All letters

May 1, 2026

The checking account was never about cash

What banks are actually defending when they compete on deposit rates

I’ve come to think of the checking account as two different assets. One is the float, which banks monetized for a century by paying depositors almost nothing while earning meaningfully more on overnight reserves. The other is transaction history: twelve months of behavioral data that tells you more about creditworthiness than a credit bureau pull. Banks have treated the second as a byproduct of the first. That assumption is worth reexamining as the float reprices.

How banks built NIM on data they didn’t fully account for

The traditional model is straightforward: take deposits at low rates, deploy capital at higher rates, keep the spread. Checking accounts were the best version of this because they were essentially interest-free funding. During the 2022-2023 rate cycle, banks passed through 49.3% of a 525 basis point Fed rate increase to depositors. The other half stayed with the bank. That gap wasn’t built on credit skill. It was built on customer inertia.

What came with the float was never separately priced. When a bank extends a mortgage, the underwriter considers income and credit history from the bureau. What they’re increasingly also using is twelve to twenty-four months of transaction history from the checking account: the regularity of income deposits and the periods where the balance ran close to zero. The behavioral signal in that data goes beyond what bureaus capture. Payment history from a bureau shows whether you’ve repaid credit before. Transaction history from a checking account shows whether you have the cash flow stability to repay it in the first place. The borrowers most affected by that distinction are precisely the ones whose stability doesn’t show up cleanly in a bureau pull. Banks have had near-exclusive access to that signal because that’s where most Americans’ money sat.

The access was always a side effect of the float mechanism. Both assets are moving as the float reprices.

What the fintech sweep model actually does

Wealthfront generated $309 million in revenue in 2025. Roughly 75% came from cash management: sitting between a network of program banks and their customers, capturing a portion of the spread banks had kept, and passing the rest to customers as higher yield.

The program banks in that network include Wells Fargo and Citibank. They pay Wealthfront a gross rate on deposited funds. Wealthfront passes 3.30% to customers and keeps the difference as a management fee. The bank holds the deposit and the regulatory overhead. Wealthfront holds the customer relationship and the revenue.

Apple ran the same play at a different scale. Its high-yield savings account launched at 4.15% in April 2023 and accumulated $10 billion in deposits in its first four months. Goldman Sachs was the program bank. Apple had the customer relationship. Goldman’s consumer division lost more than $7 billion since 2020. When JPMorgan stepped in to take the card, customers still thought of it as an Apple product.

The pattern is consistent: a technology company takes the primary relationship and a share of the spread. The bank takes the deposit, the compliance burden, and declining margin.

What banks lose beyond the funding

U.S. bank deposits fell $874 billion in the twelve months to June 2023, the largest single-year decline since FDIC records began. Almost all of it moved into money market funds and higher-yield alternatives. The funding cost impact was immediately visible in earnings.

The data impact was quieter, and will take longer to measure.

Every customer who moves their primary banking relationship to a sweep account is also beginning to move their transaction history. Not all at once. ACH infrastructure still runs through banks, and years of existing behavioral data remain on file. But new transaction data starts accruing somewhere else, and the signal that trains credit models is the forward-looking one.

The institutions that will price credit risk most accurately in 2030 are the ones with the richest transaction datasets from 2026 to 2030. That’s how credit model training works. When the primary account relationship moves, the behavioral data pipeline moves with it, on a lag.

Wealthfront filed for IPO in late 2025 with $88 billion in assets under management. The scale is still modest relative to the major banks. The trajectory is not.

The data relationship is harder to recover than the funding relationship

Banks can replace deposit funding. Wholesale markets and securitization both exist. The cost increases, but the mechanism is real. JPMorgan reduced its wholesale funding by $97 billion in a single quarter when deposit recovery made it possible.

There is no equivalent mechanism for transaction data. If you’ve lost five years of a customer’s daily financial behavior to a competing platform, you cannot buy it back cleanly. You can acquire a fintech that has it, at a premium that reflects the scarcity. You can partner with data aggregators, accepting the adverse selection problems those arrangements carry. Neither is the same as the organic accumulation of behavioral signal that comes from being the account where someone’s paycheck lands.

This is the asymmetry that deposit rate conversations tend to miss. Both assets move when the primary account relationship moves. Funding can be replaced. Behavioral data, at the scale and quality of a primary checking relationship, generally cannot.

What to do differently

Most deposit strategy conversations anchor on rate: compete on yield and compress the margin, or hold rates and accept some outflow. Both are defensible approaches to the funding problem.

The transformation question is whether you’re retaining the behavioral data relationship or just the deposit balance. Those two things are not always the same. A customer who keeps a small checking balance at your institution while moving their primary spending to a sweep account is technically still a deposit relationship. The transaction signal is somewhere else.

The strategic frame: in your next deposit strategy review, ask whether the accounts you’re retaining are primary relationships or secondary balances. That distinction matters for credit model quality in a way that deposit balance alone doesn’t capture.

The specific test for whoever has hands on the credit infrastructure: look at your last two years of credit model performance and check whether accounts with the richest transaction history (five or more years of primary banking relationship) outperform bureau-only models at the same credit tier. If they do, and you’re losing those relationships to sweep platforms, you’re not just losing cheap funding. You’re losing model performance, on a lag.

The float is repricing. The behavioral data compounds in the platform that holds the primary account. The institutions that understand which of those two assets matters more for the next decade of credit underwriting are building toward a different answer than the ones running deposit beta analysis.

- Shish

Sources

The main data points in this letter, for those who want to go further:

Want the next letter the morning it goes out?

Subscribe
The checking account was never about cash · Shish Mukherjee