Refinancing Into Higher Rates: Building the Operating Story Lenders Underwrite

$18.7 billion of hotel CMBS debt matures this year, most of it written when money was cheap. The replacement loan gets sized on trailing NOI that a lender believes. A reporting stack that produces a clean, system-generated data room every month (normalized NOI with evidence, comp-set benchmarking, a forecast track record and a rolling 13-week cash model) is how an owner moves that belief, and the proceeds with it.

Look at the building in the photo above. A lakeside grand hotel, blue awnings out, flag on the roof, a hundred-odd windows under a slate mansard. A guest sees the view. A buyer sees the rooms. A lender sees a trailing twelve months of cash flow and a roof that will need money someday, and it prices the loan on which of those it can prove.

That's the refinancing problem in one picture.

I owned ten hotels. I also founded Collective Retreats, grew it to ten locations and saw it acquired in 2025, and nothing teaches you about your own reporting like strangers reading it harder than you ever did. Whichever seat I was in, the lesson was the same. The hotel doesn't get financed. The file does.

This year a lot of files are going back to market at once, and they're going back into a rate world that didn't exist when the loans were written. What follows is about the one part of that an owner controls: the operating data package the lender underwrites, and the reporting stack that builds it every month instead of in a panic six weeks before maturity.

The Problem: Same Hotel, Different Loan

Trepp counts 596 hotel-backed CMBS loans with a combined $18.7 billion maturing in 2026. Nearly 70 percent of that balance is floating-rate debt. The FBT Gibbons team that published the count put the gap plainly: a loan written in 2016 or 2021 may carry 4 to 6 percent with a debt yield of 10 to 13 percent, and a new loan this year comes in around 6 to 7 percent.

Coupons are only half of it. The lender that writes the new loan also sets amortization, which most of the old interest-only CMBS paper didn't have. Access Point Financial, a hotel lender, says debt service on refinanced loans often runs about 40 percent higher than under the original terms.

So the sizing question changes. It used to be "what's the hotel worth." Now it's "how much loan does this NOI carry at today's coverage." HVS puts current requirements at a DSCR of 1.30x to 1.50x and loan-to-value of 55 to 65 percent for stabilized hotels, and says lenders "will underwrite the trailing 12 to 24 months of actual operating results, not projections or pro formas."

Read that last part twice. The number that sizes your loan is already written. It's in the last year of your P&L. The only open question is how much of it the lender accepts.

Here's what that looks like on an illustrative hotel. Same building, same NOI, two loans.

At those terms, every $100,000 of NOI the lender believes is worth about $840,000 of proceeds. That's the whole case for this article. The spread between the bottom two rows, about $2.8 million, isn't rate and it isn't the market. It's documentation.

The Data: Where the Pressure Sits

The lodging book isn't in crisis. Trepp's October update had the lodging CMBS delinquency rate down 24 basis points to 5.94 percent, and CRED iQ found hotels face the smallest refinancing rate gap of any property type, about 32 basis points, because many hotel loans already priced close to today's market. Averages hide the hotels that will struggle, though.

Limited service is where it shows. Trepp's lodging analysis found a 10.34 percent nonperforming rate in limited-service CMBS loans, and 19.9 percent of that balance below an 8 percent debt yield, the line Trepp uses as a common refinancing threshold. The median coverage in that segment fell from 1.81x at underwriting to 1.37x, with net cash flow 9.7 percent below underwriting.

Costs are the reason. AHLA's 2026 State of the Industry report put wages and benefits at nearly $128 billion in 2025 and projected nearly $131 billion for 2026, while PwC's May outlook called for RevPAR growth of 2.9 percent. Revenue that rises slower than labor doesn't reach NOI, and NOI is all the lender reads.

One more data point, and it's a softer one. Writing in Hotel Executive, Jacqueline Villamil and David Kasten of Strategic Solution Partners said lenders now have more applications on their desks than they can handle, and that "the first thing lenders evaluate is not just the numbers; it is whether those numbers make sense in context." A credit officer with a full desk reads the package that answers its own questions.

The Framework: What a Lender Actually Underwrites

Strip the term sheet down and a hotel lender is testing four things.

Is the trailing NOI real? That means the T-12 ties to the general ledger, the general ledger ties to bank deposits, and every add-back has a document behind it. HVS's advice to owners is to keep normalized NOI clean and well-documented, separating recurring costs from one-time items.

Is it holding up against the market? The STR or comp-set report says whether the hotel is gaining or losing share. A lender reads a falling index as falling NOI that hasn't shown up yet.

Does management know its own hotel? This is the forecast question, and owners underrate it. Lenders size on actuals, but they still ask for a budget and a three-to-five-year pro forma. A team that can show its past forecasts landed close gets its pro forma read as a plan. A team that can't gets it read as a sales document.

Can the hotel survive a bad quarter? Cash, reserves, the PIP schedule, seasonality, the franchise term. A rolling 13-week cash model answers most of it in one page.

Here's each request, and the system that should be producing it without anyone opening a spreadsheet.

The Mechanism: A Monthly, System-Generated Data Room

The fix is to stop treating the lender package as a project. Build it as a by-product of the monthly reporting the hotel already does, so on any given day the data room is at most a month stale. Five pieces.

1. One store, tied out. Connectors pull the PMS, POS, payroll, accounting GL and bank feeds into a single database keyed by date and USALI account. Reconciliation runs automatically: PMS room revenue against the GL, GL cash against bank deposits, payroll against labor lines. Any break above tolerance stops the monthly package until the controller clears it. This is the same backbone as the owner reporting pipeline in yesterday's piece, pointed at a different reader.

2. A normalized NOI engine with an evidence trail. This is where the money is. A classifier reads every GL line and every invoice above a threshold and flags candidates for normalization: the storm repair, the one-off legal bill, the pre-opening cost of a new outlet, the equipment replacement that should arguably have been capitalized. It doesn't decide anything. It builds a register, attaches the invoice, the approval and a one-line rationale to each item, and the controller accepts or rejects. When the lender asks why NOI was adjusted, the answer is a link, not a phone call. A haircut on unsupported add-backs is what costs the owner in the fourth row of the first table, and this register is what avoids it.

3. Automated benchmarking with written analysis. The STR or comp-set feed lands monthly. The system computes occupancy, ADR and RevPAR index by segment and by month, then a language model drafts two or three paragraphs explaining the movement from data already in the store: a group that washed, a new competitor, a renovation that took rooms out of service. Each sentence links to the figures behind it, the model says so when it can't find a driver, and a person approves before it goes into the room. Raw STR PDFs with no analysis are one of the things that slow lenders down; this takes the question off their desk.

4. A forecast archive and accuracy score. Every forecast and budget version gets saved with its date. Each month the system scores the 30-, 60- and 90-day forecasts against what happened (the method is in our forecast accuracy benchmarking piece). After twelve months the hotel has something most borrowers don't: evidence of how good its own predictions are.

5. A rolling 13-week cash model and a refinance simulator. On-the-books revenue from the PMS, scheduled AP, payroll runs, debt service and reserve deposits roll into a weekly cash view that refreshes every Monday. Beside it sits a simple sizing model: given trailing NOI, today's quoted rate, amortization and the lender's DSCR and debt yield tests, how much loan does the hotel carry and how big is the gap to payoff. Run it monthly from eighteen months out. If the gap is growing, the owner knows while there's still time to do something about NOI.

The data room itself is just a folder structure the system fills, with versioning and a Q&A log. Lender questions go into the log, an assistant drafts answers from the store, the controller or owner approves, and any question asked twice becomes a standing page in the room.

Forecast Accuracy as a Credit Signal

I want to spend a minute on the forecast archive, because it's the piece owners skip and the one that changes the conversation most.

A lender reading a pro forma is asking one question: do these people know what's coming? Usually there's no way to answer it, so the lender haircuts the projections and moves on. A hotel that can hand over two years of dated forecasts next to actuals removes the guess.

At Starwood I worked across more than 800 hotels as VP of sales strategy and operations, and the properties that got the benefit of the doubt in any review were the ones whose numbers had been right before. Credit works the same way.

Bias matters more than the headline error. A forecast that misses high every month tells the lender the budget is high too.

Implementation: Eighteen Months Out, Not Six Weeks

The timing is the whole game. Lenders read 12 to 24 months of actuals, so the NOI work and the documentation have to start before the window the lender will look at. Six weeks before maturity is too late to change anything but the formatting.

Eighteen months out, stand up the store and the reconciliation. Map the GL to USALI if it isn't already, connect the PMS, payroll and bank feeds, and run the tie-outs until they come back clean three months straight. Start the forecast archive the same week. It costs nothing and every month of history counts.

Fifteen months out, turn on the add-back register and go back through the trailing year. Most hotels find costs that were one-time and never labeled, and some that were expensed when the accounting standard allowed capitalizing them. HVS makes the point that major equipment replacements and damage repairs should be capitalized where standard accounting allows. Do that with your accountant, document it, and do it now, not in the package.

Twelve months out, the monthly data room should be assembling itself. Turn on the benchmarking drafter and the 13-week cash model, and start running the refinance simulator against current quotes. If the proceeds gap is bigger than the equity you're prepared to put in, that's a year to work on NOI, talk to the franchisor about the term, or line up mezzanine (HVS puts it at 12 to 14 percent) before you need it.

Six months out, the package goes to a broker or lenders as a link to a live, versioned room, not a zip file. Questions go into the log. Answers come from the store.

If you want to know which of your systems can feed a monthly data room today and which will need a workaround, that's the first thing our AI and technology scorecard and reporting engagement maps.

None of this changes the rate. It changes what the rate gets applied to, and at 1.40x and 7 percent, a hundred thousand dollars of NOI a lender believes is worth most of a million dollars of loan.

Frequently Asked Questions