How AI Creates an Offering Memorandum

The pipeline in plain terms

When people say AI creates an offering memorandum, they usually imagine one large step. In practice it is six distinct stages, and only two of them involve anything most people would call intelligence.

The stages:

1. Extraction. Reading your source documents and pulling out the data.
2. Structuring. Organizing that data into a listing record with a schema.
3. Drafting. Generating narrative sections from the structured data.
4. Assembly. Building financial tables, maps, and demographic exhibits.
5. Rendering. Applying your brand and producing the document.
6. Review. A person verifying and approving before it goes anywhere.

Understanding the split matters because the value and the risk are unevenly distributed. Stages 1 and 3 are where AI does the work and where errors originate. Stages 2, 4, and 5 are mostly deterministic software that happens to sit downstream of AI. Stage 6 is where a licensed professional takes responsibility.

Vendors that describe the whole thing as one magic step are usually hiding how stage 6 works.

Step 1: extraction from source documents

You upload what the owner sent you: a rent roll, a trailing 12-month operating statement, lease abstracts, a tax bill, maybe a prior memorandum if the asset has traded.

The system reads these and identifies the values that matter. From a rent roll, that means suite numbers, tenant names, square footage, lease start and end dates, base rent, escalations, and options, line by line. From an operating statement, income and expense categories with amounts and periods.

What determines accuracy:

• Document quality. A typed PDF exported from property management software extracts far more reliably than a scanned fax from 2011.
• Layout conventions. Standard formats are handled well. Unusual column arrangements and merged cells cause trouble.
• Annotations. Handwritten notes, strikethroughs, and marginalia are where extraction most often goes wrong.

The requirement that matters most: every extracted figure should be traceable back to the page and location it came from. Without traceability, review means re-reading the source documents, which erases the time saving. With it, review is a fast confirmation pass.

This stage typically converts an hour or more of transcription into ten to twenty minutes of review, which is the largest single time saving in the pipeline.

Step 2: structuring into a listing record

Extraction produces values. Structuring turns them into a record with meaning: property fundamentals, tenancy, financials, market context, media, and narrative fields.

This stage is mostly conventional software, and it is the part that makes everything downstream work. Once one record holds every figure, the memorandum, the brochure, the property website, and the email campaign can all derive from it. They cannot disagree, because there is only one copy of each number.

It is also where derived values get computed: occupancy from the rent roll, net operating income from income minus expenses, cap rate from net operating income and price, weighted average lease term from the tenancy detail. Computing these rather than transcribing them removes a whole category of arithmetic error.

The practical consequence shows up on revision. When the seller cuts the price two days before launch, one field changes and every derived figure and every downstream document updates. Our common OM mistakes guide covers what happens when that update is manual.

Step 3: drafting the narrative

This is the stage people mean when they say AI wrote the memorandum.

From the structured record, the system drafts the sections that are prose rather than data: property overview, location description, tenant summaries, submarket commentary, and a first pass at investment highlights.

What it does well. Descriptive sections where the correct content follows from the facts. A property overview that accurately summarizes size, construction, parking, and condition is a competent draft in seconds. Location descriptions and tenant summaries are similar.

What it does adequately. Investment highlights. It will find real patterns in the data, such as below-market rents or upcoming rollover, and state them. The output is usually correct and usually generic.

What it should not finish. The investment thesis. A generated thesis reflects what the numbers show, not judgment about which buyer to target, how aggressively to price the story, or which risk to lead with. That is the part of the document that differentiates your firm, and it deserves your time. Our how to create an offering memorandum guide covers writing it well.

The honest framing: drafting turns a blank page into a solid B-minus in seconds. Getting to an A is still your job, and you now have the hours to do it.

Step 4: financial tables, maps, and demographics

Almost entirely deterministic, and a large share of the practical time saving.

Financial tables. Operating statement, rent roll, expiration schedule, and pro forma rendered in consistent formats with correct arithmetic and labels. Formatting these by hand is hours of work per memorandum and a reliable source of typos.

Maps. Location, aerial, and trade area maps generated at appropriate zoom levels from the property address.

Demographics. Population, income, household counts, and employment pulled for the relevant radius at production time rather than copied from a prior deal. This is one of the more valuable stages precisely because manual demographic work is tedious enough that teams reuse stale figures.

Charts. Rollover schedules, expense breakdowns, and rent comparisons rendered from the same data as the tables, so they cannot disagree with them.

Step 5: branding and layout

Your template, applied automatically: cover layout, typography, color, logo placement, section structure, contact block, page numbering, and table of contents.

The important design decision here is what is locked. Disclaimer language, brand elements, and financial table structure should not be editable per deal. Narrative, imagery, and section emphasis should be. Our offering memorandum template guide covers that split in detail.

Output should include both the full memorandum and, for confidential offerings, a teaser version generated from the same record with disclosure rules applied, rather than a copy edited by hand. Hand-redacted versions are where confidential tenant detail leaks.

Step 6: human review and approval

The stage that makes the rest defensible.

Three passes: numbers verified against sources, positioning confirmed by the listing broker, and a cold read by someone uninvolved. Our offering memorandum checklist is built for exactly this.

Enforced, not assumed. Approval should be a gate the document must pass before export or sharing, not a habit. Habits fail under deadline, which is precisely when errors are most likely.

Accountability stays human. A licensed professional is answerable for what reaches an investor. No generation pipeline changes that, and any vendor implying otherwise is overselling.

What AI still gets wrong

Be specific about this during evaluation.

• Poor-quality scans. Accuracy degrades meaningfully, and the failure is often silent rather than obvious.
• Unusual lease structures. Percentage rent, complex recovery arrangements, and unconventional escalation formulas are frequently misread.
• Judgment about comparables. Which comparable sales are genuinely comparable is an opinion, not a lookup.
• Assumption defensibility. A system will happily project rent growth it cannot support. Stating and justifying assumptions remains yours.
• Local market nuance. The reason a submarket is soft, or why a corner performs better than the block, does not appear in the data.
• Tone for the specific buyer. Institutional and private capital audiences read differently.

How to evaluate this honestly

1. Bring your worst source document. A clean spreadsheet proves nothing.
2. Check extraction line by line on one rent roll and count the errors.
3. Time the full path from upload to reviewable draft, then compare to your baseline.
4. Run a revision. Change the price and regenerate everything. This is the step that separates real workflow tooling from a template.
5. Read the generated narrative critically. Is it a usable draft or a rewrite?
6. Ask how approval is enforced, not whether review is possible.

If the pipeline is not meaningfully faster on a difficult listing, you have learned that in an afternoon.

For the broader tooling picture, Offering memorandum software covers the document workflow and AI-powered commercial real estate marketing software covers extending the same record to the full campaign.

The most direct test is to run one. Try with one listing using real source documents, or book a demo and ask to see extraction on a document nobody has prepared in advance. More detail on Offering memorandum software.