Let the AI build the numbers and the draft on a traced underwriting model, and keep the recommendation with the deal team: a memo is only as good as the model under it, so start from the rent roll, T-12 and offering memorandum, not a blank chat prompt. A general assistant is fine for polishing prose; for the figures, use software where every number opens to its source, such as Titleman, which returns a live Excel model, a memo in PDF and a deck.
Updated 19 October 2026. Published by Titleman, which makes one of the tools compared.
Writing the memo was never the bottleneck. Defending it is. A committee goes to the assumptions and asks where the NOI came from, why that cap rate, which comparables support the exit. The memo's value is in how quickly those questions get answered.
The question a memo has to survive
"Where did this number come from?" For every figure there is a specific answer: a cell in the rent roll, a line in the T-12, a named comparable sale, a field in the property record, or an assumption somebody made on purpose. A tool that can show that answer changes the meeting. A tool that cannot has moved the work rather than removed it.
What goes in the memo, and who writes each part
- Deal summary: property, location, asking terms.
- Property record: owner of record, sales history, recorded debt, permits.
- Rent roll and T-12 findings, with the gaps between them.
- Comparables and a value opinion, each with a source.
- The model: cash flow, returns, sensitivities and the assumptions behind them.
- Risks, and what could not be found.
- The recommendation.
Items one to six are research and arithmetic, and AI can draft them. Item seven is judgment, and the team writes it.
Five tests before you choose software
- Open a number. Take the NOI in the draft and ask where it came from. You want the document, the page and the line.
- Break an input. Change an assumption. Does every dependent figure move, or does a paragraph now contradict the model?
- Starve it. Give it a property with thin comparables. The right behavior is a stated range and a reason it is wide.
- Check the comparables. Are they named, closed sales? Is one of them the subject property under another address?
- Read the risks. A useful section names what would have to be true for this deal to fail, not generic market risk.
The expensive error is the plausible one: a market rent carried over from a neighboring submarket, or an expense ratio right for the asset class and wrong for this building. It survives review because it looks right, unless the figure carries its source.
Options compared by tool category
| Option | Best for | What it does | Where it stops | Pricing model |
|---|---|---|---|---|
| Analyst from scratch | Teams with time and a house style | Writes and models by hand | Days per memo when the pipeline is full | Salary time |
| Word or slide template | Keeping memos in one format | Gives structure | Does not find or check a figure | Staff time |
| ChatGPT-style general assistant | Polishing prose you already wrote | Drafts text from prompts | Figures not traced; can state numbers that are not in the files | Monthly seat subscription |
| Lease-by-lease cash flow software | Institutional deals that need lease-level modeling | Projects each lease's cash flow | Does not write the memo or gather evidence | License, usually per seat |
| Titleman | A memo built on a traced model | Underwriting, comparables and value opinion; live Excel, PDF memo, deck | No recommendation or final price; a person reviews | Credits per run; ask on a demo |
Where another option fits better: for a long narrative in a fixed house style, your template and an analyst still do that part best; where the deal needs lease-by-lease modeling, use dedicated cash flow software.
Where Titleman sits
Titleman builds the memo from the documents and the property record, not from a prompt. Every input shows its source and date, every result shows its formula, and a figure that could not be found is named as missing instead of filled in. The worked example on the lenders page is a 16-page memo with a 12-sheet Excel model. The underwriting took about 58 minutes at the median on production runs in the 30 days to 18 October 2026. Runs are not deterministic, so a repeat can differ: one more reason a named person signs.
It does not replace the committee's judgment, lease-by-lease modeling where the deal requires it, or a licensed appraisal where a lender or a standard requires one.
FAQ
Can AI write an investment committee memo?
It can draft the research and numbers sections on a traced underwriting model. The writing is the easy part; a committee interrogates numbers, so each figure has to trace back to a rent roll cell, a T-12 line or a named comparable. The recommendation is written by the deal team.
What should AI IC memo software be evaluated on?
Auditability first: can you open any figure and see the document and the cell or line it came from. Then behavior under thin evidence: does it state a range and say why, or produce a confident number. Then whether a changed assumption flows through the model consistently.
Does Titleman recommend whether to invest?
No. The text gives no verdict and no final price. Numbers and methods sit in the Excel model, and the decision stays with the team.
How long does the underwriting behind the memo take?
About 58 minutes at the median on Titleman production runs in the 30 days to 18 October 2026. A median, not a promise.
Is an AI-generated memo safe to submit?
Only as a draft that a named person reviews and signs. The costly failure is a plausible figure with no source behind it. Tools that mark a missing figure as missing make that review possible.
For the wider picture see AI real estate underwriting software. Related: an AI analyst that reads a rent roll and T-12, analyzing an offering memorandum with AI and pro forma versus in-place NOI. The product: real estate underwriting AI. To see a memo draft on your own deal, request a demo.