Most underwriting models a prospect ever sees are PDFs. You can look at the number, but you cannot touch the formula behind it, and you certainly cannot watch the model fail and then watch someone fix it. This is the second thing: a real deliverable built inside Titleman for a real property, Legend Oaks, a 416-unit apartment community in Tampa, FL.
What was produced
The result is an Excel workbook, six tabs, each doing one job:
- Assumptions — every input in one place; yellow cells are assumptions you can change, green cells are sourced facts.
- Rent Roll — unit-level mix and rents, with a live check that the mix sums to the 416 units of record.
- Cash Flow — five years, phasing in loss-to-lease over the first three years rather than assuming it closes on day one.
- Debt Sizing — runs both an LTV test and a DSCR test and takes the loan the two agree on, stating in plain language which one bound.
- Exit & Sensitivity — exit value, IRR, equity multiple, a sensitivity grid, and four automated reality checks.
- Sources — every number in the workbook classified SOURCED, ASSUMED, or COMPUTED. Nothing else is allowed on that tab.
It was recalculated independently — once directly from the file with a separate formula engine, once from a second, separately written implementation of the same model — and every output matched to the cent: Year-1 NOI of $3,681,558, Year-5 NOI of $4,200,555, an operating expense ratio of 48.09%, a loan sized by DSCR (not the higher LTV figure, meaning cash flow — not collateral value — was the binding constraint) at $39,862,003 against $24,165,095 of required equity, a $38,368,632 loan balance at exit, $30,240,437 of net sale proceeds, a 1.47x equity multiple, and a levered IRR the two implementations put at 8.66% and 8.7% respectively — a rounding gap, not a disagreement.
The basis error, and how it was caught
The first draft of this workbook did not return 8.66%. It returned a 53% levered IRR, a 6.36x equity multiple, and a 2.98x debt service coverage ratio. Those numbers were the error announcing itself: nothing that good is free.
The cause was simple and easy to make. The draft had defaulted the purchase price to Legend Oaks' county-assessed value of $35.21M — already sourced, already in the spreadsheet, and convenient for exactly that reason. A Florida assessed value is an administrative number used to calculate property tax. It is not a price anyone would pay, and it has no necessary relationship to what the building's income can support. Feeding it into the price cell put the income side of the model and the price side of the model on two different bases — and the arithmetic connecting them, otherwise correct, reported the gap between those bases as if it were investment return.
The fix has three parts, all visible in the shipped file rather than described in a cover note:
- The price is now derived from the building's own Year-1 income at a stated going-in capitalization rate, with the assessed value demoted to a reference line the Sources tab says explicitly must never be used as a price.
- A peer-median comparison was removed from the exit tab. The comparison set's peer median is a median of assessed values; the exit price is a market price. Comparing them would have been the identical basis mismatch wearing a different outfit.
- Four reality checks were added to the Exit & Sensitivity tab, each of which fired on the first draft: a going-in yield outside roughly 3.5%–9%, an exit cap priced tighter than the going-in cap, a debt service coverage ratio above 2x, and an operating expense ratio under 40% — implausibly low for Florida product once insurance is included.
A model that is wrong in the flattering direction does not look wrong — it looks like a good deal. That is why the checks are built into the file, not written in a cover note.
Why this matters to someone reading it with money on the line
For a buyer or a lender, the interesting fact is not that a mistake happened — mistakes happen in every underwriting shop. It is that the mistake is now a permanent, automated check inside the file itself, not a note that depends on the next analyst remembering it. The Debt Sizing tab shows its work: it runs both the LTV and DSCR tests every time and states which one binds, which matters here specifically, because DSCR binding means the achievable loan is smaller — and the required equity larger — than a headline 65% LTV would suggest on its own.
The Sources tab draws a hard line worth taking seriously: address, unit count, assessed value, and owner of record are sourced from county assessment records. Everything else — every rent, the vacancy rate, every operating expense line, growth rates, and both cap rates — is an assumption, not an observation, and the workbook says so on the Rent Roll tab's own heading rather than in a footnote. That is the correct way to read this deliverable: as a rigorously checked structure for underwriting Legend Oaks, not yet as underwriting of Legend Oaks. Overwrite the yellow cells with the real numbers, and the same checks that caught a 53% IRR the first time will catch whatever the real numbers get wrong the second time.