Asset-Level Climate Risk: AI’s New Underwriting Edge

Swiss Re estimates 2025 global insured natural catastrophe losses reached $107 billion, with total economic losses at $220 billion. That gap isn’t just a headline—it’s the business case for changing how risk gets priced, financed, and insured.

Apollo’s latest move—expanding from “top-down” climate assessments into bottom-up, asset-level risk reviews—is a signal that the market’s center of gravity has shifted. If a private credit manager believes collateral values can swing sharply after a wildfire season or a flood event, insurers should assume the same thing: climate volatility is now a day-one underwriting variable, not a long-range scenario exercise.

This post is part of our AI in Insurance series, and I’m going to take a clear stance: asset-level climate risk analysis is becoming table stakes—and AI is the only realistic way to do it at scale. Not because it’s trendy, but because manual workflows can’t keep up with the volume, frequency, and complexity of climate-driven loss drivers.

Why “asset-level” is quickly replacing “portfolio-level” climate talk

Asset-level reviews win because they connect climate hazards to real financial outcomes: loss frequency, severity, business interruption, repair costs, insurance availability, and ultimately asset valuation.

Portfolio-level analysis (the old default) typically answers questions like: “How exposed are we to coastal wind across the book?” That’s useful for broad strategy, but it often fails at the moment underwriting decisions are made.

Asset-level analysis answers underwriting’s real questions:

• What’s the flood depth risk for this property’s exact coordinates?
• Will wildfire smoke increase business interruption risk for this facility’s operations?
• If insurance premiums double, does the deal still pencil?
• If coverage becomes unavailable, does the lender have a covenant problem?

Apollo’s framing is blunt and accurate: climate disruptions hit operating costs, supply chains, and insurance markets. That’s exactly the trio insurers live and die by.

The practical insurance impact: pricing error gets expensive fast

When climate risk is mispriced, you don’t just lose margin—you lose the option to participate. The sequence usually looks like this:

1. Underpriced risk (because the model assumes historical patterns hold)
2. Loss experience deteriorates (frequency and/or severity)
3. Rate filings spike / capacity pulls back
4. Insureds shop aggressively or self-insure
5. Market share shifts to whoever can price accurately

This is why asset-level modeling isn’t a “sustainability” project. It’s a competitiveness project.

What Apollo’s move signals for insurers (and why it matters now)

Apollo isn’t an insurer, but it’s operating like one in a critical sense: it’s trying to protect downside by improving risk selection and valuation confidence.

They’re expanding evaluations across:

• Acute hazards (floods, storms, wildfires)
• Chronic hazards (heat, drought, sea-level rise)
• Transition risks (regulatory exposure, energy sourcing, water constraints)

They’re also applying it to places where insurers already feel pain:

• Mortgage portfolios via loan-level mapping
• Hard assets where collateral value and cash-flow durability can break quickly
• Infrastructure like data centers—high capex, long holding periods, sensitive to power and water constraints

Here’s the part insurers should pay attention to: Apollo says these assessments are being integrated into every deal, across all asset classes.

That’s the new expectation in capital markets: “Show me the asset-level climate view, or I assume you don’t have one.” Insurance buyers and regulators are drifting in the same direction.

Data centers are the canary in the coal mine

Apollo specifically calls out data centers (a huge 2025–2026 investment theme thanks to AI compute demand). They focus on:

• Power costs (often the largest operating expense)
• Energy efficiency and power sourcing
• Water sourcing, use, and recycling

If you insure data centers—property, BI, equipment breakdown, cyber layering—you’ve likely seen how quickly underwriting questions expand:

• Can the facility maintain uptime during heat waves?
• Is the grid stable under peak load?
• Is there water scarcity risk that changes operating costs or triggers restrictions?

This is a perfect example of why asset-level climate risk reviews are spreading: the risk isn’t theoretical, and the financial sensitivity is high.

How AI makes asset-level climate underwriting scalable (and useful)

The value of AI in insurance isn’t “automation for automation’s sake.” It’s turning climate risk into something underwriters can actually use during quoting—fast, consistently, and with clear assumptions.

Asset-level climate assessment becomes scalable when AI helps with four jobs.

1) Data fusion: turning messy inputs into decision-ready features

Climate risk data comes in incompatible formats and resolutions: hazard maps, satellite imagery, historical claims, sensor data, building attributes, wildfire fuel models, elevation, proximity to water, and more.

AI helps by:

• Normalizing geospatial inputs into consistent property features
• Enriching missing building attributes (roof type, defensible space, construction class proxies)
• Detecting outliers and data quality issues before they hit the pricing model

Underwriting benefit: fewer “unknowns” that lead to conservative pricing or referral friction.

2) Hazard-to-loss translation: bridging science and insurance outcomes

A hazard score isn’t a loss cost. Underwriters need a defensible mapping from hazard conditions to:

• expected frequency and severity
• tail risk (PML / TVaR-style thinking)
• downtime risk (business interruption)
• secondary uncertainty (inflation, labor shortages, demand surge)

Machine learning can support this translation by learning patterns from:

• prior event loss experience (where credible)
• claims text and adjuster notes (unstructured signals)
• repair cost dynamics by region and peril

My view: models that stop at “property is high risk” aren’t good enough anymore. The winners quantify how and how much.

3) Underwriter workflow: getting answers in minutes, not weeks

Apollo is expanding asset-level diligence before closing deals. Insurers need similar speed before binding coverage.

AI-driven underwriting tools can generate:

• asset-level hazard summaries with confidence scores
• “drivers of risk” explanations (why the model thinks risk is high)
• mitigation suggestions tied to pricing credits (roof upgrades, flood barriers, defensible space)

Underwriting benefit: consistent triage—what can be straight-through processed vs. what needs specialist review.

4) Dynamic repricing: reacting to the new baseline of catastrophe losses

Swiss Re’s quote lands because it matches what many carriers have experienced: catastrophe losses are becoming baseline, not anomaly.

AI supports faster repricing and portfolio steering by:

• detecting drift (when observed losses deviate from expected)
• updating peril weights and vulnerability assumptions
• stress testing the in-force book against emerging hazard patterns

Pricing benefit: fewer lagging indicators; quicker course correction.

What “good” looks like: an asset-level climate risk playbook for insurers

Asset-level risk reviews can sound overwhelming. They don’t need to be. The best programs I’ve seen focus on a small set of repeatable, auditable steps.

Step 1: Start with decisions, not data

Pick 2–3 underwriting decisions you want to improve immediately, such as:

• eligibility and referral rules for wildfire zones
• flood deductibles and sublimits by elevation and drainage context
• BI assumptions tied to heat and smoke disruption

If you can’t name the decision, the model won’t get adopted.

Step 2: Build a minimum viable “hazard-to-coverage” mapping

Create explicit rules or model components that connect hazards to coverage terms:

• higher flood depth probability → deductible structure and sublimits
• wildfire ember exposure → roof/vent requirements and defensible space credits
• chronic heat risk → equipment breakdown considerations and BI waiting periods

This is also where explainability matters. Underwriters will trust models that show their work.

Step 3: Use AI to triage, not to replace expertise

The goal isn’t removing humans from underwriting. It’s using AI to:

• screen and score submissions
• flag missing or suspicious exposure data
• route complex risks to specialists

That’s how you scale without eroding judgment.

Step 4: Close the loop with claims and loss control

Asset-level underwriting improves fastest when claims outcomes feed back into the model.

Practical feedback loops include:

• claims severity by building attribute (roof age, construction type)
• mitigation effectiveness (did the flood barrier reduce damage?)
• repair cycle times and demand surge impacts after events

If your underwriting model never “learns” from claims, it will fall behind.

Common questions insurers ask (and the straight answers)

“Isn’t climate modeling too uncertain to use for underwriting?”

It’s uncertain, but ignoring it is a bigger error. The practical approach is to model uncertainty directly—ranges, confidence scores, scenario stress tests—and use it to set terms and appetite.

“Do we need perfect property data for asset-level reviews?”

No. You need good-enough inputs plus disciplined handling of missingness. AI-based enrichment can fill gaps, and underwriters can apply conservative rules where confidence is low.

“Will regulators expect this level of sophistication?”

Regulatory expectations vary, but the market pressure is already here: investors, reinsurers, and large insureds increasingly expect transparent climate risk assessment.

Where this is heading in 2026: underwriting that looks more like investing

Apollo’s shift is a preview of where insurance underwriting is going: more granular, more data-driven, and more explicitly tied to valuation and cash-flow durability.

The underwriting organizations that win in 2026 won’t be the ones with the most dashboards. They’ll be the ones who can answer, quickly and consistently:

“What is the asset-level climate risk, how does it change loss cost, and what terms should we write because of it?”

If your team is evaluating AI in insurance for underwriting and risk pricing, asset-level climate risk is a great place to focus because it delivers visible business outcomes: better selection, clearer pricing rationale, and fewer surprise accumulations.

If you were building a climate underwriting stack from scratch next quarter, which would you prioritize first: flood depth intelligence, wildfire defensible-space scoring, or heat-driven business interruption modeling?