Using AI for Property Investment: Screening a Deal Step by Step

A property investor can use AI to screen a deal by sifting listings against written pass-fail rules, extracting each survivor's figures into a sourced deal sheet, and calculating yield and cash flow under stated assumptions. Verify purchase costs, achievable rent, condition, operating charges, finance terms and legal restrictions by hand before relying on the result or making an offer.

The first pass is a rejection filter, not a valuation. A listing can make rent look certain while hiding an empty unit, deferred repairs or an expense paid by the owner. Keep unknowns visible as unknowns. A precise-looking AI percentage built on an unverified rent estimate is still an unverified estimate.

Follow me on Instagram@sagnikteaches

Fix the pass, investigate and reject rules first

Decide what a candidate must satisfy: property type, maximum total cash needed, minimum cash reserve after purchase, acceptable monthly shortfall and conditions that require specialist advice. The screen should return “reject”, “investigate” or “advance” to deeper review. It should never return “buy”. If an investor has no explicit threshold, the model can always produce a persuasive narrative around a weak deal.

Connect on LinkedInSagnik Bhattacharya

For an illustrative investor, the rule might be: no more than $130,000 of initial equity and costs, positive annual cash flow under the stated finance terms, and a reserve equal to six months of debt and essential property costs after completion. These are the investor's example preferences, not universal recommendations. Another investor may accept a lower current return for a different objective, but should write that objective and the permitted trade-off down first.

Subscribe on YouTube@sagnikteaches

Record the date, source and confidence for every figure. “Seller states monthly rent $1,800” is a different input from “three signed leases show $1,800 actually collected”. Ask AI to keep those separate. The difference between stated, contracted and collected rent often decides whether a screen is useful. A unit with no tenant needs an achievable rent range, supporting comparable evidence and a vacancy assumption.

Sift five listings in one pass before building any deal sheet

Most listings should fail quickly, so run the first screen on a batch rather than one property at a time. Paste the text of each listing, the rent evidence you hold and your pass-fail rules into one prompt, and ask AI to return the same columns for every listing. The value is consistency: the listing you have already fallen for gets exactly the same test as the four you have not.

For this illustrative batch, the quick test has three rules. Gross yield on the asking price must be at least 7.5%, using contracted or documented rent only. Cash needed must stay within $130,000, estimated as a 40% deposit plus a 5% acquisition allowance plus any stated works. And the seller must allow an inspection before an offer. The 60% loan implied by the deposit figure is a sorting assumption, not a finance offer.

All five listings and every figure in the table are invented for illustration. A batch prompt that produces it can be short, provided it names the evidence rule and forbids ranking.

Screen these five listings with the same three rules below.
For each listing return: asking price | rent figure | exact sentence
the rent came from | evidence type (contracted, documented, claimed,
missing) | gross yield on asking price | cash needed using my formula |
inspection access | result: advance, investigate or reject |
the rule that decided it.
Use only contracted or documented rent for the yield rule.
Do not rank listings and do not recommend a purchase.

In an illustrative first run, the model puts C at the top with a 10.7% yield, because the agent's “potential rent” line sat beside the price, and marks it “advance”. That breaks the evidence rule it was given. It also works out E's yield on $2,100, an “achievable after redecoration” figure from the agent's notes, which lifts E to about 9.5% instead of the 8.8% the tenancy schedule supports. Correct both by hand, then re-run with the source-sentence column visible. Check every “reject” as carefully as every “advance”: a misread price can kill a sound listing as easily as it can promote a weak one.

After the sift, only A goes forward to the full sheet below, and E waits for one document. Write one line on why B, C and D failed, so that a price cut or a new lease can be re-tested in minutes rather than from scratch. Requesting leases, bills and surveys for all five would have spent most of the effort on properties that fail a one-line test.

Build one deal sheet with source links and missing fields

Start with purchase price, closing and acquisition costs, immediate works, expected rent, vacancy, service charges, insurance, maintenance reserve, management, utilities, finance amount, interest, repayment structure and any known restrictions. Add document references rather than uploading every document to a public chat. Redact personal data if an approved tool must process a lease or statement.

All values in this sheet are illustrative. An asking price is not a settled purchase price, a lender quotation can change and an expense allowance is not a bill. Put an “evidence status” beside each field: documented, quoted, estimated or missing. Reject a result that silently turns estimated into documented when AI reformats the table.

A useful extraction prompt is: “Return a table of values stated in these supplied documents. For each value give the document name, page or line, date, exact wording, unit and any conflicting value. Leave missing fields blank. Do not estimate rent or condition.” The prompt output is a transcription aid. It is not proof that the supplied documents are accurate or complete.

Consider an illustrative listing that says “potential rent $2,050”, while the attached tenancy schedule says $1,700. AI should produce two entries with their sources and flag the conflict. It should not average them to $1,875. The investor needs to determine whether $2,050 depends on refurbishment, a new tenant or a marketing claim. Until then, test the current contracted figure and a separate evidenced upside case.

For a broader analysis after this initial gate, the rental deal vetting tutorial covers a deeper investment review. This screen stays deliberately narrow: can the inputs and arithmetic justify spending more time on due diligence?

Recalculate gross yield and operating income yourself

Gross yield is annual scheduled rent divided by a defined cost base. Some people divide by purchase price and others by total money invested; label the denominator every time. In the illustrative deal, $1,800 × 12 gives $21,600 annual scheduled rent. Against the $240,000 asking price, gross yield is 9%. Against $270,000 including the $12,000 acquisition allowance and $18,000 works, it is 8%. Both are mathematically true and answer different questions.

Now deduct vacancy and operating costs before finance. Assume an illustrative 5% vacancy allowance of $1,080, $2,400 maintenance reserve, $1,728 management at 8% of scheduled rent, $1,200 insurance, $1,800 service charges and $600 owner-paid utilities. Those deductions total $8,808, leaving $12,792 net operating income. Net operating income here means rent after property operating costs but before finance and personal taxes. Divide $12,792 by the $270,000 total acquisition and works cost to get about 4.74% on that cost base.

Check the arithmetic with a calculator or spreadsheet independent of the chat output. Ask AI to show each line, formula and denominator, then reproduce the formulas yourself. If an AI answer calls 9% “net yield”, you have found a classification error even if the division was right. Do not let a model remove an expense because it thinks a typical property would not incur it; the actual ownership arrangement decides who pays.

A second illustrative case is a small building advertised with $24,000 annual income. Its schedule shows one of four units vacant for three months. A simple “annual rent ÷ price” screen overstates the next year's likely receipts if the empty unit is assumed to pay immediately. Enter the expected void months and any letting costs explicitly. The screen may remain attractive, but its income line should describe the situation that will actually exist after completion.

Test finance and cash remaining after debt

In the main example, suppose a lender offers an illustrative $144,000 interest-only loan at 6%. The annual interest is $8,640. Deducting it from $12,792 operating income leaves $4,152 annual cash flow before income taxes, principal repayment and unexpected major work. The investor's initial equity and costs are $270,000 minus $144,000, or $126,000. Cash-on-cash return on that initial amount is approximately 3.30%. The monthly average of $346 disguises the fact that bills and vacancies arrive unevenly.

Interest-only finance leaves the principal outstanding. If the loan is amortising, use the lender's actual payment schedule rather than applying the interest-only illustration. Include arrangement fees, repayment charges, required reserves and renewal risk if relevant to the offer. A quoted rate that applies only before a review date is not a fixed cost for the whole holding period. An AI model can make a scenario table; the lender's written terms and a qualified finance adviser must establish what is available to this investor.

To stress the example, reduce scheduled rent by 10% to $19,440 and raise vacancy to 10% of that rent, or $1,944. At 8% management, the fee becomes about $1,555; keep the $6,000 of other operating costs. Net operating income is then about $9,941 and cash after the illustrative $8,640 interest is about $1,301. This remains positive on paper, but it leaves little room for a major repair. Ask what happens if the repair allowance is exceeded, the property stays empty longer or the lender's terms change.

An illustrative spreadsheet mistake shows why independent checks matter: AI reads a 6% annual interest rate as a 6% monthly rate and reports an enormous loss. A subtler mistake is applying 6% to $240,000 rather than the $144,000 principal. Require a formula column with units, compare the result with a hand calculation, and question any result that moves sharply when only one input changes.

For a focused explanation of denominator choices and calculation checks, the yield maths verification tutorial goes further. Here yield is one gate among several; it cannot clear a deal with missing condition, title or cash-reserve evidence.

Mark the red flags that arithmetic cannot settle

Use a short red-flag register beside the numbers. Include unavailable access for inspection, conflicting floor-area figures, undocumented alterations, abnormal service charges, a lease with unclear obligations, a tenant ledger that does not match stated rent, insurance exclusions and a works budget without a survey. AI can identify statements to investigate if you supply the relevant documents. It cannot confirm physical condition from listing photographs or supply missing legal facts.

Professional valuation guidance is explicit about inspection and checking the information relied upon. Standard VPS 4 in the RICS Red Book Global Standards says inspections and investigations must go as far as needed for a valuation that is professionally adequate for its purpose, and that the valuer must take reasonable steps to verify the information relied on. A quick AI investment screen is a different activity and must not be presented as such a valuation. Use an appropriate inspector, valuer or legal adviser when the issue is outside your competence.

For example, an illustrative property has a $15,000 lower asking price than its apparent comparables, but the owner has not provided a current service-charge budget. The correct screen status is “investigate”, not “cheap”. Ask for the budget, historic actuals and planned major works. Recalculate the income after the confirmed charge. The price difference alone cannot tell you whether the deal compensates for the obligation.

Another example: the listing says “recently refurbished”, while inspection notes mention recurring water ingress. Do not ask AI to estimate a repair amount from the adjective. Obtain a specialist assessment and written quotation. If access is refused or the seller will not provide essential records, decide whether the uncertainty is acceptable under the investment rule you wrote first.

Document review sits after this screen. The rental purchase due-diligence tutorial addresses how to read leases, surveys and searches with AI assistance. The screen should produce a list of documents to request, not a false conclusion that the documents have been checked.

Keep AI's recommendation auditable and conditional

Ask for a one-page decision note containing the input table, source status, base calculation, two stress cases, red flags, outstanding questions and decision owner. The result should state what would change its classification. “Investigate if verified rent is at least $1,800 and the repair quotation stays within the $18,000 allowance” is more useful than “promising opportunity”.

Make the final note distinguish known from assumed: “Purchase price: asking, not agreed”; “rent: seller claim, unverified”; “finance: illustrative”; “condition: inspection pending”. If the AI cannot cite the source of a number, remove that number until you can. Save the spreadsheet formulas and document links alongside the note, so another person can reproduce the decision rather than trusting a polished paragraph.

One illustrative rejection shows the value of this discipline. A candidate initially produces 4% annual cash-on-cash return using $1,900 monthly rent. The signed lease shows $1,550. With all else held constant, scheduled rent falls $4,200 a year before any management adjustment. The investor's minimum cash-flow rule may fail. The correct action is to re-run the sheet on evidenced income and either renegotiate or reject; the first AI score is not something to defend.

Review the screen again after an offer, inspection and finance proposal, because each new fact may change the result. Keep the date on each scenario. If an investor wants to compare several opportunities, use exactly the same definitions for gross rent, vacancy, operating expenses and capital required across all rows. AI can help enforce the format, but the investor must verify the inputs and decide whether the risks fit their own resources and advice.

Verify the rent line without treating an advert as evidence

Rental income deserves its own checklist because it drives every return calculation. If the unit is occupied, request the signed agreement, current payment schedule, arrears record and details of any incentive or included utility. Check dates: a rent due next month is not the same as rent collected over the last year. If the unit is empty, gather recent comparable lettings with similar size, condition and terms; then ask a relevant local professional to test the range. AI may organise these comparisons, but it cannot establish that a particular applicant will sign at the top of the range.

Suppose an illustrative owner supplies twelve bank entries of $1,600 but a marketing summary claims $1,850 “market rent”. If the current tenant remains, the base case uses the agreed and collected $1,600 unless a documented change is imminent. A separate upside case may use $1,850 with explicit assumptions about vacancy, works, letting costs and the time required. Combining the lower current expenses with the higher hypothetical rent hides the transition cost.

Ask AI to list every rent number in the supplied pack with its source and period. Then compare the extracted entries with the actual documents yourself. If a scanned schedule reads “1,600” as “1,800”, the difference is $2,400 over a full year. One character can materially alter the result. The review should include a visual check of every figure used in the calculation, not a sample of whichever figures look suspicious.

Check cost responsibility item by item

Operating cost headings are not interchangeable across properties. One owner may pay a shared-service charge while another passes it through to tenants. Insurance may be included in that charge or billed separately. Maintenance provisions may cover routine work but exclude a major roof, lift or heating replacement. Add a column for who pays, the evidence, whether the amount recurs and when it can change. That prevents double counting and omission at the same time.

In an illustrative second deal, the owner reports $3,000 annual service charges and $1,200 insurance. A statement shows that $1,200 insurance is already inside the $3,000. Deducting both understates net operating income by $1,200. The inverse error is more dangerous: assuming insurance is included when it is not. AI can flag a possible overlap from document labels; the investor should read the bill and ask the managing party which charges the owner actually bears.

Keep capital work separate from recurring expenses in the sheet. The $18,000 works allowance in the main example is part of initial capital, while the $2,400 annual maintenance reserve is an operating assumption. If a survey discovers a $9,000 urgent repair, decide whether it raises the initial works budget, replaces part of it or is a later expense. Do not put it in two places, and do not omit it because it was not in the seller's listing.

Use a stop list before spending on deeper checks

A screen can finish quickly if a hard condition fails. Write a stop list that matches your own capacity: total capital above the limit, cash flow below the minimum under the base case, insufficient reserve after completion, refusal to allow inspection, missing evidence for a key income claim or a material restriction that the investor cannot accept. Some failures are fixable by a revised price or better evidence; others end the enquiry. Keep the reason in the deal note so a later attractive headline figure does not restart a rejected case without new information.

For the illustrative $270,000 all-in deal, initial cash required is $126,000 before any separately held reserve. If the investor has exactly $126,000 available, the transaction fails a rule requiring a post-purchase reserve even though the cash-on-cash calculation is positive. AI often reports a return percentage without testing liquidity. Add “cash left after completion” as a separate field and subtract deposit, acquisition costs, works and any lender-required cash from funds actually available.

If the price falls by $10,000, do not simply add $10,000 to projected profit. Re-run all linked inputs: the loan amount may depend on price, acquisition costs may change, and the seller may have rejected a condition that gave the original proposal its protection. Record an alternative-price scenario beside the base case rather than overwriting the original. That makes the investor's reasoning visible when the negotiation moves.

Further reads

• AI to Find Investment Properties: Searches, Filters and Sanity Checks — Build a shortlist before screening.
• Property Investment AI Tools: A Stack for a Five-Property Portfolio — Plan tools for ongoing holdings.
• Rental Property Bookkeeping With AI: A Landlord's Monthly Routine — Track actual income after purchase.
• How to Build a 13-Week Cash Flow Forecast With AI Help — Understand short-term liquidity planning.
• AI Tools for Estate Agents: A Stack by Job, Valuation to Completion — See the agent-side property workflow.
• AI Commercial Real Estate Underwriting: A First-Pass Model — A worked six-suite example: clean the rent roll, reconcile the operating statement, model NOI and challenge the assumptions with AI.
• AI Property Development Feasibility: A First-Pass Appraisal — A worked first-pass appraisal for a small developer: inputs, GDV, costs, sensitivity and planning risks, with AI as challenger and a spreadsheet as calculator.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: RICS Valuation – Global Standards (Red Book, effective 31 January 2025), VPS 4 Inspections, investigations and records (checked 4 November 2026).