Altyst vs building the underwriting model in Excel

Start with what the spreadsheet is good at

Anyone selling you software will skip this section. It is the most important one, because if you do not know what you are giving up you cannot tell whether the trade is worth it.

A spreadsheet has no schema. That single property is why real-estate finance still runs on it. Any structure a deal can have, a spreadsheet can hold: a participating mortgage, an earnout on a partial condemnation, a ground lease with a reset tied to an index nobody else uses, a promote that pays differently before and after a refinance. Software that owns your model has to have anticipated the structure. A blank grid never has to anticipate anything.

Your firm's template is also more than arithmetic. It is where the house convention lives: which line items get their own row, how you treat a tax reassessment on transfer, what your investment committee expects on page one. That is institutional memory, and no import wizard has ever moved it.

And the file outlives everything. An .xlsx opens in other spreadsheet programs, sits in a data room, and goes to a lender who expects a workbook and not a login.

Hold on to all of that. The case for anything else has to be made against it, not around it.

Where the spreadsheet actually costs you

Four places, and none of them are about the math being wrong.

The retyping. A deal package arrives as a PDF offering memorandum, an Excel rent roll on a broker's own layout, a scanned trailing twelve, and a debt quote in an email. Every figure that reaches your model gets there because a person read it off one document and typed it into another. That is where the transposition happens, and it happens on the numbers that matter, because the numbers that matter are the ones there are the most of.

The tie-out that never gets done. The rent roll says one thing about in-place income and the T-12 says something slightly different, and reconciling them by hand takes an afternoon nobody has on a deal that is probably a pass. So it does not get done, and the discrepancy that would have killed the deal in week one survives into the bid.

Where a number came from. Open a model six weeks after you built it and look at the insurance line. The cell holds a value. It does not hold the page of the loss run it came from, whether it was the broker's number or the quote you got, or how confident you were. You reconstruct it from memory, or you email the analyst, or you accept it.

The fifth reprice. The first sensitivity run is fine. It is the fifth one, at nine at night, after the seller's counter, that finds the hardcoded number in the middle of a formula chain and the link to the workbook version that got renamed.

What happens to the same package in Altyst

Documents in, an editable model out, and the arithmetic done by an engine rather than predicted by a language model.

• It reads the offering memorandum, rent roll, T-12, debt quote and lease schedule in PDF, Excel, CSV, Word, plain text, or a photograph of a page. Scanned pages go through OCR.
• Extracted values arrive as proposals, each shown with its source and a confidence, for you to accept or overwrite before it enters the model. The documents are also tied out against each other, and conflicts come to you to settle rather than being silently resolved.
• Every financial result is computed by a deterministic engine in exact decimal arithmetic. Not by a model that predicts a plausible number. The same inputs always produce the same result, which is the only property that makes a downstream audit possible at all.
• Every extracted figure traces back to the document it came from and the excerpt it was read from, with a page number when that excerpt can be placed on exactly one page, and you can click a number to see the formula underneath it.
• Change any assumption and returns, cash flow, debt, and the downside case recompute together.

The part people do not expect: you still get the workbook

Altyst exports an Excel workbook, and its core tabs carry live formulas rather than pasted values. EGI is computed as gross potential rent plus other income less vacancy and credit loss. NOI is EGI less operating expenses. Levered cash flow is unlevered cash flow less debt service. The IRR row is an actual IRR over the cash-flow range. Open it in Excel, change a cell, and the sheet recalculates the way a model you built would.

Every workbook carries Cover, Model, Cash Flow, Debt, Scenarios, Sensitivities, Provenance and Model Checks. A deal picks up more depending on what it needs, among them Sources and Uses, Reprice, Partnership, After-Tax, Rent Roll, Tenant Schedule, Risks and Diligence. The Provenance tab works at the level of the assumption rather than the cell: every driving assumption, its value, whether it was your figure or one the system proposed, what kind of source stands behind it, and how firm that source is, plus any notes on how the documents were read. Document-and-excerpt attribution for a specific extracted figure lives on screen, not on that sheet.

Two details about that export, because they are the kind of thing you should be asking any vendor:

1. Not every tab recalculates. Some are computed by the engine and written in as values. The Cover tab names which tabs carry live formulas and says plainly that the rest do not recalculate, so nobody discovers it by editing a cell and watching nothing move. That line is generated by scanning the finished file for formula cells rather than typed by hand, so it cannot say a tab is live when its numbers are pasted.
2. The formulas are checked, not assumed. A separate evaluator in our test suite re-reads the workbook's own formulas, computes them, and compares the result against the engine. A workbook that wrote a formula disagreeing with the model fails the test run.

Ask every other vendor on your list for a sample export before you buy, and open it. Whether the cells contain formulas or numbers is the fastest single test of how seriously a product takes the handoff back to you.

Side by side, on the things that decide it

What to test before you switch anything

Do not take the table above on faith, ours or anyone's. Run one real deal you have already underwritten by hand, and compare.

1. Upload the actual package, including the ugly scanned page. Extraction quality on a clean broker Excel file tells you nothing.
2. Check the extracted rent roll against your own tie-out. Look specifically at unit count, down and model units, and concessions.
3. Rebuild your debt: size on LTV, on minimum DSCR, and on debt yield, and see whether the constraint that binds is the one you expected.
4. Push one assumption you know is load-bearing, and confirm the downside case moved with it.
5. Export the workbook and open it in Excel. Change a cell on a live tab. Watch what recalculates.
6. Read the Provenance tab as if you were the person on the investment committee who did not build the model.

If step 6 does not answer more questions than your own spreadsheet would, none of the rest matters.

What it costs, and what is metered

Plans are Individual $12, Professional $24, Team $99 a month, with a per additional deal price on each. The unit is one deal creation. Uploading documents, running the underwriting, editing assumptions, recomputing, and exporting are never metered, so repricing a deal eleven times costs the same as repricing it once. Unused deal credits roll over. AI document reading and property research draw on an allowance carried by each deal, included or additional, so one document-heavy deal never leaves a later one short, and when a deal has used its allowance the model, the edits, and every export keep working. The full detail, including each plan's monthly ceiling on new deals, is on Pricing.

Microsoft and Excel are trademarks of Microsoft Corporation. Altyst is not affiliated with, endorsed by, or sponsored by Microsoft. Excel is named here only to describe how underwriting models are commonly built, and how the workbook Altyst exports is meant to be used.