CUOs Don't Want Pixels Anymore
In 2025, the differentiator shifted from image quality to decision-grade outputs: versioned attributes, reason codes, confidence, and workflow integration.
The industry spent a decade competing on resolution, and resolution won. Imagery quality has reached a sufficiency threshold where incremental sharpness no longer produces incremental underwriting lift. The winners now aren’t the companies with better pictures. They’re the companies that turn pictures into decisions without requiring an underwriter to look at them.
Vendors are quietly pivoting from ‘showing you the roof’ to diagnosing roof condition, structural weakness, treefall radius, wildfire exposure, and secondary modifiers automatically.
In practice, that means a roof condition score becomes an eligibility rule, a vegetation flag becomes a referral trigger, and an address-confidence score becomes a straight-through-processing gate — without anyone opening an image viewer.
Why This Matters
The economics of property underwriting have changed faster than most carriers’ technology.
First Street Foundation’s 12th National Risk Assessment (Feb 2025) estimates premiums could rise ~29.4% over the next 30 years as climate risk is repriced—making mispricing harder to survive. At the same time, CAPE Analytics has cited research suggesting underwriters spend roughly 41% of their time on administrative/operational tasks, which makes ‘open imagery and interpret it’ a structural bottleneck. 2025 M&A underscored the direction: Moody’s announced an agreement to acquire CAPE, describing a combined property database delivering ‘instant, address-specific risk insights,’ and Nearmap announced plans to acquire itel to connect underwriting and claims workflows.
This isn’t a pivot from imagery to analytics it’s the moment analytics got good enough to be treated as decision infrastructure.
CAPE Analytics has cited research showing underwriters spend ~41% of their time on administrative/operational work rather than underwriting decisions. In high-volume channels, carriers routinely triage far more submissions than humans can fully review—so ‘open the image and interpret shingles’ becomes a bottleneck. When you’re triaging that volume, opening a satellite image and interpreting shingle condition isn’t a workflow, it’s a bottleneck that compounds across every submission you don’t have time to touch.
The consolidation activity this year made the strategic direction explicit.
In January 2025, Moody’s announced an agreement to acquire CAPE Analytics, describing the combination as creating a ‘sophisticated property database’ delivering ‘instant, address-specific risk insights. CAPE had raised about $75M in venture funding prior to the deal. The acquisition closed in Q1, and the combined entity is now integrating Cape’s Roof Condition Rating directly into Moody’s Intelligent Risk Platform. That’s not an imagery play, it’s peril modeling plus property condition in a single API call.
In May 2025, Nearmap announced it would acquire itel a provider of building material pricing and repair-versus-replace analysis for claims. CEO Andy Watt framed it as creating “a true end-to-end solution that meets the most critical data needs across insurance claims and underwriting.” The play is connecting underwriting insight to claims settlement, bridging workflows that have historically operated on different data.
The pattern: pure-play imagery providers are becoming insight platforms. Insight platforms are expanding into adjacent workflows. Carriers that haven’t figured out where they’re getting automated risk signals are running out of neutral options.
The Question This Raises
If property intelligence is moving from visualization to interpretation, two tensions emerge, and neither has a clean answer.
The first is accountability. When Cape Analytics delivers a Roof Condition Rating, or ZestyAI returns a Z-FIRE score, or Nearmap’s Betterview platform flags a property for deferred maintenance, those outputs increasingly flow directly into underwriting decisions. Nearmap’s October 2025 launch of the Betterview Accelerator for Guidewire PolicyCenter embeds property flags as “underwriting issues in the risk analysis tab”, no human interpretation required.
That’s efficient. It may also be a liability transfer you haven’t fully priced.
Regulators are catching up. Colorado’s SB21-169 pushes insurers to document governance around external data and predictive models to reduce unfair discrimination risk. And in California, aerial-imagery rules and proposals increasingly focus on notice, recency, and transparency when imagery contributes to adverse decisions. Cape Analytics noted in its 2025 outlook that carriers should “prioritize property intelligence solutions approved for use in 40 or more states for rating and underwriting”, practical advice that surfaces a harder question.
When an AI-derived insight leads to a declination, and the homeowner disputes the underlying data, who defends the model?
The answer in most vendor contracts is clear: carriers bear the regulatory risk. Vendors have limitation of liability clauses. The real question isn’t “who’s accountable”, it’s whether your E&O or D&O coverage contemplates AI-derived underwriting decisions, and whether you’ve confirmed that with your carrier. Most contracts treat the output as “information only,” while your workflow treats it as a declination reason — that gap is where the liability lives.
The second tension is the feedback loop. These models are trained on historical claims data. But carriers using them are now selecting risks differently, declining properties that would have been written five years ago, tiering renewals based on vegetation scores that didn’t exist in the training set. That changes the loss distribution going forward.
The models are training on a disappearing past. Classic selection bias. Actuaries worry about this; underwriting leaders should too. If your vendor can’t explain how they’re accounting for distributional shift in their model updates, you’re buying confidence you haven’t verified.
What the Other Players Are Doing
The competitive response has been predictable: everyone is building interpretation capabilities. The differentiators are shifting to model validation, workflow integration, and, increasingly, temporal intelligence.
The real advantage of insight-driven vendors isn’t that they see the roof more clearly. It’s that they model how risk changes between renewals, inspections, and loss events. Static imagery answers “what is it?” Underwriting needs “where is this headed?”
Moody’s/Cape Analytics is the most aggressive consolidation play. Post-acquisition, Cape’s property attributes integrate with Moody’s catastrophe models to produce combined risk scores, address-level property condition linked to peril exposure. The strategic risk for the market: if Moody’s/Cape becomes the dominant property intelligence layer bundled with cat models, carriers have fewer alternatives. That’s leverage that compounds over time.
Nearmap/Betterview is building the full-stack alternative. In October, Nearmap launched Roof Age Gen2, combining permit data, assessor records, climate data, and deep learning models to deliver roof age estimates “in under two seconds with 95% accuracy.” Zurich North America announced that same month it had integrated Nearmap’s AI-enhanced property insights directly into its U.S. Middle Market underwriting platform. That’s enterprise validation of the approach, and a reference account that matters.
ZestyAI is leaning into peril-specific models with regulatory approval as the moat. In February, NEXT Insurance adopted ZestyAI’s Z-PROPERTY and Z-FIRE models for commercial property underwriting. Southern Oak Insurance followed in September, deploying ZestyAI across its Florida homeowners portfolio. The company claims Z-FIRE is “adopted by over one-third of California’s insurance market.” ZestyAI’s 2025 State of Property Insurance survey found that only 40% of carriers have embedded AI into core workflows, which means they’re still early in adoption, with room to expand. And distribution is shifting: partnerships like ZestyAI’s integration into EarthDaily’s Ascend platform show “insight models” increasingly riding on someone else’s data rails.
Verisk entered the generative AI conversation in September with its Commercial GenAI Underwriting Assistant, a cloud-based solution that automates workflows, summarizes datasets, and delivers risk appetite insights. Verisk positioned it as augmentation, not replacement: a “Human-in-the-Loop” model. Their 2025 survey found 69% of insurance executives believe AI will have the most significant industry impact over the next five years. Verisk is betting that existing customer relationships and data assets let them catch up on the insight layer.
AIG reported in early 2025 that underwriting data collection and accuracy rates increased 15% after deploying generative AI platforms. That’s carrier-reported outcome data, not vendor claims, and it signals that the efficiency gains are real for organizations that commit.
What Leaders Should Do
If you’re running underwriting at a small-to-midsize carrier writing property risks, here’s the honest reality: the Guidewire integration story is great for carriers on Guidewire. If you’re running a legacy PAS or a smaller vendor’s system, the “workflow-embedded insight” future is harder to reach. That doesn’t mean the shift doesn’t apply to you, it means your path is different.
Audit your current property data stack. List every source: imagery providers, hazard data, replacement cost estimates, roof condition scores. Map which ones deliver raw data requiring human interpretation versus automated risk signals. The ratio tells you how exposed you are to the interpretation shift, and how much manual work you’re subsidizing.
Ask vendors about model validation with specifics. What claims data was the model trained on? How often is it retrained? What’s the demonstrated lift in loss ratio for carriers using the model versus not? How are they accounting for distributional shift as carrier behavior changes? If your vendor can’t answer these questions, you’re buying imagery dressed up as insight.
Clarify your governance and compliance posture now. Any analytical tool, whether AI-based risk scores or hazard predictors, should come with clear documentation of feature sources and statistical linkages, procedures for human override and audit trails, and protocols for responding to regulatory inquiries. Those are contract terms worth negotiating before you sign, not after a commissioner asks questions.
Confirm your E&O coverage. If you’re using AI-derived property scores in underwriting decisions, does your errors and omissions policy contemplate that? Have you discussed it explicitly with your carrier? This isn’t theoretical risk, it’s operational exposure that exists today.
Negotiate for outcome accountability. The vendors are confident enough in their models to make performance claims. Push them to back those claims contractually. If Z-FIRE really improves risk selection, ZestyAI should structure some portion of the arrangement around demonstrated outcomes. If they won’t, that tells you something about their confidence.
Watch vendor concentration. The Moody’s/Cape combination will likely bundle property intelligence with catastrophe modeling in ways that create pricing pressure, and switching costs, for carriers. If you’re evaluating long-term contracts with any of these vendors, build in flexibility for 2026.
The Measured Counter-Argument
To be fair: for some carriers, especially smaller books or less CAT-exposed portfolios, imagery-first workflows still deliver acceptable ROI. Not every organization needs to be on the leading edge of this transition. The operational lift of integrating new insight vendors, retraining underwriters, and managing model governance isn’t trivial.
The issue isn’t whether imagery-first tools work. It’s whether they scale as volatility increases, regulatory scrutiny tightens, and portfolio complexity grows. For carriers facing those pressures, the math changes. For those that aren’t, the urgency is lower, but the direction of travel is the same.
The Bottom Line
Imagery still matters, but as infrastructure, not as product. The quality threshold has been met; above it, sharper pixels don’t produce sharper underwriting. The winners in property intelligence are the vendors that deliver interpretable, defensible, temporally-aware risk signals embedded in workflows, not better pictures that require human judgment to decode.
If you can’t explain why a risk moved tiers without calling your vendor, that’s not automation. That’s dependency. Choose accordingly.
