Speed vs. Diligence Is a False Trade-Off

Every credit committee has a version of the same sentence. Go faster and something gets missed. It sounds like prudence, it is occasionally true of a specific file, and as a general rule it does not survive contact with the data.

Speed and diligence in lending are not opposing forces. Federal Reserve Bank of New York Staff Report 836, published in February 2018, found that technology-based mortgage lenders processed applications about 20 percent faster than other lenders while their default rates ran about 25 percent lower. The authors found no support for a lax-screening explanation.

The interesting part is the mechanism underneath it.

Does faster underwriting mean weaker underwriting?

Andreas Fuster, Matthew Plosser, Philipp Schnabl and James Vickery measured this across the US mortgage market in Federal Reserve Bank of New York Staff Report 836, February 2018. Using loan-level application and origination data from 2010 to 2016, they found technology-based lenders processed applications about 20 percent faster, roughly 10 days, controlling for detailed loan, borrower and geographic characteristics.

On defaults the same paper is blunt. "Faster processing does not come at the cost of higher defaults," the abstract states, and on FHA loans the authors report default rates about 25 percent lower even with detailed loan controls. Their findings "speak directly against the 'lax screening' hypothesis," and they report no evidence that the faster lenders targeted marginal borrowers.

The association runs the wrong way for the committee sentence. Faster went with better, on the riskiest slice of the book.

Where does a slow file lose information?

What makes a file slow is usually what makes it thin.

Manual intake collects evidence in the form of documents an applicant hands over. A document is point-in-time, supplied by the party with the most interest in its contents, and detached from the institution that issued it. Verifying it properly means going back to that institution, which is the work that creates the queue.

Ontario's regulator sets that standard explicitly. The Financial Services Regulatory Authority of Ontario, in guidance effective September 28, 2023 on detecting and preventing mortgage fraud, directs licensees to verify employment and income through multiple sources, contact employers directly, and review original or source documents. Read as a work order, that is days of phone calls per file. The standard is right. It is also why files sit.

Fannie Mae's Mortgage Fraud Program, in a July 1, 2021 alert, listed 63 apparently fictitious California employers appearing on loan applications, noting that "paystub templates are similar for various employers across other (involved) loan files." Those files were not underwritten too fast. They were underwritten against paper.

Equifax Canada reported on April 15, 2026 that first-party fraud rose 31 percent year over year between Q4 2024 and Q4 2025, and that inside banking and deposits, falsified financial information went from 1.5 percent of first-party cases to 21 percent. Falsified financial information is a document problem. It does not survive a pull from the institution holding the account.

What does the secondary market already treat as more reliable?

Fannie Mae answered this almost a decade ago. Under the Day 1 Certainty initiative announced October 24, 2016, lenders receive freedom from representations and warranties on income, asset and employment data validated through the Desktop Underwriter validation service, with borrowers saving time "by using electronic data versus collecting documents such as paystubs, bank statements, and investment account statements."

Two verification paths, and what each one proves. The difference is provenance:

	Document-based	Source-based
Where the data comes from	The applicant	The institution holding the account or record
Freshness	The date the document was generated	The date of the request
Typical elapsed time	Days, dependent on chase cycles	Minutes
Fannie Mae rep and warranty relief on validated components	No	Yes, per the October 2016 Day 1 Certainty announcement

The party carrying the credit risk offers relief on the fast path and not the slow one.

So what is the real trade-off?

Here is the view a cautious committee would soften. Time in queue is not a control. It is a control failure that has been renamed.

A file open for eleven days is not eleven days better verified than a file that closed in one. It is one verification, aged eleven days, beside conditions gone stale and an applicant whose interest is cooling. Time in queue belongs on the credit dashboard next to delinquency and loss rates, argued about in the same meeting. Filed under operations instead, a risk metric quietly becomes a productivity metric.

None of which makes automated verification frictionless. Some institutions will not connect. Some applicants decline and mean it. Files still fall back to documents, and the honest design question is what the fallback path does with them. Days spent waiting have almost no relationship to how much is known about a file.

What we don't know

The New York Fed result is the strongest evidence available, and it has real limits.

It is US mortgage data from a specific window: applications from 2010 to 2016, with FHA performance observed through 2017. That period had rising house prices and low unemployment, generous to almost any underwriting approach. The 25 percent default gap comes from FHA loans, not the whole market. And the authors are careful about causality: "we do not attach a strong causal interpretation to our results, given the reduced-form nature of our analysis." Faster lenders had lower defaults. The paper does not claim speed caused it.

Nothing in that study covers Canadian auto lending, equipment finance or private mortgages. On the data itself, FinRegLab's July 2019 empirical research found the predictiveness of cash-flow scores and attributes generally at least as strong as traditional credit scores and bureau attributes, standing alone. That is a real finding and a modest one. It supports the claim that source data is sound, not that speed is safe.

What the evidence rules out is the version of the trade-off most committees run on, where added days are treated as added rigour without anyone measuring what those days produced.

Common questions

Is there a trade-off between speed and diligence in lending?
Not in the strong form usually assumed. Federal Reserve Bank of New York Staff Report 836, February 2018, found technology-based mortgage lenders were about 20 percent faster with default rates about 25 percent lower on FHA loans. The paper establishes association, not causation.

Did faster lenders simply screen out risky borrowers?
The authors tested that. Federal Reserve Bank of New York Staff Report 836, February 2018, reports no evidence that technology-based lenders targeted marginal borrowers, and states that its findings speak directly against a lax-screening hypothesis. Cream-skimming does not explain the gap in that sample.

Why does manual verification create gaps?
Manual verification depends on documents supplied by the applicant, point-in-time and detached from the issuing institution. Equifax Canada reported on April 15, 2026 that falsified financial information rose from 1.5 percent of first-party fraud cases in banking and deposits in Q4 2024 to 21 percent a year later.

Does anyone treat source data as more reliable than documents?
Fannie Mae does. Its Day 1 Certainty initiative, announced October 24, 2016, grants freedom from representations and warranties on income, asset and employment data validated electronically through the Desktop Underwriter validation service, relief not extended to the equivalent paper documentation collected from borrowers.

What are the limits of the New York Fed finding?
It covers US mortgage applications from 2010 to 2016 with FHA performance through 2017, a benign credit period. The default comparison is FHA-specific. The authors explicitly decline a strong causal interpretation. It says nothing directly about Canadian auto, equipment or private mortgage lending.

---

Carousel verifies at source inside the intake flow, so speed and file quality move together. See how verification fits your flow