What is AI seismic PML report review? AI seismic PML report review is the use of large language models such as Claude and ChatGPT to extract, normalize, and pressure-test the loss estimates inside a seismic risk assessment so a CRE deal team can act on them quickly. Probable Maximum Loss, or PML, is the legacy shorthand for the earthquake damage a building is expected to suffer, and modern reports under ASTM E2026 express that risk as a Scenario Expected Loss instead. For buyers and lenders in seismic markets, reading these reports correctly is as important as any other line of AI real estate due diligence, because a single number can trigger an insurance requirement or kill a loan.
Key Takeaways
- ASTM E2026 retired the loose term PML in favor of precise metrics: Scenario Expected Loss (SEL), Scenario Upper Loss (SUL), and Probable Loss (PL), though the market still says PML colloquially.
- The number lenders care about most is the SEL-475, the expected loss from a 475-year return period earthquake, with a result above roughly 20 percent commonly triggering earthquake insurance or mitigation.
- AI extracts the SEL, SUL, return period, building stability finding, and site hazards into a normalized summary in minutes, so the deal team is not hunting through a 40-page engineering report.
- Reports run across four review levels, from a Level 0 desktop screen to a Level 3 detailed computer model, and AI should flag which level you actually received.
- AI organizes and cross-checks the report; the licensed structural engineer who stamped it, and earthquake insurance pricing, still govern the final decision.
Why Seismic Reports Are Hard to Read and Easy to Get Wrong
Seismic reports are hard to read because the terminology changed and the market never fully caught up. ASTM E2026, the Standard Guide for Seismic Risk Assessment of Buildings, recommends discontinuing the term Probable Maximum Loss and using Scenario Expected Loss and Scenario Upper Loss instead, yet investors, brokers, and even some lenders still say PML to mean any of these. A report can quote an SEL of 12 percent and an SUL of 22 percent for the same building, and a reader who does not know the difference can anchor on the wrong figure and misprice the risk.
The companion standard, ASTM E2026 and ASTM E2557, governs how Probable Maximum Loss evaluations are reported for earthquake due-diligence, particularly on CMBS loans. The practical problem for a deal team is speed: a 40-page report arrives during a tight diligence window, the key numbers are buried in tables, and the person reading it is not a structural engineer. This is exactly the kind of dense, structured document where AI earns its place, the same way it does on a AI property condition assessment building inspection CRE report.
What the Numbers Actually Mean: SEL, SUL, and SEL-475
The core metric is the Scenario Expected Loss, the mean estimated damage to a building under a defined earthquake scenario, expressed as a percentage of replacement cost. Scenario Expected Loss represents the mean of a distribution describing damage to a population of similar buildings, while the Scenario Upper Loss is a higher-confidence figure, roughly the 84th percentile, that captures a worse-than-average outcome. A building with an SEL of 10 percent and an SUL of 18 percent is expected to lose about 10 percent of its value in the scenario quake, with a realistic bad case near 18 percent.
The scenario itself is usually the Design Basis Earthquake, defined as ground motion with a 10 percent probability of exceedance in 50 years, which is equivalent to a 475-year return period. That is why agency lenders reference the SEL-475: the Scenario Expected Loss tied to that 475-year ground motion. Fannie Mae and Freddie Mac typically require a seismic assessment when the 475-year peak ground acceleration at a site exceeds about 0.15g, which captures most of California, the Pacific Northwest, and other active zones. AI can pull the SEL, SUL, return period, and peak ground acceleration into one line and tell you immediately whether the report crosses these thresholds.
How AI Reviews a Seismic Risk Assessment
AI reviews a seismic report by turning unstructured engineering prose into a structured, decision-ready summary. A disciplined prompt asks Claude or ChatGPT to extract the SEL and SUL percentages, the return period used, the ASTM review level (0 through 3), the building stability conclusion, and any site hazards such as liquefaction, landslide, or surface fault rupture. The model then states, in plain language, whether the result is below or above the lender threshold and whether earthquake insurance is implicated. What took an analyst an hour of careful reading becomes a five-minute structured output that an engineer can verify.
The highest-value checks are the ones a rushed human skips. Ask the model to confirm which review level you actually received, because a Level 0 desktop screen is not the same diligence as a Level 3 analytical model and a lender may reject the cheaper version. Have it flag whether the report assesses building stability, the question of whether the structure keeps its vertical load-carrying capacity during and after the quake, separately from damageability. Route every flagged item to the structural engineer of record. For broader context on automating this kind of report-heavy review, see our work on Claude property condition report PCR review CRE. CRE investors who want a repeatable seismic-review prompt library often build it with The AI Consulting Network.
Turning the SEL Into a Deal Decision
The SEL becomes a deal decision through three levers: insurance, loan proceeds, and price. The widely used rule of thumb is that an SEL or SUL above roughly 20 percent pushes a lender to require earthquake insurance, demand seismic retrofitting, or reduce proceeds, while a result in the single digits usually clears without special conditions. A building reporting a 28 percent SUL is not uninvestable, but the cost of earthquake coverage, often quoted as a deductible of 5 to 15 percent of the insured value, has to be modeled into net operating income and the exit, because a buyer at sale will run the same report.
This is where AI connects the engineering to the underwriting. Once the model extracts the loss figures, it can estimate the annual earthquake-insurance premium drag, reduce projected NOI, and show the effect on your cap rate and levered returns. A 50 basis point hit to NOI from insurance changes the price you can pay. The Freddie Mac and Fannie Mae Multifamily Guide set the agency standard for when these assessments are required and how results are treated, and pulling those requirements alongside the report keeps the analysis honest. When the SEL pushes a deal toward mitigation, modeling the retrofit cost against the value preserved is the same discipline The AI Consulting Network applies across due-diligence workflows.
The Limits: What AI Should Never Decide on Seismic Risk
AI should never replace the licensed structural engineer who produces and stamps the seismic risk assessment. The model reads what the report says; it does not evaluate the building, model the ground motion, or certify the loss estimate. A language model can misread a table, miss a footnote that qualifies a number, or fail to catch that the engineer used a non-standard scenario. Every extracted figure is a draft for human verification, not a conclusion.
Two judgment calls in particular stay with people. The first is whether a borderline result, an SEL near the lender threshold, warrants a more detailed Level 3 study before bidding, which is an engineering and capital decision. The second is how to price the long-tail risk of a catastrophic but low-probability event, which is an underwriting and insurance question, not a document-extraction one. Used well, AI gives the deal team a fast, consistent read so the human experts spend their time on these calls. Investors building that workflow can reach out to Avi Hacker, J.D. at The AI Consulting Network.
Frequently Asked Questions
Q: Is PML the same as SEL in a seismic report?
A: Not exactly. PML is the older, looser term for earthquake loss, while ASTM E2026 replaced it with the Scenario Expected Loss (the mean estimate) and Scenario Upper Loss (a higher-confidence estimate). The market still says PML colloquially, so always confirm which metric and return period a report is actually quoting.
Q: What SEL level triggers earthquake insurance on a CRE loan?
A: As a common rule of thumb, an SEL or SUL above roughly 20 percent prompts lenders to require earthquake insurance, retrofitting, or reduced proceeds. The exact threshold varies by lender and program, so verify the specific requirement against the agency guide or your lender term sheet rather than the rule of thumb alone.
Q: Can AI replace a structural engineer's seismic assessment?
A: No. AI reads and organizes the report a licensed engineer produces, extracting the SEL, SUL, review level, and site hazards into a usable summary. It cannot evaluate the building, model ground motion, or certify a loss estimate, and every figure it pulls should be verified against the stamped report.