FAQ
Can I use Claude to underwrite CRE deals?
Yes, but Claude should assist the workflow rather than own the final underwriting. Use it for document review, extraction, lease clause search, research, coding support, and memo drafting. Keep final assumptions, formulas, and investment judgment under source-backed review.
What are Claude’s newest abilities that matter for CRE underwriting?
The most relevant abilities are stronger agents, computer use, reasoning and math gains, Claude for Excel in financial workflows, Claude Code, extended thinking, and tool use. Anthropic says Claude Sonnet 4.5 is strong at complex agents, computer use, reasoning, and math, and its financial-services page says Claude for Excel helps analysts build and update coverage models.
Why shouldn’t Claude fully underwrite a deal pipeline by itself?
Because a deal pipeline needs more than a strong answer. It needs every important extracted fact, assumption, formula, comp, and reviewer override tied to a source and approval state. Anthropic’s own hallucination guidance focuses on reducing hallucinations, not eliminating them.
Is Claude in Excel safe for CRE financial models?
Claude can be useful around Excel, especially for updating coverage models, reviewing formulas, and pressure-testing work. But Excel is also where a single wrong cell can change a decision. Treat Claude as a reviewer and accelerator, not the final calculator.
Should Claude calculate NOI, DSCR, IRR, or debt yield?
Not as the final authority. Claude can extract facts and explain model logic, but deterministic spreadsheet or code logic should calculate NOI, DSCR, IRR, debt yield, exit cap sensitivity, and scenario results. One CRE user using Claude Code described the safer pattern as LLM extraction with deterministic logic on top.
What checks should sit around Claude before I trust the output?
Use source citations, page and row references, deterministic calculations, tool-call logs, approval gates, exception queues, reviewer states, and audit history. Claude can help move faster, but the underwriting workflow has to prove where each number came from.
Claude’s newer agent, coding, computer-use, Excel, and reasoning abilities make it useful for CRE workflows. They do not remove the need for citations, deterministic math, review states, and human ownership.
Claude deserves its own underwriting article because the product direction is different from ChatGPT. Claude’s most relevant abilities for CRE are long-context reasoning, agents, tool use, Claude Code, computer-use style workflows, and financial-services use cases like Claude for Excel.
That makes Claude a strong candidate for diligence automation and underwriting support. It can help inspect documents, build pipelines, reason over messy files, draft memos, write code around workflows, and pressure-test analysis. But the same rule still holds: the stronger the assistant, the more important the control layer becomes.
The new Claude abilities CRE teams should care about
Anthropic says Claude Sonnet 4.5 is strong at complex agents, computer use, reasoning, and math. For CRE, that maps to multi-step workflows: reading documents, comparing assumptions, using tools, checking files, and helping analysts reason through a deal package.
Anthropic’s financial-services page says Claude for Excel helps analysts build and update coverage models, separate signal from noise, and pressure-test their work. That is close to the CRE workflow pattern: live in models, update assumptions, and defend the work in committee.
Claude Code docs describe an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with development tools. That matters for CRE teams building repeatable pipelines around rent rolls, T-12s, leases, comps, and investment memo workflows.
Claude’s tool-use docs say Claude can call external tools and APIs, and the extended-thinking docs describe enhanced reasoning capabilities for complex tasks. That is powerful, but it also means teams need tool logs, approval gates, and deterministic checks around every side effect.
Real users are already naming the boundary
“Coming at this from an adjacent angle, I've been using Claude Code to build pipelines around the post-close side of CRE rather than underwriting. The pattern that's helped most: don't ask the LLM to do math. Have it extract structured facts, then run deterministic logic on top. Reliability + auditability both go way up, and the output stays defensible.”
This is the Claude workflow boundary: use Claude to extract, structure, and orchestrate, then let deterministic logic handle the math and auditability.
“Excel users work with financial models, payroll, forecasts - domains where one wrong cell out of thousands kills trust completely.”
CRE underwriting is exactly this kind of trust problem. One bad cell or missing assumption can change the investment recommendation.
Where Claude fits in CRE underwriting
| Claude ability | Best CRE use | Control before it scales |
|---|---|---|
| Long-context reasoning | Review OMs, leases, amendments, diligence notes, and market docs together | Source citations, quote/page references, confidence states |
| Claude for Excel and model support | Build, update, and pressure-test model workflows | Protected formulas, deterministic tie-outs, reviewer approval |
| Claude Code | Build extraction, comp, QA, and memo-generation pipelines | Version control, tests, audit logs, no silent production side effects |
| Tool use and APIs | Pull data, call internal tools, enrich records, route tasks | Least-privilege access, tool logs, approval gates |
| Extended thinking | Analyze messy deal context and tradeoffs | Transparent assumptions, reviewer notes, decision ownership |
Why teams use Cactus for the Claude gap
Teams use Cactus when they want Claude-level acceleration without letting an AI agent become the underwriting system of record. Claude can read context, use tools, write code, and pressure-test work. Cactus gives that work a CRE operating layer with sources, reviewer queues, model checks, and approval history.
That is the difference between using Claude to help and letting Claude own the deal. In Cactus, extracted facts, comp logic, market evidence, formulas, and memo outputs stay tied together so a team can see what changed, why it changed, and who approved it.
| Cactus layer | What it protects |
|---|---|
| Document Extraction | Keeps diligence facts tied to the original OM, rent roll, T-12, lease, or scan |
| Agent Builder | Lets teams automate recurring workflows with approvals and side-effect controls |
| Financial Analysis | Runs model logic from reviewed assumptions instead of relying on conversational math |
| Proprietary Memory | Carries forward approved comps, benchmarks, rules, and decisions from prior deals |
Do not let Claude own the model
Claude can be extremely useful around a CRE model. It can read more context, operate across tools, write code, reason through diligence, and help analysts prepare better work. But underwriting is still a system-of-record problem, not just a reasoning problem.
- Let Claude extract facts, but require source pages, rows, clauses, and comp links.
- Let Claude build or review pipeline logic, but require tests and version control.
- Let Claude help with Excel and model analysis, but protect formulas and run deterministic tie-outs.
- Let Claude call tools, but log every tool call and require approval for side effects.
- Let Claude draft the memo, but keep final assumptions and investment judgment owned by the team.
Do the Claude workflow safely
Use Claude for acceleration, Cactus for underwriting control.
Claude can reason across files, use tools, and help build workflows. CRE teams still need source-backed facts, deterministic math, approval states, and audit history that survive outside a prompt.
- Claude can extract and reason, but the deal still needs durable source links and review states.
- Tool use and agent work need permissioning, logs, and approval gates before they affect a pipeline.
- Financial outputs need deterministic calculations, not only a plausible model explanation.
- Cactus keeps Claude-style workflow speed inside a CRE-specific review system.
- Financial Analysis ties assumptions, sources, and formulas together before outputs move downstream.
- Agent Builder and Proprietary Memory help teams reuse approved workflows without losing control.
Read next