AI for Net Lease Due Diligence: What It Can Do, What It Can’t, and How to Evaluate a Tool

Net lease due diligence is mostly reading. A single-tenant acquisition can involve the lease and every amendment, a guaranty, the purchase and sale agreement, a title commitment and its exception documents, a survey, an environmental report, a property condition report, an estoppel and often a subordination, non-disturbance and attornment agreement. Each document must be read, compared to the others and reconciled with the underwriting before the deposit goes hard.

That makes diligence a natural place to apply AI, and also a place where the limits of AI matter most. This guide separates the tasks where current AI is dependable from the decisions that remain with people, and then sets out how to evaluate a tool before trusting it with a live deal. It is written for the acquisitions and asset management teams at net lease investment firms, not for brokers or lenders, though much of it applies to both.

A working definition

In this guide, “AI” means large language models and related document-understanding systems that read text and images and produce structured data or prose. These systems are good at reading and comparing at speed. They are not reliably good at knowing when they are wrong. The U.S. National Institute of Standards and Technology, in its Generative AI Profile (NIST AI 600-1, July 2024), names this risk “confabulation”: “the production of confidently stated but erroneous or false content.” Every practice recommended below exists to contain that risk.

Where AI is reliable

“Reliable” here means the task has a verifiable answer in the documents, the output can be checked against a cited source, and an error is likely to be caught by a reviewer before it matters.

Reading offering memorandums

An OM is a marketing document with a predictable structure: tenant overview, lease summary, rent schedule, property description, location and demographics. AI can extract the headline terms (tenant, guarantor, price, cap rate, lease type, term remaining, escalations, options) and screen them against a buy box in seconds. The practical value is coverage: a firm can read every offering it receives rather than the ones an analyst has time to open.

The caution is that an OM is the seller’s summary. AI can read it accurately and still return a figure that the lease will later contradict. OM extraction is a screening tool, not a diligence finding.

Extracting lease terms with citations

This is the core diligence use. Given the lease and its amendments, AI can extract rent, escalation mechanics, commencement and expiration dates, renewal and termination options with their notice windows, landlord and tenant maintenance obligations, assignment provisions, rights of first refusal or offer, co-tenancy and go-dark provisions, and the guaranty.

The output is only as useful as its citations. Each extracted term should link to the page and passage it came from, so a reviewer can confirm it in seconds rather than rereading the lease. Two points deserve particular attention:

• Amendments. A later amendment can change rent, extend the term or delete an option. The tool must read the documents in order and report which document controls each term.

• Tables and schedules. Rent schedules are often set out in tables that span pages or are embedded as images. These are where extraction most often fails, and where reviewers should look first.

Comparing estoppels to the lease

An estoppel certificate is the tenant’s statement of the lease’s status: the documents that make up the lease, the current rent, the dates, whether rent has been prepaid, and whether the tenant claims any landlord default. Comparing each statement against the lease is exacting, repetitive work done under deadline pressure close to closing, which makes it a strong fit for AI. A good tool flags every mismatch, such as a rent figure that does not match the schedule, an amendment the tenant lists that is not in the file, or a qualification added to a default representation, with citations to both documents.

Deciding what a discrepancy means, and whether to require a corrected estoppel, remains a human decision.

Flagging missing documents

AI can compare what has been received against what should exist. If the lease refers to a first amendment and a commencement date memorandum, and only the base lease is in the data room, the tool can say so. It can also compare the documents listed in the estoppel against the file, and compare the title exceptions against the documents provided. Missing amendments are one of the most common causes of an inaccurate lease file, and this is a check that people tend to skip under time pressure.

Tracking the PSA timeline

Given an executed purchase and sale agreement, AI can extract the effective date, the inspection or due diligence period, deposit amounts and when they become non-refundable, title and survey objection deadlines, and the closing date, and convert them into a dated schedule. The calculation itself should be shown, because PSA date mechanics (business days, “no later than,” extension rights) are where errors hide.

Drafting first versions

AI can produce competent first drafts of an investment committee memo, a letter of intent, a diligence issues list, a title objection letter or an estoppel request. The value is a draft that is already populated with the deal’s facts and cites them. The draft still needs an author: someone who decides what the memo argues and what the letter asks for.

Where humans must decide

The common thread in these areas is that the answer is not in the documents. It depends on judgment, on context the documents do not hold, or on professional responsibility.

Pricing

AI can assemble comps, compute returns and show how price moves with an assumption. It cannot decide what the firm should pay. That depends on the firm’s cost of capital, its portfolio concentration, its view of the market, its relationship with the broker and its competitive position on the deal.

Credit judgment

In net lease, the tenant’s credit is often the investment. AI can summarize public filings, flag a guarantor that differs from the operating entity, or surface news about store closures. Whether a tenant will pay rent for the next fifteen years, and whether a particular location matters to that tenant, is an underwriting judgment.

Legal interpretation

AI can find every clause that bears on a question. It should not be the final word on what an ambiguous clause means, how a court in the relevant jurisdiction would read it, or whether a title exception is acceptable. Those calls belong to counsel. The American Bar Association’s Formal Opinion 512 (July 2024), which addresses lawyers’ use of generative AI, emphasizes competence, confidentiality and the need to review AI output; a firm working with outside counsel should expect counsel to apply that standard.

Negotiation

AI can draft a markup and explain the difference between two versions. It cannot read the other side, decide which point to concede, or judge when a seller’s deadline is real. Those decisions shape the deal and should stay with the people accountable for it.

The decision to proceed

Ultimately, an investment committee approves a deal on the strength of the people who present it. AI can make the record more complete and easier to check. It does not carry accountability for the outcome.

A division of labor

How to evaluate an AI diligence tool

Vendor demonstrations tend to use clean, digitally generated documents. Diligence documents are not clean. Evaluate any tool on your own files, including the worst ones.

Source citations

Every extracted term and every answer should link to the exact page and passage. Test this by clicking through a sample of citations and confirming the passage actually supports the figure. A citation that points to the right document but the wrong page is not a citation. Ask what the tool does when it cannot find a term: it should say so rather than produce a plausible value.

Scanned and poor-quality PDFs

Older leases are often scanned, sometimes faxed, sometimes with handwritten initials and marginal changes. Test the tool on your oldest and least legible lease, on a rent schedule embedded as an image, and on a document with a handwritten amendment. Ask how the tool reads images: whether it relies on optical character recognition alone or reads the page layout, and how it signals low confidence.

Data privacy and model training

OMs and diligence documents are often covered by confidentiality agreements. Before uploading anything, establish in writing:

• whether your documents or outputs are used to train or improve any model, the vendor’s or a third party’s;

• which model providers process your data, and on what contractual terms;

• where data is stored and how long it is retained;

• whether your data is logically isolated from other customers’ data; and

• what independent assurance the vendor holds. A SOC 2 Type I report assesses the design of controls at a point in time; a Type II report tests their operation over a period. Ask which one the vendor has and what it covers.

Audit trail

When a figure in an IC memo is questioned six months later, the firm should be able to see where it came from, whether a person changed it, who and when. Ask whether the tool records the source of each value, edits and approvals, and whether that history can be exported.

Permissioning

Not everyone in the firm should see every deal, and an AI tool that can search across all documents can also surface documents to the wrong person. Ask how access is controlled by deal, property or team, whether the AI’s answers respect those permissions, and what the tool is allowed to do on its own, such as sending email or changing records, versus what it can only propose.

Fit with the rest of the lifecycle

A tool that produces an excellent abstract in isolation still leaves someone to re-key it into the model, the calendar and the ledger. Ask where the extracted data goes next, and whether the diligence record carries forward to asset management after closing.

Running a fair test

1. Choose three to five closed deals where you know the right answers, including at least one with amendments and one with a scanned lease.

2. Have the tool process the full document set without preparation.

3. Compare its output to your final abstracts and closing records term by term, and count both errors and omissions.

4. Click through a sample of citations for every document type.

5. Review the data handling terms with counsel before moving to live deals.

The goal of the test is not to find a tool that is never wrong. It is to find one whose errors are visible, cited and easy to correct, so that the people accountable for the deal spend their time on judgment rather than on transcription.

Where Rets fits

Rets handles the reading so your team can decide. It reads every OM in your inbox and scores it against your buy box, extracts terms with page citations, and builds the rent schedule and returns from the lease. Diligence runs on the PSA clock, estoppels are compared to the lease, and the IC memo goes to Word with LOIs drafted in track changes. Chat answers with sources, also from Claude and ChatGPT. Data is workspace-scoped, never used to train models, and covered by SOC 2 Type I.