The banking skills worth practicing when a model scores every loan

Announcing an AI partnership that reaches into wealth advice and financial-crime work, HSBC’s Group CEO Georges Elhedery described the aim as “keeping human judgement, decision-making and accountability at the core” (Georges Elhedery, HSBC (news release), opens the source). At a much smaller bank, Norway Savings Bank’s chief risk officer Dawn Mugford put the staffing side plainly: “Where it makes sense, you’d much rather have people doing analysis [and other high-value functions].” (Dawn Mugford, ProSight Financial Association (Insights), opens the source) Neither was describing a credit policy. Our interview translation: when a model can score the application, the work left for a person is the analysis behind the decision. Show you can rebuild the cash a borrower actually produces, after tax, capex and working capital, and size the debt before you lean on collateral.

When the CFPB issued guidance on AI-driven credit denials, its then Director, Rohit Chopra, said:

That was a rule for lenders’ notices, not a hiring standard, and the one regulator quoted here. Our inference: a banker who can say which factors drove a decline, how much each mattered and what would change it, without promising an approval, is more useful to the customer and to an examiner than one who says “the system decided”.

Standard Chartered’s Roger Wang builds an AI assistant for relationship managers. Describing fund figures it produced from no approved source, he says, “But the numbers were plausible, not factual.” (Roger Wang, Standard Chartered careers (employee story), opens the source) He stopped feature work and traced the cause. Our lesson for credit and operations roles: a dashboard average can look calm while one segment breaks. Ask what changed in the inputs, who was affected, and what happens to the decisions already made.

Explaining why scams aimed at older customers work, Peter Stenehjem, CEO of First International Bank and Trust, said they manipulate emotions and force action: “They get people to act quickly and urgently, and that’s how mistakes get made.” (Peter Stenehjem, ICBA Independent Banker, opens the source) He was describing customers under pressure. Our inference is that the same pressure reaches the banker: a low fraud score, a familiar voice and an important client all saying “send it today”. The skill is to slow down, test the payee and the pattern against the client’s own history, verify through a channel you already trust, and hold the line when the relationship pushes back.

Writing about a branch that an NPV analysis said to close, an analysis that gave branches no credit for the credit cards they open, Jamie Dimon wrote, “I think, for the most part, NPVs might work, but they don't always work. You need to use your common sense.” (Jamie Dimon, JPMorganChase 2024 Annual Report (letter to shareholders), opens the source) He was talking about branches, not AI. Our inference: a pricing model that sees one product can be precise and still wrong. Strong candidates add up what the whole relationship earns, test whether it would really leave, and refuse to trade away credit structure for it.