The Correction Window: What Friction Was Quietly Doing

Financial processes used to contain intervals. Days between applying and receiving money. Days for a payment to settle. A gap before a transaction posted. None of these was designed as a protection and every one functioned as one, because inside an interval a mistake can be found and undone, and outside it the mistake is permanent. Removing delay improved every measured outcome — conversion, satisfaction, cost — and removed something nobody had a line item for. The benefits of speed are measured immediately and the costs appear later under other headings, which makes the trade look better than it has been.

The intervals that existed

Interval	Then	Now	What it held
Application to decision	Days	Seconds	Review, verification
Decision to disbursement	Days	Minutes	Second thoughts, error catching
Payment to settlement	Days	Instant	Recall, reversal
Transaction to posting	Days	Immediate	Batch review
Order to fulfilment	Days	Hours	Cancellation

Not one of these intervals existed because someone wanted it. They were consequences of paper, batch processing, postal mail, and manual handling — and per our distribution analysis, removing them is where most of the genuine cost reduction in financial services has come from.

Which is why removing them was obviously right and why the removal was never evaluated as a trade. You don't weigh the benefits of eliminating an accident.

None of these intervals existed because anyone wanted it. All of them were doing something anyway.

Four things delay was doing

Catching fraud before money moved. A transfer taking two days can be recalled; one settling instantly generally can't. Per our reversibility analysis, that's the mechanism behind the payment fraud in our fraud guide — the reason those losses are so severe isn't sophistication, it's that the window closed. Detection speed determines recovery, and the time available to detect fell to near zero.

Catching operational error. A batch that sits overnight can be reviewed; one that clears immediately can't. Errors that used to be caught before leaving the building now leave it, which converts an internal correction into a customer-facing incident.

Allowing reconsideration. A borrower who applied on Tuesday and received money Friday had three days in which the urgency faded, a family member asked a question, or a better option appeared. Per our prediction analysis, borrowers are systematically optimistic at the moment of decision — and the interval was a correction for exactly that bias, applied without anyone intending it.

Forcing verification. A process taking days had room for a check that a process taking seconds doesn't. Per our verification analysis, much of this has been genuinely replaced by source-connected data — which is the one function that got a real substitute rather than simply disappearing.

Why the trade looks better than it is

The measurement structure, and it's the one our measurement analysis describes.

	Benefits of speed	Costs of removing the window
When	Immediate	Later
Measured as	Conversion, satisfaction, cost	Fraud losses, credit losses, complaints
Attributed to	The speed decision	Fraud, or the model, or the customer
Owned by	Product	Risk, operations

Row three is the mechanism. A loss that occurs because a transfer couldn't be recalled is recorded as a fraud loss. A default on a loan that a borrower would have reconsidered is recorded as a credit loss. Neither is attributed to the removal of the interval that would have prevented it, so the speed decision never appears on the wrong side of any ledger.

And row four completes it: the function that gains is not the function that pays. Which per our incentive analysis is the standard condition for a change that continues past the point where it's still beneficial.

None of this means speed was wrong. It means the evaluation was structurally biased toward it, and a decision taken under a biased evaluation may have gone further than a neutral one would.

The borrower's window

The application that deserves most attention, because it's where speed is sold as a consumer benefit and where the removed function was most protective.

Instant credit decisions with instant disbursement compress the entire sequence into minutes. What that removes:

• The interval in which urgency subsides. Per our time preference analysis, a constrained household's discount rate is highest at the moment of the crisis — and falls once the immediate pressure passes. A three-day interval let the decision be made at a lower rate than the one that prompted it.
• The chance to find an alternative, per our search analysis.
• The chance to ask someone.
• The moment where an arrangement could have been offered instead, per our arrangements guide — which would frequently have removed the need to borrow at all.

Which produces the sharpest version of the finding. Our time preference analysis argued that high-cost borrowing is frequently a rational response to a severe constraint, and that the constraint is the thing to address. Instant availability means the decision is now always made at the peak of the constraint, with no interval in which anything could address it.

That isn't an argument for slow lending. Someone facing eviction on Friday is not helped by a three-day wait — and per our time-poverty analysis, slow processes impose real costs that fall hardest on people with the least slack. It's an argument that the interval was doing something, and that nothing replaced it.

Where friction was simply a cost

Stated properly, because a report about the value of delay could easily become an argument for it.

Most friction was doing nothing useful at all. Per our time analysis, effortful processes are a regressive delivery mechanism, and per our affordability analysis, every additional step loses applicants who would have qualified.

Friction that was pure cost:

• Documentation the applicant assembles that the institution could have obtained.
• Business-hours-only processes.
• Steps requiring a branch visit for something doable remotely.
• Repeated provision of the same information.
• Waiting with no work being done — a queue rather than a process.

The distinction that matters: friction imposed on the customer versus an interval before irreversibility. The first is cost. The second is the correction window. They're entirely different things and the modernization of financial services removed both while describing all of it as removing friction.

Which gives the useful design principle: a process can be effortless for the customer and still contain an interval. Those aren't in tension — an application that takes ninety seconds and disburses tomorrow morning has removed the cost and kept the window.

Replacing the function, not the delay

The constructive position: identify what the interval was doing and provide it deliberately, rather than reintroducing general delay.

• A short cancellation period after a credit decision. The customer gets an instant answer and can cancel before or shortly after funds arrive — which preserves the reconsideration function at almost no cost to anyone who doesn't want it.
• Selective holds rather than universal ones. Per our authorization analysis, applying a delay to unusual transactions rather than to all of them targets the cases where the window matters.
• Confirmation steps where errors concentrate — new payees, changed details, amounts outside a pattern. Per our fraud guide, a cooling period on new payees defeats a whole fraud class without slowing anything else.
• Reversal rights that survive faster settlement, which per our reversibility analysis is what several jurisdictions are now considering for instant rails.
• Offering the alternative at the decision point. Where a lender can see that an arrangement would serve better than a loan, presenting it in the instant flow does what the three-day interval used to do accidentally.

The first and third carry the most value for the least cost, and both are targeted: almost every transaction stays instant, and the small share where the window mattered still has one.

And per our timing analysis, the general form of this is suspend rather than reverse — a brief hold that can be released costs little and preserves the ability to stop, which is the property that disappeared.

A test for any friction

Practical, and applicable to any process being accelerated:

1. Name what happens during the interval. If nothing, remove it — that's cost.
2. If something happens, is it done by the customer or by you? Customer effort is cost; institutional work may be function.
3. What gets caught during it? Errors, fraud, reconsideration — name them.
4. Is that outcome reversible afterwards? If yes, the window matters less.
5. Could you provide the same function without the delay? Usually yes, targeted.
6. Who measures the benefit and who bears the cost? Different answers mean the evaluation is biased.
7. Measure both sides deliberately, since one is easy and one isn't.

Steps three and six are the ones usually skipped, and they're where the analysis lives. An organization that can't name what its intervals catch has no basis for deciding whether removing them is safe.

And step seven is the honest requirement: the costs of removing a window are measurable if you decide to measure them — by comparing loss rates before and after, or by holding the interval for a random sample during a change. The reason they usually aren't measured is that nobody owns the question, not that it's hard.

The strongest objections

"This is nostalgia for slow finance." The objection to take seriously, since arguments for friction have historically been arguments for incumbents. The response is that most friction was pure cost and should have gone — the report says so explicitly — and that the argument is for targeted mechanisms rather than general delay. A cancellation period is not a return to paper.

"Speed helps the people who need help most." Substantially true. Per our time analysis, slow processes exclude people with the least schedule control, and per our time preference analysis, someone facing an imminent consequence is genuinely better off with money today. The report doesn't dispute this and notes that the removed function was the one protecting those same people from decisions made at peak constraint. Both are true and they pull in opposite directions, which is what makes it a trade rather than an improvement.

"You can't show the window was doing anything." Correct for most of it. The fraud case is well evidenced — recoverability falls sharply with time and that's not disputed. The reconsideration case is inference from the prediction literature rather than direct measurement, and it would be testable by anyone running an instant flow, which is the report's main recommendation.

Testable implications

1. Fraud losses should rise with settlement speed, across rails and jurisdictions — the best-evidenced prediction.
2. A cancellation period should be used by a small but non-trivial share of borrowers, and those who use it should have performed worse than average had they not.
3. Instant-disbursement borrowers should show worse outcomes than delayed-disbursement ones at equivalent risk — directly testable by randomizing disbursement timing.
4. Errors caught should fall as batch intervals shrink, and customer-facing incidents should rise correspondingly.
5. Offering an arrangement inside an instant flow should divert a meaningful share of applicants, per our arrangements analysis.
6. Selective holds should capture most of the protective value of universal delay at a fraction of the cost.

The third is the decisive one and it's straightforward. Randomize whether approved applicants receive funds instantly or the next morning, and compare performance. If the delayed group performs better at equivalent assessed risk, the interval was doing credit work that no model was doing — and the size of the difference is the value of what was removed.

The conclusion we'd hold: the intervals in financial processes were accidents that had become load-bearing, and they were removed by an evaluation that could see the benefits clearly and the costs only in other people's numbers. The answer isn't slower finance. It's deciding deliberately what the delay was doing and providing that on purpose, for the cases that need it.

Frequently asked questions

---

This report presents an analytical framework and the authors' interpretation; it is not financial, legal, or policy advice. The functions attributed to process intervals are argued from structure and from adjacent evidence rather than measured directly, and no estimate is offered of the magnitude of any effect described; the implications identified as testable are hypotheses.