
What AI can do with a deal
Deal analysis has a slow part and a hard part. The slow part is getting numbers out of documents and into a model. The hard part is deciding which numbers to believe. AI helps a great deal with the first and only a little with the second.
- Extraction: reading a rent roll, a trailing 12-month statement and an offering document, and laying out units, rents, lease dates, income and expense lines.
- Calculation: NOI, cap rate, and the loan metrics under the assumptions you set.
- Flags: leases expiring soon, one tenant with a large share of income, expenses that look low, income items that may not repeat.
- Scenarios: what happens to coverage if vacancy rises, rents fall or the rate is higher.
- Summary: a short written description of the deal and its open questions.
The benefit is volume and speed. An investor can screen ten deals in the time one took, and spend the saved hours on the two worth pursuing.
What to give it
Results depend on inputs. Give the tool the same documents a lender would ask for first.
- A current rent roll with unit or suite, tenant, size, rent, lease start and lease end.
- A trailing 12-month income and expense statement, month by month if you have it.
- The purchase price or your estimate of value.
- The loan you have in mind: amount, rate assumption, amortization and any interest-only period.
- Known facts that are not in the files: a tenant that gave notice, a tax reassessment after sale, planned capital work.
Clean files work better. Spreadsheets are read more reliably than scanned pages. Label whether figures are monthly or annual. If the seller's statement mixes property expenses with owner expenses, say so. What lenders look for in these two documents is covered in how lenders read a rent roll and a T-12.
A worked example
Say an AI tool reads the files for a property offered at $8,000,000 and reports NOI of $560,000. You tell it to test a $5,600,000 loan with annual debt service of $448,000 (a hypothetical figure for the example). It returns these numbers.
| Metric | Formula | Result |
|---|---|---|
| Cap rate | NOI / price | $560,000 / $8,000,000 = 7.0% |
| Loan-to-value | Loan / value | $5,600,000 / $8,000,000 = 70% |
| DSCR | NOI / annual debt service | $560,000 / $448,000 = 1.25x |
| Debt yield | NOI / loan | $560,000 / $5,600,000 = 10.0% |
On paper the deal clears common lender tests. Now you check the source. The operating statement includes a $40,000 one-time insurance reimbursement that the tool counted as other income. A lender will take it out. Underwritten NOI becomes $520,000.
With the same loan, DSCR falls to about 1.16x ($520,000 / $448,000) and debt yield to about 9.3%. If the lender requires 1.25x, the property supports debt service of $416,000 ($520,000 / 1.25), which at the same loan terms is a $5,200,000 loan. One misread line moved the loan by $400,000. The calculation was right both times. The input was wrong. For the ratios themselves, see DSCR explained and debt yield explained.
Where AI analysis goes wrong
- One-time items treated as recurring: insurance proceeds, lease termination fees, tax refunds.
- Monthly and annual figures mixed: a monthly rent column read as annual, or the reverse.
- Seller numbers taken at face value: pro forma rents presented as actual, or expenses that leave out management and reserves.
- Missing lender adjustments: lenders apply a vacancy allowance, a management fee and replacement reserves even if the owner does not pay them. See how lenders calculate NOI.
- Taxes after sale: in many places property tax is reassessed on a sale, so the seller's tax bill understates yours.
- Misread documents: scanned pages, merged cells, totals counted twice.
- Confident gaps: when a fact is missing, the tool may assume one and not say so.
- No market knowledge it can prove: a statement that rents are below market needs evidence from comparable properties.
None of these errors announces itself. The report looks equally polished whether it is right or wrong.
How to check the output
A short, fixed routine catches most problems.
- Tie the totals: total rent on the rent roll, and total income and total expenses on the operating statement, should equal what the tool used.
- Read every income line other than rent and ask whether it will repeat.
- Compare each expense line with what is normal for the property type. Ask why anything is low.
- Restate NOI the way a lender would, with vacancy, management and reserves.
- Recalculate the key ratios yourself. It takes two minutes.
- Ask the tool to list its assumptions and anything it could not find.
- Size the loan under each lender test and see which one limits it, following how to size a commercial loan.
If the deal still works after that, it is worth real time.
What it does not replace
A first-pass analysis answers one question: is this deal worth a closer look? It does not replace the lender's underwriting, which rests on an appraisal, third-party reports, lease reviews and the sponsor's financials. It does not replace your own inspection of the property or your knowledge of the market, including what a reasonable cap rate is for that asset in that location.
It does not produce a loan offer either. The rate, the proceeds and the terms come from a lender's quote. Treat AI analysis as a screening aid, and confirm terms with the lender and your own counsel before you commit money.
Where CREaiD helps
With CREaiD you can upload deal data for a quick investment analysis, then describe the deal to the Lender Matchmaker to get lenders that fit from its database of 1,500+ commercial real estate lenders.
See deal analysis, and check the results against your source documents as you would with any tool.
Key takeaways
- AI is fast at extracting numbers and calculating first-pass metrics.
- Wrong inputs, not wrong arithmetic, cause most errors.
- One-time income and missing lender adjustments are the most common problems.
- Tie totals to the source documents and recalculate the key ratios yourself.
- A first-pass analysis is a screen, not underwriting and not a loan quote.
Questions people ask
Can AI underwrite a commercial real estate deal?
It can produce a first-pass version of the numbers. Underwriting also means verifying the documents, judging the tenants, the market and the sponsor, and taking responsibility for the conclusion. That part stays with people, and the lender does its own.
What documents do I need for an AI deal analysis?
A current rent roll and a trailing 12-month income and expense statement are the minimum. Add the price or value and the loan assumptions you want tested. Leases, tax bills and capital plans improve the result.
How accurate are the numbers?
The arithmetic is usually right. The risk is in what gets fed into it: a misread figure, a one-time item or an optimistic seller number. Check the inputs against the source documents before relying on any ratio.
Can AI tell me how big a loan I will get?
It can estimate what loan the income supports under assumptions you choose for DSCR, loan-to-value and debt yield. The real amount depends on the lender's underwriting, the appraisal and the rate on the day. Treat the estimate as a range to test with lenders.
Is a spreadsheet model still necessary?
For any deal you pursue seriously, yes. A model you built or understand lets you see every assumption and change it. AI is useful for filling that model faster and for a second look at what you may have missed.
CREaiD shows the lenders that really made your kind of loan in your state, from public loan records.
