The Disappearance of Place: What a Location Was Doing

A branch was expensive and most of what it did could be done better elsewhere, which is why they closed. But it was doing three separable jobs and nobody separated them before deciding. It established identity. It held knowledge that no file contained. And it was somewhere you could take a problem to a person who could decide. The first two got replacements that are genuinely better. The third got a phone queue — and the difference between those isn't the medium, it's that one had authority and the other has a process.

Three jobs

Function	Replaced by	Better?
Establishing identity	Electronic verification, document capture	Yes, substantially
Holding local knowledge	Standardized portable data	Better for most, worse for some
Escalation to a decider	A contact centre	No
Transaction handling	Digital channels	Yes, overwhelmingly

The fourth row is why branches closed and it isn't controversial — handling routine transactions in a building was enormously expensive and the replacement is better in every respect, including for the people who used branches most.

The analytical interest is that rows one to three were bundled with row four and went with it, and they had different replacement stories. A decision made on the economics of transaction handling determined the fate of three other functions that nobody costed.

The decision was made on the economics of cashing cheques. Three other functions went with it and nobody costed them.

Identity, replaced well

The clearest gain, and worth stating plainly because the rest of the report is more critical.

Per our verification analysis, establishing who someone is was a fixed cost per relationship and a substantial part of why serving small customers was uneconomic. Electronic verification cut it dramatically — which per our pricing analysis is one of the few interventions that reduces the component actually dominating small-loan pricing.

And it's more reliable. A person looking at a document is not a good verification mechanism, and per our fraud analysis, source-connected verification catches things visual inspection never did.

The residual: people whose documents don't match cleanly still need somewhere to go. A name that doesn't align, an address that can't be verified, an identity that electronic systems fail on — per our access analysis, that population exists and is exactly the one that needed the branch for this. The replacement handles the ordinary case better and the hard case worse, which is the pattern that runs through everything below.

Knowledge, replaced differently

The most contested of the three, and the honest answer is mixed.

A local lender knew things no file contained — which trades were seasonal, which employers were stable, which businesses were solid despite thin accounts, who was reliable.

What that knowledge was worth:

• It assessed people the file couldn't. Per our coverage analysis, the file records credit accounts — local knowledge recorded everything else, informally.
• It varied between institutions, which per our monoculture analysis is exactly the property that produced second opinions.
• It didn't port, which is why it couldn't scale and why it disappeared.

And it was frequently discriminatory in ways that were serious and hard to challenge. Discretion tracked who the decision-maker recognized and trusted; a documented rule can be tested for disparate impact and an impression cannot. Any account of what was lost has to hold this alongside the loss rather than romanticizing what came before.

So the replacement was better in the two respects that matter most — reviewability and portability — and worse in one: it sees less. The trade was real and the gains were larger. The problem is that it was treated as a pure gain, so nothing was done about the population the new system can't see.

Escalation, not replaced

The function this report exists to name, because it's the one where the replacement isn't a replacement.

A branch was somewhere to take a problem. Not a routine transaction — a problem. Something had gone wrong, or your situation didn't fit, and you needed a person.

What that provided:

• A person who could see the whole situation rather than the part that fits a form.
• Frequently, authority to decide — to waive something, to make an exception, to escalate internally.
• A record of having tried, with a named person.
• Continuity, so the second visit didn't start over.
• A path for the case no rule covers.

What replaced it: a queue, a representative following a process, and an escalation path that frequently doesn't reach anyone with more authority.

And per our time analysis, the cost structure inverted. A branch visit was expensive in travel and cheap in waiting; a contact centre is free to reach and expensive in hold time during business hours — which shifts the burden onto exactly the people whose hours are least available.

Authority is the variable

The sharpest version of the finding, and it isolates what actually changed.

The significant difference is not the medium. It's the delegation.

	Branch	Contact centre
Sees your whole situation	Yes	Sometimes
Can make an exception	Frequently	Rarely
Handles the case no rule covers	Yes	No path
Cost to reach	Travel and time	Hold time, business hours
Continuity	Yes	Rarely

Row two and row three are the whole thing. A remote channel staffed by people with real authority would replace the escalation function entirely — the problem isn't that the person is on a phone, it's that they can't decide anything.

Which is the same finding our inbound analysis reached from the collections side: a caller who has to be called back frequently isn't reached again, and delegated authority at the point of contact is what converts a contact into a resolution.

And it explains why the loss is invisible to the institution. An organization measuring handle time and first-contact resolution against its own process is measuring how well the process runs, not how many situations the process cannot address. Per our measurement analysis, the customer who gave up leaves no record of what they needed.

Who bears the loss

The distributional point, and it's the same population every time.

If your situation fits the standard process, you're served better and more cheaply than before. That's most people most of the time, and it's a real gain.

The cost falls on:

• People whose documents don't verify cleanly.
• People whose circumstances are unusual — irregular income per our income analysis, complicated households, recent arrivals.
• People dealing with a complication — a bereavement per our deceased accounts guide, an authority question per our authority guide, a dispute that doesn't fit a category.
• People who need a decision no rule covers.
• People with the least discretionary time, per our time analysis.

That population overlaps heavily with the one the standard data already assesses poorly. Which produces the finding: the same households lose twice from the same change — the assessment can't see them, and the place where a person could have seen them is gone.

And per our monoculture analysis, the two losses share a cause. Standardization improved the average case by eliminating idiosyncratic judgment, and idiosyncratic judgment was what the non-average case relied on.

What would rebuild it

The constructive conclusion, and it isn't branches.

The function needed is authority at the point of contact for situations the process doesn't cover. Which is achievable remotely and cheaply:

• Delegated authority to front-line staff, within defined bands — per our inbound analysis, this is the highest-return change available in any contact operation.
• A defined path for the case that fits nothing, staffed by someone who can decide rather than by an escalation that loops.
• Continuity of handler for anything that takes more than one contact.
• Availability outside business hours, which per our time analysis is what makes a channel real for shift workers.
• Measuring what the process couldn't handle, not just how fast it handled what it could.

The last is the one that makes the rest possible, because the loss is currently invisible. An operation that counted cases where no path existed — rather than counting resolutions within paths that do — would discover the size of the gap it's been unable to see.

And the honest note on cost: delegated authority is more expensive than a script. It requires better-trained staff, it produces variation that needs monitoring, and it reintroduces some of the inconsistency that rules eliminated. That's a real trade and it's the trade that was made silently when branches closed — the argument is for making it deliberately, and for a narrower slice of cases than a branch served.

The strongest objections

"Branches were bad and expensive." Largely conceded and stated throughout. They were an inefficient way to do most of what they did, the transaction function deserved to go, and the local judgment they housed was frequently discriminatory. The argument is about one unbundled function, not about the institution.

"Digital channels serve most people better." True, and the report says so. The claim is about the tail rather than the average — and specifically that the tail is not randomly distributed but concentrated in a population that other parts of the system also serve badly.

"This is nostalgia." The risk worth naming, since arguments about what was lost frequently romanticize arrangements that were worse. The test we'd apply: the recommendation here isn't to reopen anything — it's delegated authority in remote channels, which is cheaper than a branch and would have been an improvement on branches too.

Testable implications

1. Complaint and abandonment rates should be highest for situations that don't fit standard categories, measurable by coding what customers were actually trying to do.
2. Delegating authority to front-line staff should raise resolution and lower repeat contacts by more than it costs — testable by randomizing authority bands.
3. Areas that lost physical presence should show reduced access for thin-file and unusual-circumstance applicants, beyond what the standard data explains.
4. Customers with complications should report more unresolved contacts than customers with routine needs, at the same institution.
5. Institutions measuring cases with no available path should find a material number that current metrics don't capture.
6. Extending contact availability beyond business hours should disproportionately raise resolution for hourly workers.

The fifth is the one worth doing first because it's cheap and it reveals whether the rest matters. Code a sample of contacts by whether a resolution path existed at all, rather than by whether one was followed — an operation that has never done this doesn't know how often its customers arrive at a problem the organization has no way to solve.

The conclusion we'd hold: place was doing several jobs, most got better replacements, and one got a cheaper substitute that lacks the property that made it work. Authority at the point of contact is what a branch had and a queue doesn't, and it's the part that can be rebuilt without rebuilding anything else.

Frequently asked questions

---

This report presents an analytical framework and the authors' interpretation; it is not financial or policy advice. The effects described are argued from the structure of the functions involved rather than measured, and no estimate is offered of the magnitude of any of them; the implications identified as testable are hypotheses.