A T-12 (trailing twelve months) is a property's income and expense statement for its last 12 months, usually one column per month. In commercial real estate, buyers and lenders read it to see what a property actually earned and spent, as opposed to what an offering memorandum says it could earn. The net operating income (NOI) at the bottom drives the price and the loan size, so an error in the file moves both.
At a fixed cap rate, value is NOI divided by that rate, so a 5% error in NOI is a 5% error in price. And the file was not built for you to parse. It comes out of the seller's accounting system as an Excel export or, worse, a PDF.
Here are seven checks to run before the NOI goes into a model. The first four are in code at the end.
The shape of the data
One row per line item, 12 month columns, and a Total column. Section header rows (INCOME, OPERATING EXPENSES) carry no numbers. Subtotal rows (Total income, Total operating expenses, Net operating income) sit among the line items.
Things that break a parser: negatives in parentheses, a dash for zero, dollar signs and commas stored as text, merged header cells, and a PDF that has to become a spreadsheet first. Normalize the numbers, tag every row with its section, and mark the subtotals so you never add them in as if they were lines.
Check 1: do the rows tie to the Total column?
Add the 12 months of each line and compare the result with the Total column. Use a tolerance of a dollar for rounding.
A mismatch means a month was pasted over, a number was edited by hand, or the Total column came from a different export than the months. Any of those means you cannot trust the file until someone explains it.
Check 2: do the subtotals tie?
Three tests: total income against the income lines above it, total operating expenses against the expense lines, and stated NOI against income minus operating expenses.
A summary row that disagrees with its own detail usually means a line was added to one and not the other.
Check 3: one-month spikes
Compare each month of an expense line with that line's median month. An annual insurance premium booked in a single month, a roof repair, or a tax bill paid on a cash basis all show up as a month several times the median.
A spike is a question, not an error. It can be a one-off you should take out of the run rate, a recurring cost that happens to be lumpy, or timing. Flag it, then ask which. The same check works in reverse: a line that is blank for a month may be a bill that was never booked.
Check 4: missing lines
Look for taxes, insurance, property management and repairs. A statement with no management line may belong to an owner who runs the property personally, but a new owner will probably pay a fee. No taxes line, or a tax figure that predates the sale, matters because many places reassess after a sale. A missing line makes NOI higher than the property will deliver under new ownership.
Check 5: the trend
Set the T-12 beside two shorter windows: the last 3 months times 4 (T3 annualized) and the last month times 12 (T1 annualized). A gap between them means something changed. It might be new leases, burned-off concessions or a vacancy. Seasonality can also cause it, because utilities and turnover are not flat across the year, so read the three numbers together and ask.
Check 6: items that belong below the line
Debt service, capital expenditures and depreciation are not operating expenses. They come after NOI. If they are inside operating expenses, NOI is understated. The reverse also happens: a repair relabeled as a capital item lifts NOI. Ask for the capital ledger.
Replacement reserves are a judgment call. Some NOI figures deduct them and some do not, so two properties with the same cap rate can be on different bases. Write down which one you are using.
Check 7: sign and label hygiene
Vacancy and concessions should reduce income. When one arrives as a positive number, a naive sum adds it instead of subtracting it. Look for duplicated lines, and for one-time income, such as a lease termination fee or an insurance payout, sitting inside "Other income". Also ask whether the statement is cash or accrual. On a cash basis, one month can hold a year of a bill.
Checks 1 to 4 in pandas
Feed it a CSV with the line name first, then Jan to Dec, then Total. The header rows in the sheet (INCOME, OPERATING EXPENSES) set the section.
import sys import pandas as pd M = ["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"] def num(x): s = str(x).replace("$", "").replace(",", "").strip() if s in ("", "-"): return 0.0 return -float(s[1:-1]) if s.startswith("(") else float(s) def load(path): raw = pd.read_csv(path, dtype=str).fillna("") raw.columns = ["line", *M, "total"] rows, section = [], "" for _, r in raw.iterrows(): if not any(r[m] for m in M): # header row such as INCOME section = r["line"].strip().lower() else: rows.append({"line": r["line"].strip(), "sec": section, "total": num(r["total"]), **{m: num(r[m]) for m in M}}) df = pd.DataFrame(rows) df["sum"] = df[M].sum(axis=1) return df def checks(df, tol=1.0): out = [] sub = df.line.str.lower().str.match(r"(total|net operating|noi)") inc = df[df.sec.str.startswith("income") & ~sub] exp = df[df.sec.str.startswith("operating") & ~sub] for _, r in df[(df["sum"] - df.total).abs() > tol].iterrows(): # check 1 out.append(f"{r.line}: months add to {r['sum']:,.0f}, Total says {r.total:,.0f}") for pat, lines in [("total income", inc), ("total operating", exp)]: # check 2 s = df[df.line.str.lower().str.startswith(pat)] if len(s) and abs(s.iloc[0]["sum"] - lines["sum"].sum()) > tol: out.append(f"{pat}: lines add to {lines['sum'].sum():,.0f}, stated {s.iloc[0]['sum']:,.0f}") for _, r in exp.iterrows(): # check 3 med = r[M].astype(float).median() out += [f"{r.line}, {m}: {r[m]:,.0f} against a median month of {med:,.0f}" for m in M if med > 0 and r[m] > 2.5 * med] for label, pat in {"taxes": "tax", "insurance": "insur", "management": "manage", "repairs": "repair|maint"}.items(): # check 4 if not exp.line.str.contains(pat, case=False).any(): out.append(f"No {label} line in operating expenses: NOI may be too high") return out if __name__ == "__main__": print("\n".join(checks(load(sys.argv[1]))) or "No flags.") The 2.5 times the median threshold is my choice. Tune it to the property. A flag on a file with one bad month looks like this:
Other income: months add to 18,000, Total says 18,500 Repairs and maintenance, Aug: 9,500 against a median month of 1,800 No management line in operating expenses: NOI may be too high What to ask the seller
| Flag | Question |
|---|---|
| Row does not tie to Total | Which export is this, and can you send the original from the accounting system? |
| Spike in one month | What was it, and does it recur? Is the statement cash or accrual? |
| Missing management fee | Who manages the property, and what do they charge? |
| T3 or T1 far from T-12 | What changed in the last quarter, and is it permanent? |
| Repairs looking low | Can I see the capital ledger? |
None of this uses market benchmarks. The script finds arithmetic and omissions, not opinions about what a property should cost to run. A flag is a question, not a verdict. After the arithmetic, tie rent to the rent roll, collections to bank statements and taxes to the tax bills.
Run the same checks without code
Pacific Software Ventures (PSV) publishes free CRE tools. PSV's T12 analyzer totals income, expenses and NOI, annualizes the last 3 months and the last month, and flags totals that do not add up, one-month spikes and missing lines. It runs in the browser and does not upload your numbers: T12 analyzer and guide.
Disclosure: PSV made that tool, and sells AI consulting and training to commercial real estate firms. This is general education, not investment, legal or tax advice.

