In 2019, Bank Itaú launched iti, a digital payment solution that was destined to become a full-fledged digital bank. Itaú is the largest financial institution in Brazil and creating iti was mainly a strategic move to compete against the promising fintech firms that had been stealing market share from incumbent banks.
As a big financial institution, Itaú possesses an enormous amount of financial and behavioral data from its customers. The bank detains over one third of the credit card market share in Brazil and boasts around 40 million retail customers in South America. However, Itaú has historically been little effective in creating value from that data.
One of the main factors that would differentiate this new product from the other digital solutions in Brazil was the fact that iti would leverage Itau’s data in order to create value for consumers and merchants.
I joined the project in late 2018 as head of data, with the important mission to make sure iti would derive as much value as possible from its mother company’s data. This was not a simple task, though. Our team faced many technical, legal and political challenges as we tried to tap at least part of the benefits Itaú’s data could bring. For this blog post, however, I will not focus on the challenges our team faced. Rather, I will describe three examples of the many initiatives in which we were able to extract value from the existing customer data.
- Potentializing network effects
Maximizing same-side and cross-side network effects was central to iti’s strategy. Iti allowed instant money transfer (like Venmo) and payments, without charging its retail customers. The more friends a customer have in the platform, the more people they can exchange money with. Similarly, the more of a customer’s favorite merchants that accept payments with iti, the more valuable it is for the customer. With that in mind, we looked at the data to decide what consumers and business we would invest in acquiring first. Our strategy was to start with clusters of consumers and merchants that were already in financial relationship with one another, so that our first users would see the benefits of network effects sooner. Using money transfers and payments data, we created a network graph that mapped the strength of each financial connection between consumers, and between a consumer and a business. That graph allowed us to identify big clusters of connected consumers the business with which they had most transactions. That data guided the customer acquisition team, as they decided where to invest to start growing.
- Establishing event-based marketing
Using data from the credit-card transactions, we created simple models that identified patterns of shopping that could be used to send relevant push notifications to customers and increase their use of the app as a payment solution. For example, our models identified that many customers visited café B, right after having eaten at restaurant A. Based on that knowledge, we decided to invest in the acquisition of Café B as a partner. Then, we setup an event-driven process that automatically send push notifications for Itaú’s customers right after they’ve used their Itaú credit card at Restaurant A. The push said something along the lines: “What about having your favorite expresso at Café B now? If you pay with iti, we’re giving you a 10% discount.”
We’ve used Itaú’s credit card data to find where the patterns were strongest within the strategic industries we would like to focus. You can imagine how powerful this can be when a company can see on real time one third of all credit card transactions in a country.
- Creating a recommendation system
Inspired in Netflix’s algorithm to suggest movies, we created a recommendation system that determined a list of merchants each user would most likely enjoy. The algorithm, a type of collaborative filter, uses only transaction data so that it doesn’t need to make any demographic assumption. Trying to put it in simple terms, what this model does is to predict that a user will likely become a regular customer of business A, because many other users, who usually visit the same businesses as she, also visit business A regularly. We used that prediction to make relevant recommendations at appropriate time.
Unfortunately, because I left the bank before the app had a significant number of users, I don’t have any robust data validating the benefits of those initiatives. The next step after establishing these first models would be to create a series of A/B tests to validate each approach, comparing the model suggestions with different approaches based on intuition and alternative models/algorithms.