State of Private Lending | 2027 Outlook
How private lenders are using AI, and where they still want a person in charge.
Presented by Mortgage Automator. Survey by Lux Insights.
Adoption is no longer the open question. 84% of private lenders now use AI somewhere in how they originate or service loans, and 37% describe that use as wide or extensive.
Little of this is experimental. 89% expect to use more AI in 2027, and 38% expect to use significantly more. Lenders are not running pilots. They are expanding something already in production.
The pattern holds across the border. US and Canadian lenders report adoption at effectively the same level, which is worth noting in a sector where regulatory environments, loan structures, and capital sources diverge as much as they do.
So the useful question is not whether private lenders use AI. It is where they are willing to let it work on its own, and where they still want a person signing off before anything happens. That line is what the rest of this report is about.
The same lenders who trust AI to read documents are more careful about letting it inform credit decisions.
The spread from the top of that list to the bottom is seventeen points, and nothing on it falls below 64%. That makes this a gradient rather than a verdict. Lenders are not withholding trust from AI so much as rationing it against what a mistake would cost.
The tasks at the top share a quality: when the AI is wrong, somebody notices, and the fix is cheap. A misread document gets corrected. A reporting figure gets reconciled against the ledger. Toward the bottom of the list, errors are quieter and more expensive. A weak underwriting signal does not announce itself. It shows up in the portfolio eighteen months later.
Among the 41 lenders using AI for both tasks, 80% trust the document review output and 56% trust the underwriting output. Small base; supporting detail only.
37% of lenders have adopted AI widely or extensively. Set against the lenders who have barely started, they look less like early adopters of a technology and more like a particular kind of business.
| Heavy AI users | Barely started | |
|---|---|---|
| Enterprise lenders (over $500M) | 48% | 15% |
| More willing to take on risk than a year ago | 77% | 33% |
| Very optimistic about 2027 | 38% | 12% |
| Would recommend private lending to a peer | 78% | 48% |
| Expect a significant increase in AI use | 49% | 12% |
They are bigger, with 48% in the enterprise tier against 15%. They are more willing to take on risk than they were a year ago, more optimistic about 2027, and more likely to recommend private lending to someone else in the industry. They also intend to go further: 49% expect a significant increase in AI use, against 12% of the group that has barely begun.
What that adds up to is not that AI made these lenders ambitious. It is that ambitious, growing lenders have made AI part of how they run.
The more revealing finding sits underneath. Only 17% of all lenders run a largely integrated platform. Among heavy AI users it is 37%. That is better, but it still leaves most of them adding AI on top of systems that do not talk to each other. 27% of heavy users are running mainly on spreadsheets and email.
All lenders
Heavy AI users
AI is arriving in private lending faster than integration is.
55% of lenders saw an attempted fraud in the past year: 110 of the 200 surveyed.
Among lenders who faced attempted fraud, AI tools and human review were the two most common ways it was caught, and 4 in 10 used both together.
| How they caught the attempt | Lenders | Share |
|---|---|---|
| AI tools and a person reviewing files | 44 | 40% |
| AI tools, but no person reviewing files | 28 | 25% |
| A person reviewing files, but no AI tools | 25 | 23% |
| Neither (rules-based alerts, a third-party alert, or found after funding) | 13 | 12% |
Because lenders could name more than one method, these figures overlap rather than compete. On average each lender named about two, and only 17% relied on a single one. The most common answer was both at once.
That is the finding. Detection did not come down to better tooling or better judgment. It came down to both, working the same files.
Lenders expect the problem to get harder rather than easier. 56% are very or extremely concerned about AI-enabled fraud heading into 2027.
Two lenders put the problem in their own words:
The rise of AI generated document fraud and the challenge of updating the underwriting system fast enough without slowing down loan originations [worries me].
Fraud detection is getting harder as fake documents can look convincing even when basic checks appear clean during normal underwriting reviews.
The lenders describing that problem are not asking for less AI. They are describing why the review step is critical.
Delegate, Review, Decide.
The framework below sorts AI use cases by what is at stake when the AI is wrong, not by how much lenders currently trust it. Those two things usually move together. Where they come apart is where the thinking is.
Compliance monitoring is the most trusted use case in this survey, at 81%. It still sits in Review, because a compliance error carries regulatory consequences and somebody needs to have signed off on it. Trust and stakes are different measures, and this framework follows stakes.
Some use cases are also more than one task, so they appear on more than one level. Pulling data off a document is different from deciding what it means: whether a title exception is a problem, or whether the lender is actually named as loss payee (the callout above on Lendr’s AI Underwriter shows how much that varies by field). Servicing works the same way, since a payment reminder is not a default notice. And a borrower update is only routine if it commits the lender to nothing. A payoff figure, extension terms or a rate belongs in Review, however routine it looks.
% = lenders using AI for that use case who trust its output. Hover or tap for the part of the use case that sits at that level.
The question for 2027 isn’t whether to use AI. Most lenders already do. It is whether the AI they use can be trusted with more over time.
Four things will separate the lenders who get real value from it.
Pulling data off documents, sending payment reminders, building internal dashboards, drafting routine borrower updates. When AI gets these wrong, somebody notices and the fix is cheap. That is where the time comes back. Just be precise about what counts as routine. An update that commits the lender to a payoff figure, extension terms or a rate is not routine, and a person should review it.
Lenders told us clearly where they want a person in charge. Three calls should always end with a named person signing off: the credit decision (approve, decline, final terms), moving money (funding, draws, payoffs, and especially wire instructions, a favorite target for AI-generated fraud), and default and foreclosure. AI can prepare all of it. The goal is not to remove that judgment. It is to put better information in front of it sooner. Private lending is still a relationship business. Whether a borrower will pick up the phone when a project goes sideways is a judgment call. AI can help inform it, but it should not make it.
Only 17% of lenders run on one connected platform. AI added on top of spreadsheets, side tools and one-off builds is hard to check, hard to audit, and somebody has to keep it running. Trust grows when the AI’s suggestion and the person’s decision sit on the same record as the loan.
Override rate is a good start, but a team that rubber-stamps everything also has an override rate of zero. So track it per field, not just per task, and pair it with a second number: how often something the AI got wrong slips through and shows up later, at closing, in servicing, or in an investor report. Track how often your team overrides the AI on a task, and how often mistakes slip through anyway. Once both stay low, you can move it up a level. And keep spot-checking after you do. Delegate still means someone is sampling the work.
Fill in your details to read the rest of Trust Issues. It's free.
You already have access. Pick up where you left off.
Read Unlock the Full InsightFaster, busier, still stitched together. 87% of lenders saw more applications and 85% funded more loans, but only 17% run on a largely integrated platform. See where you stand compared to similar-sized lenders and whether you agree with where the market is headed.
You're on the list. We'll email you when it's out.
Free live webinarDecember 1, 2026, 2pm ET
A walkthrough of the survey findings and what they mean for lenders heading into 2027. Find out what surprised us and compare your business to our survey data in real time.
Survey of 200 private lenders in the United States and Canada (120 US, 80 Canada), conducted by Lux Insights between August 26 and September 20, 2026. Margin of error ±5.8% at 90% confidence. In order to qualify, respondents must be a private lender, be a senior operator or higher, be a decision maker for purchasing technology, reside in US or Canada, be 18+ years of age. Responses collected through a 15-minute online survey.
