What is AI multifamily refinance underwriting? AI multifamily refinance underwriting is the use of large language models and document-aware AI tools to size take-out loans by DSCR constraint, evaluate supplemental loan eligibility, and compare agency, life company, and CMBS refinance options against a property's current trailing operating performance. The refinance underwriting use case differs meaningfully from acquisition underwriting because the property has actual operating history (not a broker pro forma), the loan-sizing math is constrained by lender stress tests as well as DSCR, and the borrower's goal is usually cash-out or rate-and-term optimization rather than purchase. For pillar context, see our AI multifamily underwriting guide; for a deeper dive on debt analysis tools, see ChatGPT vs Claude vs Gemini debt analysis comparison.
Key Takeaways
- Refinance underwriting starts from the property's actual T12 operating performance, not a broker pro forma; AI should rebuild NOI line-by-line from actuals before sizing the loan.
- DSCR is the binding constraint on most multifamily refinances; Fannie Mae and Freddie Mac typically require 1.25 to 1.30 DSCR, CMBS requires 1.20 to 1.25 DSCR, life companies require 1.30 to 1.40 DSCR.
- Supplemental loans (a second mortgage on top of the existing senior agency loan) can unlock additional cash-out without refinancing the senior; Freddie and Fannie supplementals require the combined LTV and DSCR to satisfy current underwriting standards.
- AI handles the comparative analysis well: same property, same NOI, three different lender stress tests, three different proceeds outputs.
- Common AI errors include using pro forma NOI instead of T12 NOI, missing the lender's debt yield constraint, and forgetting that supplementals are sized on top of an existing senior loan that is already drawn against the property's value.
Why Refinance Underwriting Is Different from Acquisition Underwriting
Acquisition underwriting is forward-looking: the buyer projects what the property will do under new ownership, applies a business plan, and underwrites to an exit. Refinance underwriting is backward-looking and operationally focused: the existing owner has a trailing 12 months (sometimes 24 months) of actual operating data, and the new loan is sized to a stressed version of that data. The lender doesn't care what the borrower thinks rents will be next year; the lender cares what rents actually were last year and how much that supports today.
This changes the AI workflow in three ways. First, the input to the underwriting is the T12 or T24 actuals, not a marketing pro forma. Second, the lender applies its own underwriting overlays (haircuts to vacancy, normalized management fee, normalized property taxes at current millage), which AI should replicate rather than the borrower's own assumptions. Third, the output is loan proceeds, not equity returns; the question is how much money the property can support at what rate and term, not what the unlevered IRR looks like.
Sizing the Senior Refinance Loan with AI
The senior refinance loan is the new first mortgage. AI should size it to the binding constraint among three lender tests.
- LTV constraint: Loan amount divided by current appraised value. Agency loans typically allow up to 75 to 80 percent LTV for stabilized assets, life companies 60 to 70 percent, CMBS up to 75 percent.
- DSCR constraint: NOI divided by annual debt service must equal or exceed the lender's minimum (1.25 to 1.40 depending on lender).
- Debt yield constraint: NOI divided by loan amount must equal or exceed the lender's minimum debt yield (typically 7.5 to 9.5 percent for multifamily). This protects the lender even if interest rates rise during the hold.
AI should iterate through all three constraints and report which one is binding. A typical small refinance might be DSCR-constrained at 6.5 percent interest rates (with current rates above the historic average), then shift to LTV-constrained when rates fall. The binding constraint determines how much equity the borrower can pull out. The right AI prompt: "Size the loan to the lesser of 75 percent LTV, 1.30 DSCR, or 8.0 percent debt yield using a 6.50 percent rate, 30-year amortization, and the T12 NOI of $[X]. Report the binding constraint and the resulting proceeds, LTV, DSCR, and debt yield."
How Lenders Adjust the T12 NOI
The NOI a lender uses to size a refinance is rarely the unadjusted T12 NOI. Lenders apply standardized underwriting overlays.
- Vacancy floor: Even if the property has a 2 percent actual vacancy, the lender will underwrite to a 5 to 7 percent vacancy minimum. AI should haircut income accordingly.
- Management fee normalization: If the property is self-managed at zero cost, the lender will impute a 3 to 5 percent management fee. If a related-party management fee is below market, the lender will normalize it up.
- Property tax reassessment: If taxes haven't been reassessed recently, the lender may project the next reassessment.
- Replacement reserve: Lenders require $250 to $300 per unit per year as an underwriting reserve regardless of what the owner actually spends.
- Repair and maintenance smoothing: If the T12 shows abnormally low R&M (because the owner deferred maintenance), the lender will normalize to a market level.
AI should bake all five overlays into the underwriting before sizing. The gap between the borrower's pro forma NOI and the lender's underwritten NOI is commonly 8 to 15 percent. Borrowers who don't account for this overestimate their refinance proceeds.
Supplemental Loan Analysis
A supplemental loan is a second mortgage layered on top of an existing senior agency loan, allowing the borrower to access additional proceeds without paying off the senior. Both Fannie Mae and Freddie Mac offer supplemental programs with specific rules.
- Eligibility: The existing senior loan must typically be at least 12 months old (sometimes 24 months for Freddie). The property must have demonstrated stabilized performance since closing.
- Combined LTV: Senior plus supplemental cannot exceed the program's maximum LTV (typically 75 to 80 percent of current value, not original value).
- Combined DSCR: Combined debt service from both loans must satisfy the program's minimum DSCR (typically 1.25 to 1.30).
- Subordination: The supplemental sits behind the senior in any default scenario, so the pricing reflects this; supplementals typically price 100 to 200 basis points wider than the senior.
AI is particularly useful for supplemental analysis because the math is multi-step: compute current combined LTV at current appraised value, compute the maximum combined LTV the program allows, the difference is the supplemental capacity, then test whether that supplemental amount also passes the combined DSCR test. The right prompt: "The senior loan has an outstanding balance of $[X] with annual debt service of $[Y]. Current appraised value is $[Z]. T12 NOI is $[N]. Compute the maximum supplemental loan amount under [Fannie Mae/Freddie Mac] supplemental program rules assuming a 75 percent maximum combined LTV and a 1.30 minimum combined DSCR at a [rate] interest rate and 30-year amortization."
Comparing Agency vs Life Company vs CMBS Refinance Paths
For most stabilized multifamily refinances, the borrower has three main lender channels. AI handles the side-by-side comparison faster than any spreadsheet.
- Agency (Fannie Mae / Freddie Mac): 5 to 10 year terms (sometimes longer), 30-year amortization, full or partial interest-only options, prepayment via defeasance or yield maintenance, recourse-free for most loan sizes, rates typically the tightest spread to Treasuries.
- Life companies: 7 to 25 year terms, lower LTV (60 to 70 percent), often more conservative DSCR (1.30 to 1.40), generally non-recourse, very competitive rates for trophy assets but tight underwriting standards.
- CMBS: 10-year terms, full term interest-only common, defeasance prepayment, non-recourse with standard carve-outs, looser underwriting overlays than agency but more onerous loan documents.
The borrower's decision tree depends on hold period, prepayment flexibility, and how aggressively they need to maximize proceeds. AI's job is to run the same property through all three lender models and produce a clean comparison: proceeds, rate, term, prepayment, recourse, monthly debt service, and cash-on-cash return on remaining equity for each option.
For more sophisticated debt structuring across lender types, see our deep dive on AI for CMBS loan underwriting and industry research from Cushman & Wakefield on multifamily capital markets.
The Cash-Out Refinance Workflow
The complete AI-driven cash-out refinance workflow has four steps. Step one: load the T12 (and T24 if available) actuals and the current loan amortization schedule into the AI session. Step two: have AI rebuild the underwriting NOI applying standard lender overlays. Step three: size the senior refinance to the binding constraint and compute cash-out proceeds (new loan minus existing loan payoff minus closing costs). Step four: separately, size a supplemental loan if the borrower wants to keep the existing senior in place.
For an owner-operator running this on multiple assets across a portfolio, the AI compresses what used to be a 2-week analysis cycle into a single afternoon. CRE investors looking for hands-on AI implementation support for portfolio-level refinance optimization can connect with The AI Consulting Network for tailored workflows.
Common AI Errors in Refinance Underwriting
Five mistakes recur often enough to be worth naming explicitly.
- Using pro forma NOI instead of T12 NOI. Refinance lenders underwrite to history, not projections. Forward growth gets little weight in the loan sizing.
- Missing the debt yield constraint. AI often skips debt yield because it's a less familiar metric. In rising-rate environments, debt yield is the binding constraint more often than DSCR.
- Forgetting lender overlays. The borrower's NOI and the lender's underwritten NOI differ. AI should apply the lender's overlays explicitly.
- Mis-modeling supplementals. Supplementals sit on top of the existing senior, not in place of it. The combined LTV and combined DSCR are what matter, and combined DSCR uses combined debt service from both loans.
- Ignoring prepayment penalties. If the existing senior loan has 3 years left of yield maintenance protection, refinancing now may cost more in prepayment penalty than the rate improvement saves. AI should always compute the net present value of the refinance including prepayment costs.
Frequently Asked Questions
Q: When should I refinance vs take a supplemental loan?
A: If the existing senior loan rate is meaningfully below current market rates and has prepayment protection remaining, supplementals are usually the right answer. If the senior loan is at or above current market rates, a full refinance can be more efficient. AI can compute both options side-by-side in minutes.
Q: How does AI know what underwriting overlays each lender uses?
A: Most agency lender overlays are documented in publicly available program guides. AI prompts should reference the specific program (Fannie Mae DUS, Freddie Mac Optigo SBL, Freddie Mac TAH, etc.) and apply the program-specific rules.
Q: Can AI handle the comparison between fixed-rate and floating-rate refinance options?
A: Yes. AI can model the floating-rate path with an assumed forward curve and compare lifetime debt service to a fixed-rate alternative. The output is the breakeven SOFR path at which the two options produce equivalent total cost.
Q: Is AI accurate enough to use for actual loan submission?
A: AI's underwriting output is a useful starting point but should be validated by the borrower's mortgage broker or in-house capital markets team before submission. AI handles the mechanics; humans handle the relationships and the credit narrative.
Q: How does AI handle complex situations like partial prepayments or rate buydowns?
A: These require explicit prompting. AI won't volunteer to model a partial prepayment scenario unless the user asks for it. The clearer the prompt, the cleaner the output.