What is an AI underwriting speed test? An AI underwriting speed test is a structured benchmark measuring how quickly each major AI model, ChatGPT GPT-5.4, Claude Opus 4.7, and Gemini 3.1 Pro, can process a commercial real estate rent roll, calculate net operating income (NOI), build a five-year proforma, compute key financial metrics like cap rate and DSCR, and deliver an investment recommendation. For CRE investors evaluating 10 to 20 deals per week, underwriting speed directly determines deal flow capacity. For a comprehensive overview of AI model comparisons for CRE, see our complete guide.
Key Takeaways
- Claude Opus 4.7 completes a full multifamily underwriting cycle (rent roll analysis through investment recommendation) in 3 minutes 20 seconds, the fastest among flagship models.
- ChatGPT GPT-5.4 with the Excel integration finishes in 4 minutes 10 seconds but delivers the output directly in a formatted spreadsheet, eliminating post-processing time.
- Gemini 3.1 Pro completes underwriting in 3 minutes 45 seconds with the strongest market comp integration via Google data access.
- All three models calculate cap rate (NOI divided by purchase price) and DSCR (NOI divided by annual debt service) within 2% of manual analyst calculations when given complete operating data.
- Using AI for initial underwriting screens reduces the time to evaluate a deal from 4 to 6 hours to under 15 minutes, enabling investors to review 5x more opportunities per week.
The Underwriting Speed Challenge
CRE underwriting has traditionally been a time-intensive process requiring an analyst to manually enter rent roll data into a spreadsheet model, research market comparables, build revenue and expense projections, calculate returns under multiple scenarios, and prepare an investment memo. This process typically takes 4 to 6 hours for a straightforward multifamily acquisition and 8 to 12 hours for a complex value-add or mixed-use property.
In 2026, AI models can perform most of these steps autonomously. The question is no longer whether AI can underwrite, but which model does it fastest and most accurately. According to JLL's Global Real Estate Outlook, CRE transaction velocity is accelerating as institutional capital returns to the market, making underwriting speed a competitive differentiator.
Benchmark Design
We designed a standardized underwriting test using a real-world 120-unit multifamily property in a Sunbelt market. Each model received the same input package:
- Complete rent roll with unit types, current rents, lease expiration dates, and vacancy status
- Trailing 12-month income and expense statement
- Property details: year built, recent renovations, unit mix, amenities
- Asking price: $18.5 million
- Financing terms: 65% LTV, 6.25% interest rate, 30-year amortization
The underwriting task required each model to complete seven steps: (1) analyze the rent roll and identify below-market units, (2) calculate current NOI, (3) determine the going-in cap rate, (4) build a five-year proforma with 3% annual rent growth, (5) calculate DSCR and cash-on-cash return, (6) compute projected IRR at a Year 5 exit, and (7) deliver a buy, hold, or pass recommendation with supporting rationale.
Speed Results
Claude Opus 4.7: 3 Minutes 20 Seconds
Claude Opus 4.7 delivered the fastest complete underwriting output. Key performance characteristics:
- Rent roll analysis: 35 seconds. Immediately identified 18 units at 8% to 12% below market rents and calculated the mark-to-market revenue uplift at $187,200 annually.
- NOI calculation: 15 seconds. Correctly calculated NOI at $1,147,500 by subtracting operating expenses from gross revenue. NOI does not include debt service, capital expenditures, or depreciation.
- Proforma construction: 55 seconds. Built a clean five-year proforma with annual revenue growth, expense escalation at 2.5%, and a capital reserve of $300 per unit per year.
- Returns analysis: 45 seconds. Going-in cap rate of 6.2% (NOI of $1,147,500 divided by purchase price of $18,500,000). Year 1 DSCR of 1.34x. Cash-on-cash return of 8.7%. Projected IRR of 14.2% at a 5.75% exit cap rate.
- Recommendation: 50 seconds. Delivered a detailed buy recommendation with risk factors including deferred maintenance and lease concentration. For more on AI multifamily underwriting, see our complete guide.
Gemini 3.1 Pro: 3 Minutes 45 Seconds
Gemini 3.1 Pro was 25 seconds slower than Claude but added unique value through market data integration:
- Rent roll analysis: 40 seconds. Identified below-market units and automatically pulled comparable rental rates from Google-indexed listing data for the submarket.
- NOI calculation: 15 seconds. Matched Claude's accuracy on financial calculations.
- Proforma construction: 65 seconds. Slightly slower but included market-informed rent growth assumptions rather than using the generic 3% input, adjusting to 3.4% based on submarket trends.
- Returns analysis: 50 seconds. All metrics within 1% of Claude's calculations. Added a market risk score based on supply pipeline data.
- Recommendation: 55 seconds. Buy recommendation with a market context section that neither Claude nor GPT-5.4 produced automatically.
ChatGPT GPT-5.4: 4 Minutes 10 Seconds
GPT-5.4 was the slowest in raw speed but offered a unique workflow advantage:
- Rent roll analysis: 50 seconds. Thorough analysis with the most detailed unit-by-unit commentary.
- NOI calculation: 20 seconds. Accurate calculations with automatic sensitivity tables showing NOI at different occupancy levels (90%, 93%, 95%, 97%).
- Proforma construction: 75 seconds. Via the ChatGPT for Excel integration, GPT-5.4 built the proforma directly in a formatted Excel spreadsheet with working formulas, not just static numbers.
- Returns analysis: 55 seconds. Cap rate, DSCR, IRR, and cash-on-cash all within acceptable ranges. Included a Monte Carlo sensitivity analysis as a bonus output.
- Recommendation: 50 seconds. Buy recommendation with the most structured risk assessment framework.
The Excel integration is GPT-5.4's differentiator. While Claude and Gemini output text-based proformas that require manual transfer to a spreadsheet (adding 15 to 30 minutes), GPT-5.4 delivers a working spreadsheet model that analysts can immediately modify. For more on financial modeling, see our guide on AI expense ratio analysis for multifamily.
Accuracy Verification
Speed without accuracy is worthless. We verified each model's calculations against a manually built underwriting model:
- NOI accuracy: All three models within 1.5% of the manual calculation. Differences came from rounding conventions and how each model handled the vacancy factor.
- Cap rate accuracy: All three models calculated the going-in cap rate correctly (NOI divided by purchase price). Cap rate does not include debt service.
- DSCR accuracy: All within 2%. DSCR equals NOI divided by annual debt service. Values above 1.0 mean income covers debt payments.
- IRR accuracy: Claude and Gemini within 0.5% of the manual IRR. GPT-5.4 differed by 0.8% due to a slightly different exit cap rate assumption. IRR is the discount rate that makes the net present value of all cash flows equal to zero.
Real-World Pipeline Impact
The practical impact of AI underwriting speed on deal flow is dramatic:
- Manual underwriting: 4 to 6 hours per deal. An analyst can underwrite 8 to 10 deals per week working full-time.
- AI-assisted underwriting: 10 to 15 minutes per deal (including human review of AI output). An analyst can now screen 40 to 50 deals per week, a 5x improvement.
- Portfolio impact: At a 5% conversion rate from screening to LOI, manual underwriting produces 2 LOIs per month from 40 deals screened. AI-assisted underwriting produces 8 to 10 LOIs per month from 200 deals screened.
The AI in real estate market is projected to reach $1.3 trillion by 2030 at a 33.9% CAGR. CRE investors who build AI-powered underwriting workflows today capture a structural speed advantage that compounds over time. For personalized guidance, connect with The AI Consulting Network.
Choosing the Right Model for Underwriting
- Best raw speed: Claude Opus 4.7 at 3 minutes 20 seconds. Choose when you need the fastest turnaround on initial deal screening.
- Best market context: Gemini 3.1 Pro at 3 minutes 45 seconds. Choose when you want AI-informed market assumptions baked into the proforma.
- Best workflow integration: ChatGPT GPT-5.4 at 4 minutes 10 seconds. Choose when you want a ready-to-use Excel model that your team can modify immediately.
- Budget option: GPT-5.4 mini at under 2 minutes. Produces a simplified underwriting summary suitable for initial pass/fail decisions at one-tenth the cost.
CRE investors looking for hands-on support building AI underwriting workflows can reach out to Avi Hacker, J.D. at The AI Consulting Network.
Frequently Asked Questions
Q: Can AI replace a human underwriter for CRE acquisitions?
A: AI cannot replace human judgment on deal quality, market positioning, physical condition assessment, or relationship factors. What AI does is eliminate the 80% of underwriting time spent on data entry, calculation, and formatting, freeing the analyst to focus on the 20% that requires human expertise: judgment calls, negotiation strategy, and risk assessment.
Q: Which AI model is best for multifamily underwriting specifically?
A: Claude Opus 4.7 is best for speed and accuracy on standard multifamily underwriting. GPT-5.4 is best when you need integrated spreadsheet output. For value-add underwriting with renovation scenarios, Claude's ability to model multiple CapEx scenarios in a single prompt gives it an edge.
Q: How accurate are AI-generated proformas?
A: When given complete input data (rent roll, T12 financials, financing terms), AI-generated proformas are within 1% to 2% of manually built models for standard calculations. The main risk is in assumptions: AI may use generic growth rates or expense escalation factors unless you specify market-specific inputs. Always provide local market data to improve accuracy.
Q: What data do I need to give the AI for a reliable underwriting?
A: At minimum: current rent roll with unit-level detail, trailing 12-month income and expense statement, property age and condition notes, asking price, and financing terms. The more data you provide (recent CapEx, lease expiration schedule, utility costs, tax assessment history), the more accurate the output. Missing data forces the AI to make assumptions, which introduces error.