BrokerVA
BrokerVA Team · October 27, 2026

Will AI Replace Underwriters? Why 70% Isn't the Whole Solution

Will AI Replace Underwriters? Why 70% Isn't the Whole Solution

The evidence so far doesn't show it. MQube, a UK mortgage vendor, says its AI reached the same outcome as human underwriters in 70% of nearly 1,000 complex cases, but that measures agreement in a test, not replacement. Here is what the study shows, as part of our series on the human element in mortgage.

What Did MQube's Study Find?

According to Financial Reporter, the study examined nearly 1,000 complex mortgage cases that experienced human underwriters had already decided. Using the same information given to those underwriters, MQube's AI re-underwrote the cases and reached the same outcome as the humans 70% of the time. The company also says AI could cut underwriting time from four hours to 25 minutes.

MQube's CEO, Stuart Cheetham, framed it as a shift in role. The point, he said, is "improving speed and consistency and importantly not removing human judgement." IFA Magazine reported in March 2026 that the analysis suggests human underwriters could move from processors to auditors.

Is "70% of Underwriting" the Same as "70% Agreement"?

No, and the coverage shows the difference. Financial Reporter describes the AI reaching the same outcome as human underwriters in 70% of cases. IFA Magazine describes AI accurately performing around 70% of traditional underwriting tasks.

Those are different claims. "The AI agreed with the human decision in 70% of cases" is a statement about outcomes in a test. "AI performs 70% of the work" is a statement about how work is divided. Neither says AI can underwrite 70% of loans without a person, and the study as reported doesn't claim that.

What Can't the Study Tell Us?

This section is our analysis. Several limits matter, and most apply to any vendor study:

  • It's a vendor's analysis. MQube sells the technology it tested, so it has an interest in the result.
  • It's UK data. Underwriting rules, products, and investor guidelines differ from the U.S., so treat the number as an illustration, not a U.S. benchmark.
  • The human decision was the benchmark. The AI was judged by whether it matched experienced underwriters, which assumes the human outcome was right. The coverage doesn't say who was right in the 30% where they differed.
  • They were complex cases, which is interesting, but the reports don't say how the cases were selected or how the 70% breaks down by type.
  • A re-underwrite isn't live work. The AI saw the same file the human saw, after the fact. Real files change, arrive incomplete, and generate exceptions.
  • Agreement isn't accountability. Matching a decision doesn't answer who answers for it.

None of this makes the study wrong. It makes it a useful data point about speed and consistency, and a thin basis for staffing decisions.

Why Do Accountability and Borrower Expectations Still Matter?

Because decisions have owners. Freddie Mac's Guide Section 1302.8 has required an AI governance framework since March 3, 2026, and Fannie Mae's Lender Letter LL-2026-04 took effect August 6, with documented oversight and accountable people at the center. See Who Is Accountable When AI Touches a Mortgage File?.

Borrowers' expectations point the same way. In Cotality's 2026 survey, 55% of U.S. buyers preferred working with a person to secure a mortgage. Cotality sells AI-enabled products, so weigh its interest. See our summary.

What Does the Claim Mean for Roles?

Our view is that the likely shift is from doing every step to reviewing, handling exceptions, and answering for results. That is MQube's own framing: processors becoming auditors. If that proves right, the skills that matter more are judgment, exception handling, and knowing where a wrong answer costs the most.

It's also consistent with Freddie Mac's finding that technology works best as a strategic enabler alongside changes in process and staffing, in its 2024 Cost to Originate Study. Freddie Mac sells digital tools, so weigh that. For the role-shift view on conditions, see AI Clears a Condition in Under 2 Minutes. What Should Your Team Do Instead?.

What Should Operations Teams Do With a Claim Like This?

These steps are our recommendations.

  1. Treat vendor numbers as hypotheses. Ask what was measured, on what, and against what benchmark.
  2. Test on your own files, with a baseline. See How to Run a 30-Day Pilot With a Mortgage Operations Provider, which works for tools as well as providers.
  3. Decide where the human checkpoints go. See How to Add AI Without Losing the Human Check.
  4. Build review skills, because reviewing well is a skill, not a formality.
  5. Track exceptions. How often and why a person changes the output tells you where the real work is.

How Does BrokerVA Think About This?

Credit decisions stay with the lender's underwriters, and accountability for the loan stays with you and your licensed people. BrokerVA's five core roles, Setup, Disclosure, Closing, Funding, and Post-Closing, support the stages around them. Our Setup role, for example, reviews files for completeness and prepares them for underwriting submission, and the later roles handle the disclosure, closing, funding, and post-closing work. We may be able to tailor roles to your business, and we have partnered with Addy AI for LOA services.

Whatever happens to underwriting, someone still has to make sure files are complete, disclosures are valid, figures match, and records hold up. Our goal is to lower your cost to originate. Every client gets a dedicated Coach for a 6-week implementation, then ongoing coaching and regular check-ins. We can't promise a specific saving, since it depends on your operation. On a call, we'll walk through your numbers and show you our cost calculator live. Pricing is customized to your volume and the roles you need. See how BrokerVA works.

Frequently Asked Questions

Can AI do 70% of mortgage underwriting? A UK vendor's study, as reported, found its AI reached the same outcome as human underwriters in 70% of nearly 1,000 complex cases. That measures agreement in a test, not the share of loans AI can underwrite without a person.

Will AI replace mortgage underwriters? The study doesn't say so. MQube's own framing is that underwriters' roles could shift toward review, and its CEO said the aim is not removing human judgment.

Does the study apply to U.S. lenders? It's UK data from a vendor, and U.S. rules and guidelines differ, so treat it as an illustration.

What should lenders do about claims like this? Ask what was measured and against what, test on your own files with a baseline, and decide where human checkpoints belong.

Sources

Want to see where your files need people, whatever the technology? Book a call with BrokerVA and we'll walk through your numbers and show you our cost calculator.

BrokerVA | NMLS #1977844