The bank branch and the manager who could make an exception

There is a short answer about the local bank branch and a useful one, and they are not the same. What follows is the useful one.

Discretion sat locally and it was real

For much of the twentieth century a branch manager held delegated authority to approve loans and overdrafts within limits. That authority rested on soft information: knowledge of a customer's trade, their family, the state of the local economy and whether they were good for it.

Credit scoring moved the decision to a model applied centrally, using data that can be verified and compared across millions of cases. The gain in consistency is genuine, and so is the loss of any route for a case the model cannot see.

Soft information does not travel

The knowledge that a particular builder always repays late in winter and clears by spring cannot be entered into a scoring field. Research on relationship lending has generally found that local knowledge matters most for small businesses and opaque borrowers, though findings vary.

What is easy to forget is that when decisions centralise, that category of borrower loses first, because they are precisely the ones the data describes poorly. This is a well-recognised problem in banking research rather than a nostalgic complaint.

The bias cuts both ways

Manager discretion also allowed prejudice, favouritism and lending based on who somebody's father was. Scoring models are consistent, auditable and can be tested for discriminatory outcomes, which discretionary judgement cannot.

For a while, models also encode the biases of their training data, and that failure is harder to see because it looks like arithmetic. Neither system is neutral, and pretending either one was is the main error in this argument.

Closures concentrate on the places least able to absorb them

Branch networks have shrunk sharply in most developed markets, with closures falling disproportionately on smaller towns and lower-income areas. Access to cash, in-person help and business banking are the practical casualties, and cash-dependent groups skew older and poorer. Several countries have introduced shared banking hubs, post office banking arrangements or access-to-cash rules in response.

The policy responses are recent, uneven and their adequacy is not yet established.

The branch was also a security theatre that worked

A heavy building, a counter and a person you could name made a bank feel accountable in a way an app does not. Fraud has moved almost entirely into channels where the customer is alone with a screen and under time pressure.

Look closer and authorised push payment fraud in particular exploits the absence of a moment where someone could ask why you are transferring the money. Some banks have reintroduced friction and warnings for exactly this reason, effectively rebuilding the pause the counter used to impose.

The record here is thinner than it should be, and much of it was never written down.

What the counter did for the unbanked

Opening an account, resolving an identity problem or explaining an unusual situation is much harder without a person and a desk. Digital-only onboarding excludes people with irregular documents, no fixed address or limited digital confidence. Basic bank account obligations exist in several jurisdictions precisely to address this, with mixed effectiveness.

A service that assumes a smartphone and a passport is not universal, however convenient it is for those who have both.

The takeaway

The model is fairer to the average case and has nothing at all to say about the unusual one, which used to be exactly what the manager was for.

What replaced it works. That is not the same as it being nothing.