What can AI for mortgage lenders do, and where do the rules draw lines?
By Yath Regmi, Founder, Kodec
Verified on Updated Published
Short answer
AI for mortgage lenders works best on paperwork and follow-up. Keep a person on anything that approves, denies, prices or promises a loan. Sorting documents and reminder calls carry the least risk. Freddie Mac now expects written AI governance from lenders, including for vendor tools, and a law-firm summary says Fannie Mae does too. Ask every vendor where borrower files go before you sign.
When this is the right call
Skip AI for now if your files are few and your processor has spare hours. The tools cost money and time to set up, and a small shop may save more by fixing a slow handoff. Also wait if you cannot say who owns your AI policy. Freddie Mac sellers have had to follow an AI governance section since March 3, 2026. If you sell to Fannie Mae, its letter took effect August 6, 2026, per a law-firm summary.
AI for mortgage lenders: hours per loan, before and after AI helps with the paperwork
Example
| Step | Minutes per loan today | Minutes per loan with AI plus review | Minutes saved |
|---|---|---|---|
| Sort and name borrower documents | 20 | 5 | 15 |
| Build and check the condition list | 40 | 25 | 15 |
| Chase the borrower for missing items | 45 | 25 | 20 |
| Answer status questions | 20 | 10 | 10 |
| Total | 125 | 65 | 60 |
Every input here is invented, so replace them with your own time study. The total is 1 hour saved per loan. At 40 loans a month that is 40 hours. At an Example loaded cost of $35 an hour, that is $1,400 a month. The review time stays in the right column because a person still checks the output. Compare that to a published tool price such as $99 a month for follow-up calls, or a quote you request for document tools.
Side by side
Ocrolus
| Job it does | Document classification and condition checks against Encompass data.12 |
|---|---|
| Published price | None published. Page says "Schedule demo".12 |
| Where your data goes | Vendor-hosted. Page names Encompass integration and says data capture accuracy is insured by Lloyd's of London.12 |
| Loan system fit | Native Encompass integration.12 |
| Rate quotes | Not its job. |
| Who sells this number | Ocrolus. |
Lead Friendly
| Job it does | AI voice calls for rate-quote follow-up, document chasing and past-borrower outreach.13 |
|---|---|
| Published price | $49, $99 or $199 a month, plus $3 a month for a phone number (page updated 2026-10-28).13 |
| Where your data goes | Vendor-hosted. Customer must scrub Do Not Call lists and hold consent records.13 |
| Loan system fit | No native Encompass, Calyx or Arive link. CSV import. Two-way sync is on its roadmap.13 |
| Rate quotes | The agent never quotes rates, payments, APRs or approvals.13 |
| Who sells this number | Lead Friendly. |
Microsoft 365 Copilot
| Job it does | General assistant inside Microsoft 365 for drafting, summaries and searching your own files.14 |
|---|---|
| Published price | $18 per user a month yearly, promo through 2026-12-31 for existing Microsoft 365 Business customers. $25.20 monthly billing.14 |
| Where your data goes | Stays in the Microsoft 365 boundary and is not used to train the models, per Microsoft.14 |
| Loan system fit | No loan system link. Works on documents and email you already hold.14 |
| Rate quotes | Can draft text. A person must supply and approve any rate. |
| Who sells this number | Microsoft. |
Salesforce Agentforce
| Job it does | Build your own agents, such as a customer-facing question bot, on Salesforce.15 |
|---|---|
| Published price | $2 per conversation, or $125 per user a month for add-ons. Page says prices can change.15 |
| Where your data goes | Not stated on the pricing page. Ask Salesforce.15 |
| Loan system fit | Needs Salesforce and custom build work.15 |
| Rate quotes | You decide what the agent may say. That choice is yours and your compliance team's. |
| Who sells this number | Salesforce. |
| What it is | Ocrolus | Lead Friendly | Microsoft 365 Copilot | Salesforce Agentforce |
|---|---|---|---|---|
| Job it does | Document classification and condition checks against Encompass data.12 | AI voice calls for rate-quote follow-up, document chasing and past-borrower outreach.13 | General assistant inside Microsoft 365 for drafting, summaries and searching your own files.14 | Build your own agents, such as a customer-facing question bot, on Salesforce.15 |
| Published price | None published. Page says "Schedule demo".12 | $49, $99 or $199 a month, plus $3 a month for a phone number (page updated 2026-10-28).13 | $18 per user a month yearly, promo through 2026-12-31 for existing Microsoft 365 Business customers. $25.20 monthly billing.14 | $2 per conversation, or $125 per user a month for add-ons. Page says prices can change.15 |
| Where your data goes | Vendor-hosted. Page names Encompass integration and says data capture accuracy is insured by Lloyd's of London.12 | Vendor-hosted. Customer must scrub Do Not Call lists and hold consent records.13 | Stays in the Microsoft 365 boundary and is not used to train the models, per Microsoft.14 | Not stated on the pricing page. Ask Salesforce.15 |
| Loan system fit | Native Encompass integration.12 | No native Encompass, Calyx or Arive link. CSV import. Two-way sync is on its roadmap.13 | No loan system link. Works on documents and email you already hold.14 | Needs Salesforce and custom build work.15 |
| Rate quotes | Not its job. | The agent never quotes rates, payments, APRs or approvals.13 | Can draft text. A person must supply and approve any rate. | You decide what the agent may say. That choice is yours and your compliance team's. |
| Who sells this number | Ocrolus. | Lead Friendly. | Microsoft. | Salesforce. |
Vendor by vendor
Lead Friendly
$49 Starter (150 voice minutes), $99 Pro (400 minutes), $199 Agency. Extra minutes $0.15. A phone number is $3 a month (page updated 2026-10-28, checked 2026-11-14).13
Good
Microsoft 365 Copilot
$18 per user a month on yearly billing, a promo that ends 2026-12-31 for existing Microsoft 365 Business customers. $25.20 on monthly billing (checked 2026-11-14).14
Good
Not so good
- It knows nothing about your loan system. It helps with drafts and summaries, not pipeline actions.
- The $18 price is a time-limited offer, so budget for the list price of $21.14
Salesforce Agentforce
$2 per conversation, or $125 per user a month for Sales and Service add-ons. Page says prices are subject to change (checked 2026-11-14).15
AI for mortgage companies: custom builds and your own private model from Kodec
Kodec is a technology and operations firm for owner-run companies in mortgage, real estate, home services and law. It builds internal tools, AI agents and automation, and in-house AI systems, meaning a self-hosted model or an internal assistant that runs on your own data. That matters when you do not want borrower files going to a vendor. Kodec connects the pieces to your loan system, and tracks hours and dollars against results. It fits companies of $1M to $20M that need a tool built around their own process. If you are a large lender, a firm like Accenture, Deloitte or IBM has more people for an enterprise rollout. If a tool in the table already does the job, buy it. Another option for custom work is Withn, an applied AI firm in California that lists a $1,875 two-week assessment (checked 2026-11-14).
What we found
Who wrote this number
AI for mortgage loan officers and processors: what is safe to automate
This is our sorting, based on the rules cited below. It is not legal advice, so show it to your compliance lead. Reg B says reasons for a denial must be specific. The CFPB's 2023 chatbot report warns that language-model bots can give wrong answers on fees and trap people in loops with no path to a human.156
| Group | Tasks |
|---|---|
| Safe to automate (a person spot-checks) | Naming and sorting borrower documents; building a missing-items list; reminder texts and calls to borrowers who opted in; summarizing a file for the processor; internal pipeline reports; drafting a loan officer's follow-up email for the officer to send. |
| Needs a human before it happens | Comparing a document to the loan system and clearing a condition; reading income from tax forms; answering a rate or fee question; sending the borrower anything that states a term; scrubbing call lists and checking consent; anything touching a complaint or dispute. |
| Do not automate | The decision to approve, deny or counteroffer; the written reasons for a denial; promising a rate or approval; any tool that receives borrower files before you have vendor terms and a security review. |
Who wrote this number
The CFPB circulars everyone cites were withdrawn in 2025
Most AI-and-lending pages still cite CFPB Circulars 2022-03 and 2023-03. The CFPB's own Withdrawn Guidance page lists both as withdrawn on May 12, 2025. Circular 2022-03, issued May 26, 2022, said a creditor cannot use a complex model if that stops it from giving specific reasons for a denial. Circular 2023-03 was issued September 19, 2023, and its CFPB page is marked archived. What still stands is the Reg B text: 12 CFR 1002.9(b)(2) says the statement of reasons "must be specific and indicate the principal reason(s) for the adverse action," and the notice is due within 30 days of a completed application. The withdrawn circulars are not a safe harbor, and the rule text did not change. We could not open the Federal Register notice, so the withdrawal date rests on the CFPB page.4231
Verified terms
Most prices are not published, and blog prices disagree
Of the four tools in the table, only three show a price, and the document tool, Ocrolus, shows none. For Capacity, another document and condition tool, third-party blogs gave $49 per user, $1,200 to $3,000 a month, and "custom". We did not use any of them. Published prices we did confirm on 2026-11-14 run from $49 a month for follow-up calls to $18 per user for a general assistant. A lender should ask for a written quote that names the loans per month, the users and the setup fee before comparing.1213141516
Who wrote this number
What a private or in-house model changes for borrower data
Borrower files hold names, income and account numbers. The FTC Safeguards Rule names mortgage lenders and brokers as covered. It requires encryption of customer information, multi-factor sign-in for anyone using an information system, and that you choose capable service providers, require safeguards by contract, and check them over time (16 CFR 314.4(c)(3), (c)(5), (f)). Firms with customer information on fewer than 5,000 consumers are exempt from some parts, but not encryption or multi-factor sign-in. Freddie Mac's Bulletin 2025-16 (effective March 3, 2026) added an AI governance section to its Guide. Fannie Mae's Lender Letter LL-2026-04 (issued April 8, 2026, effective August 6, 2026) asks for written AI policies, holds vendor AI to the same standard, and says to describe your AI use on request. We read the Freddie text. We read Fannie through a law-firm summary because its page blocked us. A private model does not remove these duties. It gives you one fewer outside party to list, review and contract with, since the files stay on systems you control. Whether that is worth the build depends on how many files you handle.678910
How we checked
We searched the question and five variants on 2026-11-14 and read vendor pages, CFPB pages, the FTC rule text, the Freddie Mac bulletin PDF and an industry survey summary. Search results were mostly vendor-run lists, so we used them to find tools, not as facts. Prices come only from the vendor page and carry the date we checked. The Fannie Mae letter and the Federal Register withdrawal notice could not be opened directly. Those two points are marked in the text. We gave no legal advice and did not test any tool on real borrower files. The hours table is an Example.
Questions
Reg B says a denial notice must give specific principal reasons, and the CFPB page lists no exception for models. A tool can help your underwriter, but a person should own the decision and the reasons. Ask your compliance lead before any model touches a credit decision.
No. The CFPB lists Circulars 2022-03 and 2023-03 as withdrawn on May 12, 2025. The Reg B rule on specific reasons for denial is still in the regulation. The withdrawal is not a license to use opaque models. Check the CFPB site again before you rely on this.
Yes. Freddie Mac's Bulletin 2025-16 took effect March 3, 2026. Fannie Mae's Lender Letter LL-2026-04 took effect August 6, 2026, per a law-firm summary. Both ask for written AI governance and cover vendor tools. Read the letters for your own seller agreement, since we summarized them, not interpreted them.
It can, but it is the riskiest of the common uses. The CFPB's 2023 chatbot report warns of wrong answers on fees and no route to a human. Some vendors, like Lead Friendly, bar the bot from quoting rates. Route rate questions to a licensed loan officer.
It applies to the customer information you hold, wherever it goes. Mortgage lenders and brokers are covered. If a vendor tool receives borrower files, you must choose it with care, require safeguards by contract and check it over time under 16 CFR 314.4(f). It does not name AI.
Sources
- 1.CFPB, Regulation B 1002.9 (notification) accessed Nov 14, 2026
- 2.CFPB Circular 2022-03 (withdrawn 2025-05-12) accessed Nov 14, 2026
- 3.CFPB Circular 2023-03 (withdrawn 2025-05-12) accessed Nov 14, 2026
- 4.CFPB Withdrawn Guidance list accessed Nov 14, 2026
- 5.CFPB, Chatbots in consumer finance (2023-06-06) accessed Nov 14, 2026
- 6.16 CFR 314.4, FTC Safeguards Rule (Cornell LII) accessed Nov 14, 2026
- 7.16 CFR 314.6, Safeguards Rule exceptions (Cornell LII) accessed Nov 14, 2026
- 8.FTC, Safeguards Rule: what your business needs to know (Dec 2024) accessed Nov 14, 2026
- 9.Freddie Mac Guide Bulletin 2025-16 (2025-12-03) accessed Nov 14, 2026
- 10.Pease Bell summary of Fannie Mae LL-2026-04 (primary page blocked) accessed Nov 14, 2026
- 11.HousingWire on the MBA, AARMR and BCG AI survey (Sept 2026) accessed Nov 14, 2026
- 12.Ocrolus mortgage AI accessed Nov 14, 2026
- 13.Lead Friendly for mortgage accessed Nov 14, 2026
- 14.Microsoft 365 Copilot pricing accessed Nov 14, 2026
- 15.Salesforce Agentforce pricing accessed Nov 14, 2026
- 16.Withn pricing accessed Nov 14, 2026
Where to go next
Want to see which of your own calls turned into paying customers? See how Leverage works.
Related answers
- MortgageVendors
Bankrate vs Zillow mortgage leads: which is better for loan officers, and where does LendingTree fit?
Updated Nov 14, 2026
- All industriesAI & operations
What can AI agents for business operations actually do?
Updated Nov 14, 2026
- All industriesAI & operations
How to choose an AI implementation partner (questions to ask before you sign)
Updated Nov 14, 2026