The lead story: how one bank or fintech really makes its money, rotating through launches, economics, strategy, pricing, operations and incumbents vs challengers.
The numbers. For every dollar of revenue, here's how much each bank spent on running costs in its latest reported quarter (Nubank Q2 2026; Bank of America and Citi Q3 2026):
Nubank: 19.5 cents (its reported efficiency ratio for Q2 2026), on a record $1.1B in net income and a 33% return on equity (Nu).
Bank of America: 66 cents (its reported efficiency ratio for Q3 2026): $19.7B of noninterest expense on $30.0B of revenue, for $6.9B in net income and a 12.6% ROTCE. Excluding a $1.2B regulatory settlement charge, the adjusted ratio was 62% (Bank of America).
Citi: 58.5 cents (its reported efficiency ratio for Q3 2026): $13.95B of operating expenses on $23.8B of revenue, with an 11.5% RoTCE (Citi).
Why the gap is so wide.
No branches, one core platform. Nu serves 139M customers across Brazil, Mexico and Colombia from an app. 83.5% of them are active monthly (Nu).
High-yield revenue. Nu's net interest margin is 22.9%, driven by credit cards and unsecured loans. When each dollar of assets earns that much, the cost ratio looks small almost automatically.
AI on the cost line. AI agents now handle more than 60% of customer-support conversations in Brazil "at or above human parity," and Nu's in-house model, NuFormer, runs underwriting on three loan portfolios (same source).
Why it's not a fair fight (yet). These are three ratios with different definitions and accounting regimes (Nu reports under IFRS, the US banks under US GAAP), and Bank of America's headline figure includes a $1.2B regulatory settlement charge that it excludes from its adjusted figures: the adjusted 62% is a non-GAAP measure. Its group ratio also includes a trading business and an investment bank, where pay is the main cost. Look at its Consumer Banking segment alone and the ratio is 50%: $5.8B of expenses on $11.6B of revenue, run through 3,531 financial centers (Bank of America). Nu's ratio also moves around: it rose from 17.6% in Q1 as marketing, real-estate and international costs landed in Q2. And a high-margin book carries higher credit risk: Nu's 90+ day NPLs were 6.9%. Even so, Bank of America's Q3 2026 net income ($6.9B, after the charge) was about 6x Nu's Q2 2026 ($1.1B). Scale still wins on absolute dollars.
Our take. The lesson for incumbents isn't "be Nubank." It's to find the parts of your bank where cost-to-serve is set by habit rather than by need. Servicing chats, card disputes and onboarding are where Nu's approach transfers best. Citi is the bank to watch here: its ratio improved from 64.7% a year ago (a quarter that included a one-off $726M goodwill charge), and it's aiming for 11–13% returns on the way to a 14–15% target (Citi).
What happened. Trust Bank, the digital bank backed by Standard Chartered and FairPrice Group that opened in September 2022, turned a profit in March 2026. That's just over three years after launch. It still expected a loss for Q1 as a whole, and it didn't disclose the base numbers for March. But the direction is clear: in March, risk-adjusted revenue (revenue after credit losses) grew 67% year on year while costs fell 9%.
The full-year picture. For FY2025, revenue rose to S$135M (from S$97M), costs fell to S$146M (from S$157M), and the loss narrowed to S$54M from S$93M (The Business Times). Our take: revenue growing about 39% while costs shrink is the combination every bank CFO wants and few get.
How the funnel works. The cost side starts with acquisition:
Referrals. Every customer gets a referral code. At launch, when a friend signed up, both got a FairPrice voucher instantly, in the app. Today the reward is an in-app cashback scratch card (Trust, Refer and Earn). More than 70% of new sign-ups came through referrals (Fintech News SG).
Speed beats size. Trust's CMO, speaking to Campaign Asia, put it this way: customers are comfortable with a smaller reward if it arrives instantly. She put Trust's cost to acquire a customer at roughly one-seventh of the traditional industry norm (Campaign Asia, 2025), a figure Trust's CEO has also cited (The Straits Times). It's the bank's own claim.
Distribution it didn't pay for. FairPrice is Singapore's largest retailer, so Trust started with in-store promos and shopper data from day one (same source).
From sign-ups to revenue. Cheap customers only matter if they stick around and use the product. The signals Trust points to:
Cards are used about 25 times a month per customer, and 30% of deposit balances sit with customers who have their salary paid in (BT).
Roughly 170,000 customers treat Trust as their main bank. More than a million customers make it Singapore's fourth-largest bank by customer count (Fintech News SG).
Loans rose to S$1.2B from S$800M, while credit losses grew more slowly than the asset book (29% vs 46%), per the CEO (BT).
Our take. A referral reward is marketing you only pay for when it works. The clever bit isn't the reward, it's how fast it arrives. If you're an incumbent whose "refer a friend" bonus shows up 60 days later on a statement, you're paying a similar price for much less effect. The open question for Trust is the next step: turning a million cheap sign-ups into more of those 170,000 main-bank relationships. That's where deposits, card spending and lending all come from.
