Renew Risk

Machine learning is named as part of the method, with technical profiling describing large cloud simulations utilising machine learning alongside proprietary hurricane and earthquake models and enhanced credit analysis. The nine month model development cycle against an industry norm exceeding three years implies computational methods doing substantial work.

Against that, the company's own framing is deliberately deep science and science first rather than artificial intelligence led, and catastrophe modelling rests on engineering, meteorology and statistical simulation of physical hazard, which is a discipline that predates and does not depend on modern machine learning. Strip the models and the underlying physical science remains, which is the substance a reinsurer is buying.

The output is a loss distribution that an underwriter, actuary or credit committee interprets, not a decision, and the company frames it accordingly as enabling participants to make data driven decisions and allowing insurers to understand risk and confidently provide capacity.

Catastrophe models sit inside established actuarial governance where model output is one input to a rate and a capacity decision made by qualified people, so the human role is preserved by the structure of the market rather than by a vendor promise. What is absent is any statement of uncertainty communication, which matters because these models cover assets with little loss history and the confidence around an output is as important as the number.

The validation approach is stated and it is the right one for this discipline: models were developed and validated in collaboration with early market adopters to ensure they reflect real world underwriting and risk management needs, which means practitioners who carry the losses tested them before release.

Credentials support it, with the chief product officer a chartered engineer and fellow of the national engineering academy who won a resilience award at the industry's own catastrophe modelling awards. The problem statement is also unusually well argued, that assets are larger, more exposed and not accurately captured by historical data, which is the correct justification for simulation over empirical loss experience. What is missing is documentation: no validation methodology, sensitivity analysis or independent review is published.

Two insurance market participants are named. A major British insurer's head of renewable energy and engineering is quoted on the model launch, describing the firm as a lead market in this class, and a specialist renewable energy underwriter entered a published partnership to enhance risk analytics, with its chief executive and team named. The company states models are live with global reinsurers.

Product coverage is evidenced by launch rather than claim, spanning first in region catastrophe models for the United Kingdom, Ireland and Europe, on top of existing models for Taiwan, Japan and the United States. Around 6.7 million pounds has been raised across two rounds. What is absent is scale: no customer count, insured value modelled or portfolio figure is published.

No data boundary statement was located. Models were developed and validated in collaboration with early market adopters, which means underwriting expertise from specific insurers shaped tools now sold to their competitors, and that is the ordinary economics of catastrophe modelling rather than a defect.

What is not addressed is the exposure side: an insurer running its portfolio through the platform discloses what it insures and where, and nothing states how that data is separated between carriers competing for the same renewable programmes.

The cleanest privacy position in this index by subject matter. The inputs are physical asset characteristics, geographic location, engineering specifications and hazard science, and the outputs are loss distributions for infrastructure, so no personal data enters the payload at any point and no individual is the subject of any assessment. What is commercially sensitive is a client's exposure portfolio, since the assets an insurer is modelling reveal what it is writing. Held at B because no data processing terms, retention schedule or subprocessor list was located.

No attestation, certification, trust centre or enumerated framework was located. Global reinsurers running live models and a specialist underwriter in a published partnership mean vendor assessment has been passed, and the insurance market's own supplier standards are demanding, so the assurance exists privately while nothing is published for a prospective carrier to read.

No supervisor, statute or instrument is named, and the omission is specific rather than general. European insurance capital rules require that any model used in determining capital requirements be validated and documented to a defined standard, with governance around changes, and a catastrophe model feeding an insurer's view of risk sits directly inside that framework. A vendor selling into that market would ordinarily state how its models support those obligations, and nothing published does.

No individual is assessed and the adapted exposure is consequential at a different scale. A catastrophe model determines whether a renewable project can be insured and therefore whether it can be financed, so a model that overstates hazard prices projects out of a region while one that understates it leaves carriers and lenders exposed to losses they did not reserve for.

Coverage compounds it: models exist for the United Kingdom, Europe, Taiwan, Japan and the United States, which means projects in unmodelled regions face underwriters with no quantitative basis and correspondingly less appetite, so model availability shapes where energy transition capital can flow. No validation results, uncertainty ranges or coverage roadmap were located.

No guarantee, indemnity or falsifiable commitment was located, which is conventional for catastrophe modelling since no provider warrants a loss estimate and carriers retain their own view of risk. The insurer is nonetheless the party with recourse, since it can test the model against its own experience and adjust.

The project developer whose asset is modelled has none: a plant assessed as higher risk faces higher premiums or unavailable cover, and nothing describes whether a developer can see the assumptions driving that assessment or challenge them.

No hazard data source is named, which is the material gap for this product, because catastrophe models are built on meteorological reanalysis datasets, historical event catalogues and terrain data whose provenance determines both accuracy and licensing, and a buyer assessing model quality would start there. No compute provider is identified despite large cloud simulation being central to the approach, and no subprocessor list was located. Model development is the company's own, drawing on its engineering and science team.

No integration is published, and for a catastrophe model that is the central practical question. Insurers run exposure management and pricing through established platforms and consume third party models through defined interfaces, so whether these models are available in those environments determines how easily an underwriter can actually use them. No exposure management system, modelling framework or data standard is named, and no developer documentation was located.

Large cloud simulations are referenced, so the computation is clearly hosted, and no provider, region selection, residency commitment or private deployment option is named. Exposure is lower than for platforms holding personal data, and an insurer submitting its portfolio for modelling is disclosing commercially sensitive exposure information whose processing location its own risk function would expect to be stated.

No pricing, packaging or basis of charge was located. Catastrophe models are conventionally licensed per peril and per region, and this company now offers several across multiple territories, so the commercial structure is visibly modular and undescribed. Nothing indicates whether charge scales by model, by exposure modelled or by seat.

Five buyer types are served, spanning insurers and reinsurers, insurance brokers, banks, developers and asset managers, and the company is explicit that they need the product for distinct purposes, pricing, capacity, financial planning and risk management.

Peril coverage spans windstorm, hurricane, earthquake and severe convective storm; geographic coverage spans the United Kingdom and Ireland, Europe, Taiwan, Japan and the United States; and asset coverage extends from offshore wind to onshore wind, solar, tidal and hydrogen. That is genuine breadth on three axes at once within one domain, and the domain itself is the limit.