Drop the PDF
The model reads the offering memorandum and pulls keys, T-12 revenue and expense lines, ADR, occupancy and comp RevPAR. Every field comes back with the page it was found on and a confidence score.
Upload the offering memorandum. Agents read it, a deterministic engine sizes the loan and builds the USALI proforma, and you get a live Excel model and a lender or investor package. The run persists, so the broker package, the credit memo and the variance report are views of it rather than copies.
The actual workspace, not a mockup. Courtyard Example, Lake Nona, 120 keys, Orlando MSA.
Three product lines
One record carries the deal from the first read of the OM through the hold, so nobody rebuilds the model at the handoff.
Underwriting
LiveAgents pull the T-12, the STR report and the rent roll, and every figure sits beside the page it came from until a person approves it. Operations, debt, capex, stress and returns all seed from that one read.
Asset Management
In developmentAsset management usually starts over at closing, in a new spreadsheet that never reconciles with the underwrite. Here the acquisition model stays the plan, and monthly actuals post against it on the same record.
Market Intelligence
Market data livePick the hotel out of the STR inventory and the market pull runs with the extraction: the submarket, the competitive set, seasonality and the federal economic panel, attached to the deal rather than pasted into it. Submarket performance forecasts are still being built.
Selling hotels? Broker listing pages run on the same record: underwrite the listing once, send one link, and see who opened it. Listing is free.
For brokers →Why bother
Most of it is typing numbers out of a PDF and checking they were typed correctly. Then the lender does it again, the committee does it again, and asset management starts over at closing: six rebuilds of the same asset, none of which reconcile.
Today
With PATL
You still decide whether the deal is any good. You just get to that question a lot sooner.
How it works
The model reads the offering memorandum and pulls keys, T-12 revenue and expense lines, ADR, occupancy and comp RevPAR. Every field comes back with the page it was found on and a confidence score.
Nothing reaches the model until you approve it. Each value sits beside its source page, so you can check a number against the document itself.
The engine builds RevPAR, runs the USALI departmental P&L, sizes debt against all three tests, and carries a ten-year cash flow through to IRR and equity multiple. Then you export a live Excel model and a branded PDF package.
Starting a deal
Pick the hotel out of the STR inventory or drop the offering memorandum in. The extraction, the staffing model and the market pull all run at once, and you can close the tab while they do. The deal is waiting when they finish.
Language models are good at finding a number on page 14 and unreliable at arithmetic that has to tie out. So a model does the reading, and everything downstream is computed by a deterministic engine in exact decimal, with no floating point anywhere in it. Run the same deal twice and you get the same numbers, which matters because a proforma that moves between runs is no use to a credit committee.
How PATL reads a deal
What happens, step by step, when you upload a confidential offering memorandum to PATL.
01 · Upload
Offering memorandum, T-12 and STAR report. The files a seller asked you to keep confidential.
02 · Read in private
An AI model trained on hotel documents reads every page on PATL’s own servers. No outside AI company sees your file.
On the roadmap03 · Calculate
Builds the proforma, sizes the loan and runs ten years of cash flow. Same inputs, same answer, every time.
Live today04 · Share
Every number links to the page it came from. Lenders and investment committees see the same file.
Live todayAll of this can run inside your own cloud account, under your own encryption keys.
On the roadmapSees only a tricky page, with names, addresses and deal terms blacked out first.
On the roadmapIn plain English: like a firm’s back office
Your own staff reads the file behind a locked door.
A specialist sees one blacked-out page, never the whole deal.
The accountant runs every number by the book and keeps the receipts.
TodayThe reading step runs on Anthropic’s Claude, under commercial terms that do not allow training on your files, and nothing it reads counts until a person on your team approves it. The private reader replaces that step. Anthropic is listed in our privacy policy.
PATLCalculator: live today · Private reader: on the roadmap
The engine
Every module is hotel-specific, and each one is tested on its own.
Revenue
Occupancy ramp with month-level disruption windows, ADR growth, comp penetration index, and a seasonality curve you can pull straight from market data.
Channels
Each channel carries a share and a commission. The blended ADR drag falls out of the mix, and direct is the implied remainder, so the cost of distribution stays visible.
Departments
Rooms and other operated departments with their own direct expense, then A&G, sales and marketing, utilities, POM and IT to GOP, then fixed charges to EBITDA and NOI.
Costs
Every expense category runs as a single base rate or as a named build-up (percent, per occupied room, per available room or fixed) with its own escalation.
Brand
Royalty, brand marketing, reservation, loyalty and technology charges, each on its own basis. The output is the all-in brand load per year, which a single 5% royalty line hides.
Labor
Positions, counts and wages with a burden engine over them, allocated across departments, so a staffing change moves the P&L the way it moves the payroll.
Capital
The property improvement plan lands on the months you say it lands on, flows into uses and equity, and ongoing capex lands in the cash flow.
Debt
The loan is the minimum of the LTV, DSCR and debt-yield constraints, and the model reports which one binds. Interest-only and amortizing schedules with per-year coverage.
Returns
Exit at cap through selling costs and loan payoff to net equity, levered and unlevered IRR, equity multiple and cash-on-cash, plus two-axis grids that re-size the loan in every cell.
Costs
Each USALI category runs as one base rate or opens into named line items, each on its own basis with its own escalation. Open a category and the total is still the same number the P&L carries.
Costs · the departmental build-up
Debt
LTV, DSCR and debt yield each produce a number, the model takes the minimum, and it says which one bound. When one loan is not the structure, the stack takes tranches senior to junior, each sizing against the cumulative position above it, which is how subordinate paper is actually quoted.
Debt · the sizing constraint that binds
Stress
Each scenario re-runs the whole model with one thing changed, then gets a verdict: it fails when the equity does not return or the loan cannot cover itself, and warns when coverage slips under the covenant. Write your own with the same vocabulary, and it is scored and exported the same way.
Stress · scenarios against the base case
What you get
Both exports are versioned, attached to the deal, and stamped with who generated them.
See one first
A 120-key select-service hotel: operating statement, KPIs by year, revenue and channel mix, staffing, capital plan, exit and debt, a stress grid that re-sizes the loan in every cell, and the monthly appendix. The figures are illustrative, and there is no email form in front of it.
Every export is versioned per deal, so "which model did we send them in March" has an answer.
Model Context Protocol
PATL ships a hosted MCP server. Point Claude at it and ask about your pipeline in plain language. It reads your deals, runs the engine and cites the figures, under your permissions and without a copy-paste.
The assistant decides what to ask. The engine decides what is true. No figure it reports is generated by a language model. The model calls a tool, and the arithmetic is the same deterministic engine that produces your lender package.
Who it is for
Brokers and credit teams
Owners and operators
Funds and sponsors
Pricing
Listing is free for brokers. Every plan and add-on is on the pricing page.
Trust
It arrives under an NDA and it stays that way, with the isolation written into the database schema.
Every table is scoped to your organization and enforced by row-level security in Postgres, forced on at the database rather than by a WHERE clause somebody has to remember to write.
Uploaded OMs are cached per organization, so an identical document uploaded by another firm never touches your extraction, and yours never touches theirs.
The assumption set, the engine version and the timestamp are frozen on every run. You can always show which model produced which package, and when.
Every extracted field carries a data-rights class, so third-party licensed figures stay usable inside the app and stay out of your exports.
Questions
No. A model reads the PDF and proposes values, each one shown next to the page it came from. You approve or correct every one before it reaches the deal. From there the arithmetic is a deterministic engine, and the same inputs always produce the same outputs, to the cent.
Today the pages are read by Anthropic’s Claude, under commercial terms that do not allow training on your files, and the document is never shared with another customer. We are building a private reader that runs on PATL’s own servers, so no outside AI company sees the file and only redacted pages ever leave. How PATL reads a deal.
You catch it at the review gate, which is exactly why the gate exists. Fields come back with a confidence score and a source page, so the ones worth double-checking announce themselves. Nothing is promoted into the model until you approve it.
Yes, and it is a working model. The P&L cascade ships as live formulas pointing at the assumption cells, so a lender can change a rate in the workbook and watch it recalculate.
There are three ways in: upload an OM, search 261,000 US lodging records and seed the deal from the property record, or type the assumptions yourself.
Yes. Download the sample investor packet: twenty pages generated by the engine from a sample 120-key select-service deal, with the figures marked illustrative. There is no email form in front of it.
Not yet. Underwriting is live today, and market data is live on every deal. Asset management, where monthly actuals post against the plan a hotel was bought on, is in development. Ask about early access.
No, and every package says so on its face. It is an underwriting model that shows its work: assumptions, sources and engine version.
Hotels first, because operating assets are where a lease-based model breaks down and hotels are the hardest version of that problem. Short-term rental investors have their own plan on the pricing page.
Get started
The scanned one, with the T-12 as an image and the expense detail in a footnote, is the one worth testing this on.