In late 2023, a mid-market commercial real estate fund underwriting a 15,000-square-foot office-to-mixed-use conversion in suburban Atlanta ran the numbers on a deal that looked solid. The spreadsheet was clean. Debt service coverage ratio (DSCR) was 1.34x, well above lender minimums. Operating expense estimates matched comparable properties within 5%. Lease-up timeline was conservative—18 months to 85% occupancy. Comparable sales in the neighborhood had averaged $95/sf over the past two years.

The math checked out. The deal closed in Q1 2024. Eighteen months later, the project was on track to underperform pro forma by $2.1M annually, and the fund was negotiating with lenders about modified terms.

What went right with the math, and what went wrong with the analysis, is a teachable lesson about the gap between accurate computation and sound judgment.

The underwriting team's arithmetic was solid. They:

This is how you underwrite real estate. Every number is defensible. The model rolls forward 10 years without a logical gap.

The model was right. The analysis missed everything.

1. Comparable sales were cherry-picked (but not deliberately). The $95/sf comp set came from three recent sales in the same neighborhood submarket. All three had been repositioned by their previous owners—one had been vacant for 14 months before sale, one had been rented to a single tenant at $42/sf and was re-leased to tenants at $68-78/sf post-acquisition, and one had been owner-occupied office with immediate conversion to mixed-use. None were truly comparable in terms of current market conditions. The fund's broker had done the comparable search, selected the most recent arm's-length transactions, and submitted them as support. No one asked whether the properties had been distressed sales, tenant-replacements, or use changes. The $95/sf wasn't invented; it was just not representative of what a stabilized asset in that market was actually trading for. Actual 2024 comparable sales showed the true market at $78-82/sf.

2. The lease-up timeline assumed a neighborhood that wasn't changing. The suburb was in the path of a major office-to-residential gentrification wave driven by Atlanta's accelerating tech hiring. The underwriting team modeled 18 months to 85% occupancy based on historical office leasing velocity in that zip code (which was real, and averaged 20-22 months). But they didn't cross-check whether tenant demand profiles were shifting. By mid-2024, Class A and high-end Class B office space near tech corridors was leasing fast; Class B space in secondary locations was moving slower. The fund's asset was right in the crosshairs of this shift, but the timeline didn't reflect it. Actual lease-up took 28 months to 78% occupancy.

3. The operating expense benchmark didn't account for the property's specific cost drivers. The two comparable office buildings they benchmarked against were both Class B office, single-use, with standard HVAC and parking. The mixed-use comparable was newer construction (2018) with modern systems. The Atlanta property being modeled was a 1998 office conversion with original HVAC equipment, single parking garage (no overflow), and façade issues that would require attention in year 3. Operating expense comps don't flag obsolescence risk; they just average line items. The actual building's year-1 operating costs were $22/sf, and maintenance reserves ballooned to 1.8% by year 2.

4. Expense forecasts didn't differentiate between controllable and market-driven costs. Property taxes, insurance, and utilities are not under management control, and they were rising in Atlanta at 3-4% annually during 2024. The model used 2.5% rent growth and 2.0% expense growth—a compression dynamic that favors the owner if it holds. It didn't. By year 2, the property's insurance premium had risen 18%, and the city's property tax assessment was up 5%. Rent growth lagged at 1.8% annually.

5. The capital plan omitted a known future cost. The building's parking garage had a structural assessment flagged for 2025-2026 work (common in Atlanta's humid climate and aging infrastructure). The estimate was $800K-1.2M. The underwriting included a 1% capex reserve but didn't explicitly line-item this known requirement. When year 2 arrived, the capex reserve ($150K/year) fell $650K short, and the fund had to reprogram cash flow.

None of these errors were math errors. Arithmetic was right. Analysis was incomplete.

Real estate underwriting models are built for closure, not for capturing uncertainty. A spreadsheet shows a single pro forma path: a loan amount, an interest rate, rent levels, operating expenses, capital needs, and a resulting IRR. The model is built to answer yes or no to a single question: "Does this debt service?"

But the model is a path, not a range. And markets are ranges. The comps you find are a sample, not a census. The expenses of comparable properties are data points from specific moments, not futures. The lease-up timeline is based on historical velocity, which may not predict new conditions.

Teams typically acknowledge this in a memo ("conservative assumptions," "moderate rent growth," "industry-standard reserves"), but the memo lives outside the model. The model is deterministic. The reader's memory of the caveat fades. By the time the asset performs, the caveat is forgotten, and the model is treated as prediction.

The fund's CFO, reviewing the miss 18 months post-closing, commissioned a post-mortem. The findings:

This is not a call for more complex models. It's a call for audit trails and explicit uncertainty capture inside the model. A good underwriting spreadsheet should be able to answer: "What comps drove this price assumption? What if I use different comps? What expenses are controllable, and which are market-driven? What capital needs are known but not yet due?"

Real estate math can be right and the analysis still miss. The gap isn't usually in the formula. It's in what goes into the formula.

Sources

90-Day Prediction (December 30, 2026)

By year-end 2026, at least one institutional real estate investor (REIT, private equity fund, or pension board) will publish a case study or quarterly report explicitly flagging "comp selection bias" or "backward-looking operating expense assumptions" as a material driver of recent underperformance. Verification: search for published investor report (10-K, investor letter, or audit memo) citing these terms in connection with a specific property acquisition decision in 2023-2024.

Call-to-Action

If you're reviewing real estate analysis—whether you're an investor, a lender, or a fund manager—the model's logic is only as good as the data underneath. Cokas.io builds verification layers into real estate underwriting: it surfaces the comps, the benchmarks, and the assumptions behind your numbers, so you can see what's defensible and what's not before you close.