1. Capture Borrower Intent Without Inventing Eligibility
Good early intake collects name and contact, purchase versus refinance, general property and location, approximate timing, preferred meeting, and a high-level loan goal. AI may summarize, classify, route to the loan officer or team, schedule, and create a CRM task. Do not promise approval, program eligibility, rate, monthly payment, closing date, cash-to-close, or savings. Those come from the approved application, rater, and lender systems.
AI can help
- ✓Summarize and classify borrower intent
- ✓Route to the right loan officer or team
- ✓Schedule a meeting and create a CRM task
Human / specialist keeps
- ●Determine eligibility and program fit
- ●Own the quote, rate, and approval decision
- ●Run the licensed and carrier or lender process
Warning
Do not promise approval, program eligibility, rate, monthly payment, closing date, cash-to-close, or savings. Those come from approved mortgage systems.
2. Help Explain the Application Process—Do Not Fill Facts by Guessing
AI can explain the approved general process: common documents, the upload process, stage meaning, who to contact, how to schedule, and what happens next. The borrower provides and validates the actual application facts. Do not infer income, assets, employment, occupancy, debts, citizenship or residency, property value, or credit characteristics. The LOS owns the submitted data.
AI can help
- ✓Explain the application process and common documents
- ✓Answer administrative process FAQs
- ✓Guide scheduling and next steps
Human / specialist keeps
- ●Collect and validate actual application facts
- ●Own the LOS submission and data integrity
- ●Handle eligibility and program questions
Note
Do not infer income, assets, employment, occupancy, debts, citizenship, property value, or credit characteristics. The borrower provides and validates them; the LOS owns the submitted data.
3. Use AI to Organize Documents and Questions
Documents may include pay stubs, W-2s, tax returns, bank and asset statements, ID, purchase agreement, insurance, business financials, and other underwriting documents. AI may classify, identify document type, summarize, flag a missing page or period, prepare a processor checklist, and draft a borrower request. Do not let general AI independently determine income eligibility, asset acceptability, fraud, or underwriting sufficiency. Sensitive financial documents require approved secure handling.
AI can help
- ✓Classify uploads and identify document type
- ✓Flag missing pages or periods
- ✓Draft processor checklists and borrower requests
Human / specialist keeps
- ●Determine income, asset, and underwriting sufficiency
- ●Enforce secure document handling and retention
- ●Own fraud and eligibility decisions
Warning
Do not let general AI independently determine income eligibility, asset acceptability, fraud, or underwriting sufficiency. Sensitive financial documents require approved secure handling.
4. Never Let a General AI Model Become the Pricing Engine
Rates and fees can depend on loan program, market, lock period, credit, LTV, property, occupancy, points, investor, loan amount, state, and pricing adjustments. Borrower-specific rate and price must come from an approved PPE, lender, investor, or loan-officer workflow. AI may explain general concepts such as fixed versus adjustable, points, and the rate-lock process from approved educational content. Never invent a live borrower rate.
AI can help
- ✓Explain general rate and pricing concepts from approved content
- ✓Prepare borrower questions about the rate-lock process
- ✓Draft educational explanations for review
Human / specialist keeps
- ●Produce borrower-specific rate and price from the approved PPE
- ●Own the lock, investor, and pricing decisions
- ●Verify every rate and fee statement
Warning
Never invent a live borrower rate. Borrower-specific rate and price must come from an approved PPE, lender, investor, or loan-officer workflow.
5. Give the Loan Officer a Better Borrower Brief
Inputs include the CRM, verified application status, approved LOS fields, document status, communications, calendar, and open tasks. The output is the borrower goal, current stage, documents missing, questions, open conditions, and next action. Do not turn model inference into qualification or approval likelihood. The brief prepares the loan officer; the LOS and underwriting systems own the loan truth.
AI can help
- ✓Assemble borrower context and open conditions
- ✓Surface missing documents and next actions
- ✓Draft follow-up tasks from verified state
Human / specialist keeps
- ●Own qualification and approval likelihood
- ●Verify LOS and underwriting status
- ●Approve the next action and communication
Note
Do not turn model inference into qualification or approval likelihood. The brief prepares the loan officer; the LOS and underwriting systems own the loan truth.
6. Automate Borrower Updates From Real Loan State
Good updates include application received, documents requested, processing, appraisal ordered where confirmed, conditions requested, and closing appointment confirmed. Rules trigger from verified LOS or workflow state, and AI converts that state into clear language. Do not communicate approval, clear-to-close, rate lock, appraisal completion, or condition satisfaction unless approved systems confirm it.
AI can help
- ✓Convert verified loan state into clear borrower language
- ✓Draft status updates from real workflow triggers
- ✓Prepare reminders and next-step messages
Human / specialist keeps
- ●Confirm approval, clear-to-close, and rate lock from approved systems
- ●Own appraisal and condition satisfaction status
- ●Approve every borrower-facing status statement
Warning
Do not communicate approval, clear-to-close, rate lock, appraisal completion, or condition satisfaction unless approved systems confirm it.
7. Treat AI/ML Underwriting as a Governed Lending System, Not a Chatbot Feature
AI and ML can be used in modern origination and servicing, and current Fannie Mae and Freddie Mac governance frameworks recognize this. Underwriting and credit systems need formal governance, approved data, applicable-law compliance, model and vendor oversight, security, documentation, monitoring, and integration with official decision systems. Do not recommend uploading a borrower file to a general chatbot and asking whether to approve the loan.
AI can help
- ✓Support governed origination and servicing workflows
- ✓Summarize conditions and documents for staff
- ✓Prepare underwriter context for review
Human / specialist keeps
- ●Own the underwriting and credit decision
- ●Apply formal governance and model oversight
- ●Integrate with official decision systems
Warning
Do not upload a borrower file to a general chatbot and ask whether to approve the loan. Underwriting is a governed lending system, not a chatbot feature.
8. Keep Action-Taken and Adverse-Action Notices Inside the Approved Compliance Process
Current Regulation B includes action-taken and adverse-action notification requirements, and current CFPB resources maintain model and sample notification forms. Do not let AI independently select a denial reason, rewrite a regulated reason, create an unsupported adverse-action explanation, change notice timing, or determine final action. The approved creditor system and process control the decision, reason, notice, timing, and record. Actual valuation belongs to approved appraisal or valuation systems and the creditor or investor process.
AI can help
- ✓Summarize valuation documents for staff
- ✓Prepare compliance context for review
- ✓Organize condition and document checklists
Human / specialist keeps
- ●Select denial reasons and produce adverse-action notices
- ●Own the decision, notice, timing, and record
- ●Control appraisals and valuations through approved systems
Warning
Do not let AI independently select a denial reason, rewrite a regulated reason, or determine final action. The approved creditor system and process control the decision, notice, and record.