An overdraft fee customers cannot see coming is a design defect

Public debate about overdraft in the United States has mostly been about price: how large a fee should be, whether it should exist at all and which rules should limit it. Federal policy on the question has moved more than once in recent years and may move again. I think the price debate, though legitimate, misses the design problem underneath it. The customers who pay most overdraft fees are usually not choosing to borrow. They are surprised, because the balance the bank showed them was not the balance against which it charged them. My position is that a bank or credit union should remove the surprise before it argues about the price, and that a fee customers could not reasonably have seen coming is a defect in the product, whatever the fee schedule says.

I learnt this at the counter. In 1999, at a credit union in Duluth, a member came in to ask why she had paid three overdraft fees in one afternoon. Her pay had been deposited on the same day as three small purchases, and the order in which the items posted overnight took her below zero before the deposit counted. I could explain the rule to her. I could not explain why she should have known it.

The balance on the screen is not the balance that posts

Twenty years later, at Corriston Bank, I led a review of our checking fees. In 2019 overdraft and returned-item fees brought in US$96 million, and 8% of checking accounts paid 79% of them. When we sampled those accounts with our complaints team, the pattern was familiar. Most fees followed a moment when the available balance in the app had been accurate on its own terms and misleading in practice: pending debit card holds that had not yet posted, a cheque deposit with part of its funds held, a scheduled bill payment the customer had forgotten and a posting order that differed from the order in which things had happened.

None of this was hidden in the legal sense. Every rule was in the account agreement, and our disclosures had been reviewed many times. But a disclosure describes a mechanism, and customers do not experience mechanisms. They experience a number on a screen. If that number is regularly not the one the bank will charge against, the bank has built a product that produces fees from ordinary behaviour.

Our first repair was to the number. We introduced a projected end-of-day balance in the app, showing pending card authorisations, payments scheduled for that day and held deposit funds as separate lines beneath the available balance. It took longer to build than any fee change, because our core processing had never been asked to share that information in real time.

A day to put it right

The second repair was time. From November 2020 we stopped charging overdraft fees until the end of the next business day, so a customer who saw a negative balance could move money and avoid the fee altogether. We added a buffer, so that no fee applied to an account overdrawn by US$50 or less, and we abolished the returned-item fee, on the reasoning that declining a payment and then charging for the decline was difficult to defend to anyone.

These changes were not free, and the people who opposed them were not wrong about the cost. Our chief financial officer expected fee income to fall by more than half, and in 2021 it fell to US$41 million. He argued for a smaller change: a lower fee with the timing rules left as they were. My view was that a lower fee charged in the same circumstances would still charge customers for the bank's own presentation, only more cheaply. Our chief executive approved the larger change, partly because we could show the board that customers who paid many fees were closing their accounts at twice the rate of others, and taking their direct deposits with them.

In the first full year, complaints about fees fell 64%, closures among the heaviest fee payers fell 31% and losses written off on overdrawn accounts fell from US$9.8 million to US$6.1 million. The bank gave up real revenue. It also lost less to customers who had been leaving angry and leaving overdrawn.

What a fee should still pay for

I am not arguing that overdraft coverage should be free or abolished. Some customers value having an essential payment covered rather than returned, and an institution that takes credit risk on that payment may reasonably charge for it. Federal rules have long required customers to opt in before a bank can charge overdraft fees on everyday debit card and ATM transactions, and many customers make that choice deliberately. A fee can fairly pay for a service a customer understood and chose.

What a fee should not pay for is the gap between what the bank displayed and what it posted. My test is whether a reasonably attentive customer, looking at the bank's own app at the moment of the transaction, could have known that it would cause a fee. Where the answer is no, I would remove the fee before debating its size.

There are mistakes a board considering a similar change should know about. We changed the fees in November 2020, but the projected balance did not reach every customer until the following April, so for five months we had a more generous policy that customers still could not see. I should have put the display first. We also gave branch staff a two-page note on the next-day window and no tool to check whether a particular fee would apply, which turned a customer-friendly rule into a source of argument at the counter.

The same reasoning now shapes my work at Lanternfield, where the shared app shows credit union members pending items and a projected balance by default, and where several of our clients have adopted cure windows of their own. Credit unions differ in their fee policies, and they should. They should not differ in whether a member can see what is about to happen.

The question I would put to a bank or credit union board is not whether its overdraft fee is competitive. It is how many of last year's fees followed a moment when the institution's own app showed a positive balance. If the answer is most of them, the institution does not have a pricing problem. It has a product that charges customers for its own design.