FAQ
Can I use ChatGPT to underwrite CRE deals?
Yes, but use it as an assistant, not the final system of record. ChatGPT can help read files, explain variances, draft memo language, and research markets. The final facts, assumptions, and calculations still need sources, tie-outs, and review.
What are ChatGPT’s newest abilities that matter for CRE underwriting?
The most relevant abilities are deep research, agentic web and task workflows, spreadsheet support, file analysis, and stronger reasoning around messy information. OpenAI says deep research can find, analyze, and synthesize hundreds of online sources, while ChatGPT agent combines web interaction with analysis and task execution.
Why shouldn’t ChatGPT fully underwrite a deal pipeline by itself?
Because scaled underwriting is not just writing a clean answer. It is preserving every rent, expense, lease clause, comp, debt assumption, reviewer override, and formula that moves the investment decision. OpenAI platform safety guidance still treats evaluation, monitoring, and safeguards as part of reliable AI use.
Can ChatGPT use Excel, deep research, or agents for CRE underwriting?
Yes, and those are the features CRE teams should watch closely. AI tool workflows and spreadsheets are useful together, but outputs still need review before anyone relies on formulas, assumptions, or analysis.
Where does ChatGPT make mistakes in financial models?
The risk is usually not that every answer is bad. The risk is that most of the work looks right, while a small number of cells, rows, assumptions, or source claims are wrong. In CRE, a small error can change NOI, DSCR, debt yield, IRR, exit cap sensitivity, or the story in the IC memo.
What checks should sit around ChatGPT before I trust the output?
Use source links, row-level citations, deterministic formulas, variance checks, exception queues, confidence states, reviewer approvals, and audit logs. ChatGPT can speed up the work, but the underwriting workflow should prove where the numbers came from and who approved them.
ChatGPT can research, analyze files, work with spreadsheets, and run agentic tasks. CRE underwriting still needs source trails, deterministic math, reviewer approvals, and human judgment before it scales.
ChatGPT has become much more than a chat box. For a CRE team, the newest useful abilities are obvious: deep research for market context, agentic workflows for web tasks, spreadsheet support for model work, file analysis for OMs and diligence material, and clearer explanations for investment memos.
That makes ChatGPT more useful for underwriting. It also makes the controls more important. The faster a tool can move from source material into a spreadsheet or memo, the more dangerous it becomes when the output is clean but not reviewable.
The new ChatGPT abilities CRE teams should care about
OpenAI says deep research can find, analyze, and synthesize hundreds of online sources into a cited report. For CRE, that maps to market context, comp research, tenant background, submarket risk, and competitive supply checks. It is research leverage, not underwriting ownership.
OpenAI’s platform tools documentation shows how models can use tools for multi-step work. That matters because CRE teams live in models, trackers, lender templates, and investment committee materials.
Spreadsheet and tool workflows are directly relevant to underwriting, but CRE teams still need deterministic formulas, tie-outs, and review before relying on model outputs.
Real users are already naming the boundary
“Whenever I try to have it analyze data however, it either goes completely off the rails or it reduces the dataset without telling me. So when it comes to number crunching, nothing beats ye olde spreadsheets.”
This is the ChatGPT underwriting boundary: useful for communication and task support, risky when it quietly changes or reduces the dataset behind a financial decision.
“1% wrong doesn’t fly in IB.”
CRE models have the same tolerance problem. If one assumption, formula, or lease term is wrong, the output can still look professional while the decision is wrong.
Where ChatGPT fits in CRE underwriting
| ChatGPT ability | Best CRE use | Control before it scales |
|---|---|---|
| Deep research | Market context, tenant research, competitive supply, policy or local risk | Cited source list, date stamps, analyst review, no unsupported comp claims |
| Spreadsheet support | Formula help, cleanup, scenario setup, model explanation | Deterministic tie-outs, protected formulas, cell-level review, version history |
| Agentic web tasks | Gather public data, navigate portals, update trackers | Approval gates, audit logs, credential controls, exception queues |
| File analysis | OM summaries, T-12 labels, rent-roll extraction, lease clause search | Page and row citations, confidence states, reviewer approval |
| Memo drafting | IC memo structure, risk sections, variance explanations | Human-owned assumptions and final investment judgment |
Why teams use Cactus for the ChatGPT gap
Teams use Cactus when they want AI speed without turning underwriting into a chat transcript. Cactus is built around the control problem in this article: documents, extracted facts, assumptions, market context, reviewer notes, and outputs stay connected to the deal.
That matters because ChatGPT is strongest as a support layer for research, summaries, explanations, and drafting. Cactus is where the work becomes reviewable. It keeps citations, confidence states, deterministic checks, and approvals attached before an answer becomes an IC memo or model output.
| Cactus layer | What it protects |
|---|---|
| Document Extraction | Keeps rent roll, T-12, OM, and lease facts connected to page or row references |
| Financial Analysis | Maps reviewed facts into model assumptions and Excel-ready outputs instead of trusting chat math |
| Market Intelligence | Attaches comp and market context with source links, timestamps, and analyst review |
| Proprietary Memory | Reuses approved assumptions, benchmarks, and prior decisions across future deals |
Do not let ChatGPT own the model
The right workflow is not ChatGPT versus analysts. It is ChatGPT plus a CRE-native control layer. Let the model help gather, structure, and explain information. Let deterministic logic calculate NOI, DSCR, debt yield, IRR, exit cap sensitivity, and scenario outputs. Let humans approve assumptions before they move downstream.
- Ask ChatGPT to extract facts, but require links to source pages, rows, clauses, or comps.
- Ask ChatGPT to draft a market narrative, but require the market source and date beside every claim.
- Ask ChatGPT to explain model outputs, but keep formulas in deterministic spreadsheet or code logic.
- Ask ChatGPT to help update a tracker, but require approval before any downstream side effect.
- Ask ChatGPT to summarize risk, but keep the investment decision owned by the team.
How Cactus turns this into defensible underwriting
A comparison only matters if it predicts which workflow can stand up in diligence. Cactus is built for teams that still need ARGUS-level analysis, but want source-backed extraction, premium third-party market intelligence, public records, review controls, and reusable firm memory around every assumption.
- Extract relevant facts from OMs, rent rolls, T-12s, leases, PDFs, spreadsheets, and customer templates.
- Check rents, expenses, growth targets, cap rates, sales comps, and site context against market intelligence from premium data providers, public records, and firm history.
- Surface conflicts, confidence states, reviewer comments, and assumption overrides before the model becomes the memo.
- Populate Excel or Cactus models, then store approved facts, templates, comps, and decisions in Proprietary Memory for the next deal.
The point is not to make the model less sophisticated. The point is to make the source, market check, assumption owner, review state, and output path visible before the number reaches a partner, lender, client, or investment committee.
Defensible underwriting
Defend every number before it reaches IC.
Cactus gives CRE teams ARGUS-grade underwriting intelligence with document extraction, market checks, source trails, reviewable assumptions, Excel-ready outputs, and Proprietary Memory around the workflow.
- Rent rolls, T-12s, OMs, comps, and assumptions live in separate files.
- Market evidence gets copied into the model without a durable source trail.
- Reviewer decisions disappear after the memo, email thread, or spreadsheet version changes.
- Extract deal facts from OMs, rent rolls, T-12s, leases, PDFs, spreadsheets, and customer templates.
- Check rents, expenses, growth targets, sales comps, and other assumptions against market intelligence.
- Populate Excel or Cactus models and preserve approved logic as Proprietary Memory.
Read next