling ticket routing to the best available agent who could answer the client’s request. This was one of the first products that Nubank made. An internal product. The reason: customer experience was such a pain with incumbent banks that providing an amazing customer support was central to Nubank’s strategy. The result: Nubank had NPS scores that other companies, even outside of financial services, would never even dream of achieving, which led to product-led growth and goodwill with users, allowing Nubank to grow to millions of users without having spent a dime in marketing.
Use case 2) Function and department-specific dashboards
At Nubank, with some very specific restrictions, anyone can access sanitized customer and transaction data through an online platform. During my time there we used Metabase and people would freely give SQL tutoring to anyone who was interested in exploring the data to gather insights. Customer support agents, designers, product marketers, product managers, business developers, business analysts, engineers, and many other functions could use the same tool for their decisions.
Customer support agents would map quality in a data-driven way through these dashboards, for instance, at an individual agent level, to understand which internal courses would best suit the challenges they were currently facing in order to improve their quality. Prioritization of trainings and support script reviews data-driven way, mapping not only the volume for each contact reason, but metrics such as wait time, support time, customer satisfaction, and many others.
Use case 3) A/B testing
Finally, with the use of these dashboards combined with more advanced tooling, like frontend analytics and real-time database analysis, teams could perform A/B tests with ease. I saw this firsthand when I helped Nubank’s digital account as a product manager, starting off in a growth hacking squad, basically the A/B testing squad. Anything was up for grabs in terms of testing, as long as it helped move the needle on the metrics we defined as key for our squad. With Nubank’s massive client base, in the millions, we were able to sample a small batch of users for any test and achieve statistical significance. When I left the company to pursue my MBA, we were still struggling to find the best approach to gather results and learnings from these tests in one standardized way in order to avoid duplicating tests. This is always an important process consideration when creating an A/B test oriented company.
With great power comes great responsibility
This is a ton of data. Some aspects of this information are sensitive, especially in financial services and a tough regulatory environment. Even more so with Brazil’s current equivalent to GDPR taking effect. Who can access what information?
The same way Nubank set up its tech stack for massive growth, it also built a solid data engineering team to create what current is Nubank’s data policy and data infrastructure. With a tight control on roles and data access, Nubank was able to scale its team to more than 2400 employees combining a variety of functions, while still ensuring that employees had the data they needed to perform their jobs.
The story of Nubank might as well be the story of how they achieved to hire the best of the best in technical talent to achieve this feat of engineering. Tech talent is hard to come by. And while many VCs in the past didn’t dare to invest in Latin American companies because of the perceived tech talent scarcity, Nubank understood the hidden potential of the region and reaped the benefits of having done so, building an incredibly scalable and secure platform not only for end-consumers but also for its internal customers.