AI in Acquisitions: What It Catches in a T-12, and What It Misses

Rate snapshot

10-year Treasury5.11%September 23 close, the latest posted print
2-year Treasury4.85%September 23 close; 26 bp inside the 10-year
Fed funds target range3.75% to 4.00%Raised 25 bp September 16, 2026
Coupon used in the sizing below7.46%10-year plus an assumed 235 bp industrial spread

An offering memorandum lands with a T-12, a rent roll, and 22 leases. An AI tool reads all of it in about four minutes and produces a clean, structured, correctly extracted underwriting file. On the deal below, that file supports an NOI of $1,435,912 and it is wrong by $388,778.

Not because the extraction failed. The extraction was fine. It is wrong because 68% of the adjustments that matter are not in the documents at all. They come from a county assessor’s reassessment practice, an insurance renewal quote, a submarket vacancy survey, and a view on what it actually costs to manage a 22-tenant building. None of that is in the PDF the broker sent, so none of it is in the output.

That is the useful frame for AI in acquisitions. It compresses the part of the work that was already cheap and leaves the expensive part exactly where it was. On this deal the difference between those two categories is worth $2,300,529 of loan proceeds.

What the offering memorandum says

A 142,000 square foot multi-tenant small-bay industrial park with 22 tenants, marketed at a 6.75% cap on the seller’s stated NOI.

As presentedAmount
Gross potential rent$1,718,000
Expense reimbursements$232,000
Other income$41,000
Vacancy and credit loss at 2.0%-$39,820
Effective gross income$1,951,180
Operating expenses excluding management$486,000
Management fee at 1.5% of EGI$29,268
Total operating expenses$515,268
NOI as presented$1,435,912
Asking price at a 6.75% cap$21,272,775
Price per square foot$150

Nothing in that table is fabricated. Every line traces to a document. It is still not the number a lender will underwrite.

The normalization, line by line

AdjustmentEffect on NOIWhere it comes from
NOI as presented in the offering memorandum$1,435,912
Remove one-time lease termination fee booked in other income-$64,000T-12 line item, March
Remove reimbursements from a tenant that vacated in month 9-$38,500Rent roll against T-12
Reset vacancy and credit loss from 2.0% to 6.0%-$73,490Submarket survey
Reset management fee from 1.5% to 3.0% of EGI-$23,988Market standard; seller self-manages
Deduct replacement reserves at $0.15 per SF-$21,300Lender requirement, not in the T-12
Reassess real estate taxes at the transaction price-$118,000County millage applied to the sale price
Mark insurance to the current renewal quote-$27,500Binder dated this month
Normalize repairs and maintenance to a three-year average-$22,000T-12 against T-36
Normalized NOI$1,047,134
Total reduction-$388,77827.1% of the presented NOI

None of these adjustments is aggressive. Every one of them is standard practice, and a lender’s credit committee will make all eight whether the buyer does or not. The only question is whether the buyer finds them before or after signing a contract.

What it does to the loan

At the asking price of $21,272,775On the OM’s NOIOn normalized NOI
NOI$1,435,912$1,047,134
65% LTV on the ask price$13,827,304$13,827,304
1.30x DSCR at 7.46%, 25-year amortization$12,499,558$9,115,262
9.5% debt yield$15,114,866$11,022,466
Loan amount$12,499,558$9,115,262
Binding test1.30x DSCR1.30x DSCR
Equity required at the ask price$8,773,216$12,157,513
The consequencesAmount
Loan proceeds lost to normalization$3,384,297
Additional equity the buyer must fund at the ask price$3,384,297
Value at the normalized NOI and the same 6.75% cap$15,513,101
Implied overpayment at the ask price$5,759,674
Overpayment as a share of the ask27.1%
Price per SF at the normalized value$109

A buyer who underwrites the OM’s NOI and gets a term sheet sized off it will discover the gap in week six, when the lender’s own analyst reruns the numbers. By then the buyer is hard on a deposit and negotiating from behind.

The part that matters for AI

Split the same eight adjustments by whether the information needed to make them was in the package the seller sent.

In the documents the seller providedEffect on NOI
Remove one-time lease termination fee-$64,000
Remove reimbursements from the vacated tenant-$38,500
Normalize repairs and maintenance to a three-year average-$22,000
Subtotal-$124,500
Not in the documents; requires outside knowledgeEffect on NOI
Reset vacancy to the submarket rate-$73,490
Reset management fee to market-$23,988
Deduct replacement reserves-$21,300
Reassess real estate taxes at the transaction price-$118,000
Mark insurance to the current renewal quote-$27,500
Subtotal-$264,278
The splitValue
Total adjustment-$388,778
Share found in the documents32.0%
Share requiring information the seller never sent68.0%
Loan on NOI adjusted only for what the documents show$11,415,791
Loan on fully normalized NOI$9,115,262
Proceeds difference the outside knowledge is worth$2,300,529

An AI tool that reads every page perfectly captures $124,500 of the $388,778. It will not know that the county reassesses at sale, because that fact lives in a millage table and a transfer practice, not in the T-12. It will not know what insurance actually renewed at, because the binder is dated after the package was assembled and was never in it. It will not challenge a 2.0% vacancy assumption, because 2.0% is what the document says and the document is its entire world.

The largest single adjustment on this deal, the $118,000 tax reassessment, is also the one furthest outside the documents. That is not a coincidence. The items a seller’s package is least likely to reflect are exactly the items that move NOI most, because a seller assembling a package has no incentive to volunteer them.

Simplifications: this is a constructed example built to be representative, not a specific transaction. The eight adjustments and their magnitudes are illustrative assumptions chosen to reflect common normalization items; actual adjustments vary enormously by market, asset, seller and lender. The tax reassessment assumes a jurisdiction that reassesses on transfer, which many do not. The 235 basis point spread, 1.30x DSCR, 65% LTV, 9.5% debt yield and 25-year amortization are modeling inputs rather than quoted terms. Value is computed by capping NOI at a single rate, which ignores lease term, tenant credit, rollover timing and capital needs, all of which a real appraisal weighs. No transaction costs, closing costs or capital reserves beyond the $0.15 per SF replacement reserve are modeled.

The one check that finds the most money

If you only run a single reconciliation, run this one: take the rent roll, annualize the contractual rent, and tie it to the rental revenue line on the T-12. On a stabilized building those two numbers should land within about a percent of each other.

When they do not, the gap is always telling you something specific. Rent roll higher than the T-12 usually means a tenant moved in partway through the year, or a lease is signed but not yet paying, or somebody is in arrears. T-12 higher than the rent roll usually means income is in the trailing period that will not repeat, which is exactly how the $64,000 termination fee and the $38,500 of reimbursements from a departed tenant surfaced on this deal.

That single check accounted for $102,500 of the $124,500 that was findable in the documents. It takes about twenty minutes by hand and an AI tool can set it up in one. What the tool cannot do is decide what an unexplained variance means, because the answer usually lives in a conversation with the property manager rather than in a file.

How to structure the work

The sequencing matters more than the tooling. Run extraction first and treat its output as a draft dataset, not an underwrite. Reconcile second, because reconciliation is what converts extracted numbers into trustworthy ones. Only then apply the adjustments, and apply them in two passes: the ones the documents support, then the ones that require you to go get information nobody sent you.

Keeping those two passes separate is the discipline worth adopting. It makes visible, on every deal, how much of your underwriting rests on the seller’s package and how much rests on your own work. On this deal the answer was 32% and 68%. A screening model that stops after the first pass is not conservative, it is incomplete by a known and measurable amount.

It also changes what you ask for in diligence. If you know before you tour the building that the tax reassessment and the insurance renewal are worth $145,500 of NOI between them, those two items go at the top of the diligence list instead of the bottom, and they get resolved in week one rather than week six.

What to let AI do, and what to check by hand

TaskLet AI draft itWhat a human has to do
Extract the rent roll into a structured tableYesSpot-check five leases against the source PDFs
Categorize twelve months of operating expensesYesConfirm the category map matches your own chart of accounts
Abstract escalations, options and expense stopsYesRead the three largest leases yourself, in full
Tie annualized rent roll to the T-12 revenue lineYesInvestigate every variance over 1%, personally
Flag non-recurring income itemsYesDecide what is actually non-recurring; that is judgment
Reassess property taxes at the sale priceNoCall the assessor or pull the millage table
Mark insurance to current costNoGet a real quote from a real broker on this building
Set the vacancy and credit loss assumptionNoUse your own submarket data, not the seller’s
Confirm occupancy and tenant healthNoEstoppels, site visit, and a look at who is actually paying
Build the normalized NOI bridgeYesOwn every line of it in front of your investment committee

Verification rules worth writing down

  1. Every extracted number carries a source: document name and page. If the tool cannot produce that, the number does not go in the model.
  2. Reconcile before you analyze. Annualized rent roll should tie to the T-12 rental revenue line within about 1%. A larger gap is a finding, not a rounding issue.
  3. Never let the tool infer a missing value. A blank cell is information. A silently interpolated one is a defect you will not notice.
  4. Check the square footage three ways: rent roll, offering memorandum, and appraisal or survey. They disagree more often than anyone expects, and every per-foot metric depends on it.
  5. Count the tenants. If the rent roll lists 22 and the lease folder holds 19, ask what happened to three before you ask anything else.
  6. Run the expense ratio against your own comparable set. An industrial building at a 26% expense ratio is either exceptionally run or missing a line item.
  7. Treat any item the tool flags as “one-time” as a question, not an answer. Some one-time items recur annually.
  8. Re-run the whole normalization after diligence, with actual quotes and actual assessments substituted for assumptions. The first pass is a screen, not an underwrite.

Why this matters more right now

Two forces are pushing in opposite directions. The Federal Reserve’s July 2026 loan officer survey reported that “Moderate and modest net shares of banks reported having eased standards for loans secured by nonfarm nonresidential (NFNR) properties and multifamily properties, respectively.” Easier standards mean more deals get term sheets.

At the same time, the Mortgage Bankers Association reported that “Seventeen percent ($875 billion) of the $5.0 trillion of outstanding commercial mortgages held by lenders and investors is scheduled to mature in 2026.” A large maturing cohort means more properties coming to market, marketed by sellers under time pressure, with packages assembled to sell rather than to underwrite.

More deals moving faster through a market with more motivated sellers is precisely the environment in which a 27.1% NOI overstatement gets signed. Speed is the thing AI is best at, and speed is not what this problem needs.

How QuadBlock can help

QuadBlock Capital provides bridge, transitional and permanent financing from $250,000 to $30 million on multifamily, industrial, retail, self-storage, specialty assets and ground-up SFR and townhome communities, in every state except Oregon and Nevada. We normalize the T-12 before we quote, which sometimes means telling a sponsor the deal sizes smaller than they hoped, early enough for it to be useful. Send the OM, rent roll and T-12 to us at quadblockcapital.com and we will show you the bridge from the seller’s number to ours.

This article is for general educational purposes and is not tax, legal, or investment advice, and it is not a rate quote or a commitment to lend. The property, offering memorandum, adjustments and all figures in the worked example are constructed illustrations, not a real transaction, and are not representative of any specific deal, seller or market. Normalization practice varies by lender, market and asset class; property tax reassessment on transfer applies in some jurisdictions and not others. Spreads, coupons, DSCR, LTV, debt yield and amortization assumptions are modeling inputs, not quoted terms. Treasury yields and the Fed funds target range cited are as of the dates shown and change daily. Nothing here is a forecast of future interest rates. All figures in the worked example are illustrative and do not represent QuadBlock Capital loan terms, pricing, or leverage.

Sources

  1. U.S. Department of the Treasury, Daily Treasury Par Yield Curve Rates, September 2026
  2. Federal Reserve, FOMC statement, September 16, 2026
  3. Federal Reserve, July 2026 Senior Loan Officer Opinion Survey on Bank Lending Practices
  4. Mortgage Bankers Association, 17 Percent of Commercial and Multifamily Mortgage Balances to Mature in 2026
  5. Mortgage Bankers Association, Commercial and Multifamily Mortgage Debt Outstanding Crosses $5 Trillion in First-Quarter 2026
  6. Freddie Mac Multifamily, package requirements including certified property financials
Scroll to Top