Digital Twins in Commercial Property: From Buzzword to Underwriting Input

The digital twin concept has accumulated extraordinary momentum. McKinsey estimates global digital twin investments will exceed $48 billion by 2026, with 70% of C-suite technology executives at large enterprises already exploring or investing in the technology (McKinsey, 2024). Accenture has described digital twins as "the next wave of digital change in the insurance industry." Guidewire has argued that expanding the definition of "inspection" to include digital twins could increase the inspection rate of commercial properties from 10% to 100% (Guidewire, 2023).

These are significant claims. And the underlying technology — combining satellite imagery, aerial photography, IoT sensor data, AI-driven data extraction, and real-time environmental monitoring into a continuously updated virtual model of a physical asset — is genuinely powerful.

But for commercial property underwriters, the question is not whether digital twins are technically impressive. The question is whether the data they produce changes the number on the quote. And in most current implementations, the honest answer is: not yet — at least not in the way the technology's advocates suggest.

What digital twins promise — and where the gap is

Digital twins offer a structurally different approach to this problem. Rather than relying on periodic physical site visits and static submission data, a digital twin of a commercial property aggregates multiple data layers — satellite and aerial imagery, street-level photography, IoT sensor feeds, open data sources (crime, fire service, flood, weather, environmental), AI-extracted data from historical risk reports, and real-time monitoring — into a continuously updated virtual model of the property and its risk environment.

Intelligent AI, which developed through Lloyd's Lab and now integrates with Guidewire InsuranceSuite, describes its platform as creating digital twins from up to 300 data sets, covering COPE data (Construction, Occupancy, Protection, Exposure), geocoding, satellite imagery, flood maps, construction details, fire risk, and crime analysis (Intelligent AI; Insurtech Analyst, March 2026). The proposition is a 360-degree view of risk across 100% of a portfolio, delivered at a fraction of the cost of physical surveys.

ZestyAI, operating primarily in the US residential and commercial property market, uses AI and computer vision to build a digital twin for every building across North America, encompassing over 200 billion property insights. Its models have been approved by regulators for rating and underwriting across wildfire-prone states, and the company reports that insurers using its platform enabled coverage for over 511,000 previously uninsurable properties in 2024, with a target to double that in 2025 (ZestyAI; InsurTech Digital, April 2025).

Tensorflight, acquired by Nearmap in 2024, uses satellite, aerial, and ground-level imagery to automate commercial property inspections. The company works with three of the five largest US commercial property insurers and reports reducing the cost of in-person property inspections by over 25% (Tensorflight; Duck Creek; FinTech Global).

These are real capabilities delivering real operational value. But there is a critical distinction between using digital twin data for operational efficiency — faster triage, richer submission data, automated COPE extraction — and using it as a direct underwriting input that changes how risk is priced.

The underwriting input question

For a data source to become an underwriting input — one that directly influences pricing, terms, or risk selection — it needs to meet a higher bar than operational utility. An underwriter needs to believe, with actuarial confidence, that the data changes their assessment of expected loss frequency or severity for that specific risk.

Most current digital twin implementations in commercial property deliver value at the portfolio level: identifying accumulation risk, flagging properties with characteristics correlated with higher loss, improving the completeness and speed of submission triage, and enabling virtual inspection of properties that would otherwise receive no survey at all. This is valuable. But it is not the same as an underwriter looking at a digital twin of a specific property and deciding to price it differently because of what the twin tells them.

Where digital twins do change the price

There are three areas where digital twin data is beginning to cross the threshold from operational enrichment to genuine underwriting input — where the data doesn't just describe the property more efficiently, but reveals risk that was previously invisible or unquantifiable.

1. Peril-specific, property-level risk scoring. ZestyAI's wildfire model, Z-FIRE, illustrates what this looks like in practice. The model doesn't just identify that a property is in a wildfire zone — it predicts which specific properties are most likely to survive or be destroyed, based on property-level characteristics including building materials, defensible space, vegetation proximity, and roof condition. During the January 2025 Los Angeles wildfires, Z-FIRE correctly identified 94% of the area impacted by the Palisades Fire and 87% of the Eaton Fire area as high or very high risk. Critically, even in areas with 60% overall destruction, the model showed that highest-risk properties were 50% more likely to be destroyed than lowest-risk ones — demonstrating property-level differentiation within the same catastrophe zone (ZestyAI). This is not portfolio enrichment. This is a data input that changes which properties an underwriter accepts and at what price. The model has regulatory approval for rating and underwriting across wildfire-prone US states, and ZestyAI reports 62x better risk segmentation than traditional rating methods.

2. Continuous condition monitoring via IoT. The escape-of-water peril in commercial property offers a clear example of how sensor-based digital twins can move from monitoring to underwriting input. Water damage is the leading driver of non-catastrophic commercial property claims. The traditional underwriting approach — assessing plumbing age, building maintenance records, and occupancy type — captures some of the risk, but cannot detect the slow leaks, failed valves, and burst connectors that cause the majority of losses. Chubb has deployed thousands of water leak sensors in US commercial properties, with sensor-activated shutoff valves that can isolate a leak before it causes significant damage. Insurance Journal reports that customers using active monitoring devices with shutoff capability could see premium discounts of up to 8% (Nordic Semiconductor / Insurance Journal). Insight Risk reports that its bundled IoT sensor and water damage prevention solution has reduced water damage claims by up to 90% in construction projects (Eyrus / Insight Risk). When a technology can demonstrate that level of claims reduction, it moves from a risk management recommendation to a pricing factor.

3. Real-time natural catastrophe exposure. The combination of flood forecasting, seismic monitoring, and weather data with property-level digital twins creates the potential for dynamic exposure management — not just knowing what a property looks like, but knowing in real time what environmental hazards it faces. This connects to the predict-and-prevent model discussed in the parametric insurance context: if a digital twin integrates with flood prediction systems providing 48-hour advance warnings, and the property owner demonstrably acts on those warnings to move stock and activate defences, the expected loss from a flood event is genuinely lower than for an identical property without that capability. The Previsico–Descartes partnership described elsewhere on this site illustrates how flood intelligence and parametric triggers can combine. The digital twin becomes the connective layer — the persistent, property-level data model that integrates the prediction, the mitigation action, and the insurance trigger into a single, auditable risk picture.

The maturity spectrum

It is useful to think about digital twins in commercial property insurance as sitting on a maturity spectrum, where each level delivers increasing value but also requires increasing evidence:

Level 1: Submission enrichment. The digital twin automates and enriches COPE data, reducing the time underwriters spend sourcing missing information and increasing the proportion of submissions that can be triaged efficiently. This is where most current implementations sit. It is operationally valuable and addresses a genuine bottleneck — the 40–50% of submissions arriving incomplete — but does not change the underwriting decision itself.

Level 2: Portfolio-level risk insight. The digital twin provides aggregate views across the portfolio — accumulation exposure, concentration of specific construction types, proximity to natural hazard zones, identification of properties with characteristics correlated with higher historical loss. This supports portfolio management, reinsurance purchasing, and regulatory reporting, but still operates at the portfolio rather than individual risk level.

Level 3: Property-level risk scoring. The digital twin produces a risk score for a specific property that has a demonstrated, validated correlation with actual loss outcomes. The underwriter uses this score to accept, decline, or price the risk differently. This is where the data becomes a genuine underwriting input. ZestyAI's wildfire, hail, and wind models operate at this level. Reaching it requires loss data, actuarial validation, and in many jurisdictions, regulatory approval.

Level 4: Dynamic, continuous underwriting. The digital twin integrates real-time sensor data, environmental monitoring, and predictive analytics to enable continuous risk assessment throughout the policy period — not just at quote and renewal. Pricing adjusts based on the property's current condition and exposure. IoT-based water leak detection with premium discounts is an early example. This level represents the ultimate promise of the technology, but requires infrastructure, data governance, and commercial model innovation that most insurer-technology partnerships have not yet achieved.

The chicken-and-egg problem — and how to solve it

This creates a well-documented chicken-and-egg problem in insurance technology partnerships. Technology companies need claims data to validate that their digital twin data predicts loss. Insurers are reluctant to share claims data until the technology has demonstrated its value. The result is a stalemate that keeps promising technology stuck at Level 1 or 2.

There are four approaches that have shown traction in breaking this cycle:

Pilot with loss data access. The most direct route. Structure a partnership where the insurer provides anonymised historical claims data for a defined portfolio segment, and the technology company demonstrates retrospective correlation between its digital twin risk scores and actual loss experience. If the analysis shows that properties the digital twin scored as high-risk had meaningfully worse loss outcomes, the case for Level 3 integration is made. This is how ZestyAI built its validated models — by training on actual claims data and demonstrating lift against real outcomes.

Use prevention as the proof. If the digital twin includes an active risk reduction component — IoT sensors that detect and prevent water leaks, flood forecasting that enables asset relocation — the claims impact is measurable directly. The insurer doesn't need historical correlation analysis. They can observe the difference in claims frequency and severity between properties with the prevention technology deployed and those without. Insight Risk's reported 90% reduction in water damage claims on monitored construction projects is an example of this approach (Eyrus / Insight Risk).

Start with a peril, not a platform. Rather than offering a comprehensive digital twin of the entire property, focus on a single peril where the data has the clearest causal link to loss outcomes. Wildfire risk based on defensible space and building materials. Escape-of-water risk based on plumbing age and sensor data. Flood risk based on elevation, drainage, and forecasting. A narrow, peril-specific digital twin with strong loss correlation is more valuable to an underwriter than a broad, multi-dimensional model with no validated relationship to claims.

Leverage Lloyd's Lab and insurer innovation programmes. Intelligent AI developed through Lloyd's Lab, which provides structured access to insurer partners and data environments. Programmes like these exist specifically to bridge the evidence gap between technology capability and underwriting adoption. Technology companies that approach these programmes with a clear hypothesis — "our data predicts this specific type of loss better than current methods" — are more likely to emerge with the validated evidence they need.