Insurity

The removal test settles this at the same level as every other core platform in this lane. Strip the models out and a policy administration system, a billing engine, a claims system and a marine and specialty estate all keep running, because carriers bought them to run the business rather than to score it.

What sits above that line is genuine rather than decorative: Valen builds custom predictive models on pooled consortium data and returns scores that drive risk selection and pricing, and Insurity Predict packages the same capability for underwriting. That franchise predates the current wave and is sold on its own terms, which is more than a bolted on assistant.

Against it, the newest work in the Borealis release is assistive rather than determinative, covering a premium audit assistant for policyholders and conversational querying inside the geospatial product. The company's own May 2026 positioning reinforces the grade rather than lifting it, since it aims artificial intelligence at reducing the cost and time of configuring products in the core, which is a build economics argument rather than a claim that models decide.

One published capability is a real autonomy claim and it is stated plainly: the specialty suite prioritises the most profitable submissions and automatically processes or declines straightforward risks using configurable rules. Automatic decline is the sharper half of that sentence, because a submission refused without a human is an outcome the applicant experiences, and the customer implementation documented with a workers compensation and commercial lines carrier names straight through processing as an explicit goal of the deployment.

The control is carrier configured rules, which puts the boundary in the buyer's hands rather than the vendor's, and that is the honest reading of the design. What is absent is any published account of where the vendor recommends the boundary sits, what referral triggers exist, whether a declined submission is flagged for review, and what the assistive features may complete without a person. The default in this lane is that the carrier owns the setting and is left to work out the safe value alone.

This is the axis where the vendor publishes more than its peers rather than less. True rating transparency is named as a delivered capability in the Andromeda release rather than described as a principle, and the modelling practice is documented at a level a model risk function can actually use: calibrated models built and deployed in an average of 12 to 18 weeks, production ready models in 4 to 6 weeks, and models described as custom built to a stated objective rather than sold as a fixed score.

Publishing a build and deployment timeline is a modest disclosure that says something real, because it implies a defined fitting and validation cycle rather than a static product. The reservation is that the validation content of that cycle is undescribed, with no published account of out of sample testing, monitoring for drift after deployment, revalidation cadence or what happens when a fitted model and a carrier's own loss experience disagree. The grade reflects disclosure of process shape without disclosure of process rigour.

Scale is stated precisely and repeated without variation across separate announcements, which is a mild check on it: 22 of the top 25 United States property and casualty carriers, 7 of the top 10 managing general agents, more than 400 cloud based deployments and over 500 carriers and managing general agents on the software.

Customers are named rather than described, including Glatfelter Insurance Group, Markel, UBIC, LUBA Casualty and Frank Winston Crum Insurance, and one is documented at use case level, with a workers compensation carrier's own account of moving to a predictive model and processing higher volume as a result. What holds the grade below the top band is the shape of the headline numbers.

Three times the industry average growth rate over five years, loss ratios 3 to 10 percent better than industry average and a 5 percent improvement in claims costs are all stated as customer outcomes without a named measurement basis, comparison universe, period or attribution method. Those are the figures a buyer would most want audited and they are the ones carrying the least method.

Stewardship here has a specific shape that the vendor discloses in outline and leaves undefined in substance. The Valen Data Consortium is a pooled asset built from contributed carrier premium and loss experience, sized publicly at 109 billion dollars of premium, and models fitted on it are sold back into the same market.

Publishing the existence and the size of that pool is more disclosure than most data advantaged vendors offer, and the earlier consortium figure for commercial automobile was also published, which allows a buyer to see the asset growing rather than being asserted.

What is missing is every control question that follows from the structure: the de identification method applied before contribution, whether a contributor can opt out of having its data inform competitor models, what happens to contributed data on termination, and whether the pool feeds any generative capability.

The assistive features added in Borealis raise a separate untreated question, since a policyholder facing premium audit assistant handles insured party information directly and nothing published describes its data handling.

Two passes located no privacy policy content, data processing terms, retention schedule or subprocessor list, and no statement of how policyholder data flows between a carrier's core instance and the analytics estate. That gap matters more here than for a pure core vendor because of the consortium structure: carriers contribute premium and loss data into a pooled asset, models are built on the pool, and scores are sold back to the contributing market.

The pooling itself is disclosed openly and sized publicly, which is the honest part. What a buyer cannot establish from anything published is the boundary around its own contribution, specifically whether a carrier's data informs models sold to a direct competitor, what de identification is applied before pooling, and whether contribution is a condition of purchase. Those are answerable questions and nothing published answers them.

Two passes across the site, the press archive and third party coverage located no trust centre, no service organisation control report, no information security certification and no security page of any kind, which is the weakest published position of the core platforms indexed in this lane.

The grade sits at the middle band rather than lower on the inference standard already applied elsewhere in this index, namely that buyers of this profile conduct their own security review before onboarding. Stated penetration of 22 of the top 25 United States property and casualty carriers means the controls have been examined repeatedly by parties with the leverage to demand evidence and the expertise to read it, so the controls very probably exist and are simply not published.

That reasoning supports the grade and should not be mistaken for the evidence itself, and the practical consequence for a buyer is that nothing can be assessed before a non disclosure agreement is in place.

The company is a software and analytics supplier rather than a licensed insurance entity, and carries no licence of its own, which is the ordinary position for this lane. The regulatory surface it does touch is filing rather than authorisation: built in regulatory intelligence automatically updates rates, rules and forms so carriers can expand into new markets while staying compliant, which places the product inside the state by state filing machinery of United States property and casualty insurance.

That is a meaningful and unusual claim, because a vendor maintaining rate and form currency on a carrier's behalf is assuming work with a regulatory consequence if it is wrong. Nothing published sets out how that intelligence is maintained, how quickly changes propagate, what happens when a filing is rejected, or where responsibility rests between vendor and carrier when a rate in production is out of date.

Two passes located no responsible artificial intelligence statement, fairness testing description, bias audit or model governance policy, which places this vendor level with the rest of the core platform tier and below the pure play insurance specialists that publish one. The absence is more consequential here than the uniform grade suggests, because the products most exposed are pricing and risk selection models.

A rating model has to be explainable to a state regulator before it is accurate to an actuary, and the specific hazard is a model proxying for a prohibited rating factor through correlated variables, which pooled consortium data across many carriers makes easier rather than harder to do inadvertently. The company markets true rating transparency as a named capability, so the vocabulary of explainability is present in its positioning while the evidence of fairness testing is not. A buyer should treat those as separate questions and ask for the second.

Recourse is unaddressed and the exposure runs to parties outside the contract, which is the pattern across this tier. The models score risks for selection and pricing, and the specialty suite can decline a straightforward submission automatically.

The person or business on the other side of that outcome pays a higher premium, is declined, or is never quoted, and has no relationship with this vendor, no notice that a consortium model informed the decision, and no published route to see or contest the inputs.

Between vendor and carrier the position is equally undefined, with nothing published on liability allocation for a model that misprices, a rate that is out of date after an automatic regulatory update, or an automatic decline that proves wrong.

The consortium structure sharpens the point rather than softening it, because a model fitted on pooled industry experience can produce a correlated outcome across many carriers at once, so an error does not stay contained to the carrier that bought it.

Provenance is clear for the part of the estate that predates the current wave and opaque for the part that does not. The predictive models are the vendor's own, fitted on a pooled data asset it owns and sizes publicly, which is a straightforward and verifiable supply chain with no third party model dependency to disclose. The newer conversational and assistive capabilities are the gap.

An always on assistant guiding policyholders through premium audits and conversational querying inside the geospatial product both indicate a language model underneath, and no provider, model family, version or hosting arrangement is named anywhere published, nor is any statement made about whether prompts or insured party data reach a third party service. Naming the cloud provider for the analytics products while leaving the model provider for the assistants unnamed is the specific asymmetry a buyer should raise.

The vendor is the core system, which settles the axis at the top band on the same reasoning applied to the other enterprise platforms in this lane. Depth is evidenced rather than asserted: more than 400 cloud based deployments, an estate covering policy administration, billing, claims, rating and marine and specialty operations, and internal integration documented at product level, with the analytics models connected into the data management and policy decisioning platforms so scores reach the point where a risk is selected or priced.

Outward integration is claimed as the largest open network of data providers in property and casualty analytics, and the data hub architecture is described as open and cloud native rather than closed. The one structural qualification a buyer should note is that the strongest evidence concerns integration inside the vendor's own estate, and interoperability with a rival core is not addressed anywhere published.

Deployment is cloud only and the company treats that as a differentiator rather than a constraint, describing itself as cloud native and claiming more cloud based deployments than any other core system provider in its market. The underlying infrastructure is named openly for the analytics products, with Valen Analytics and Insurity Predict stated as deployed on Amazon Web Services, and naming the hyperscaler is a disclosure many peers withhold. Beyond that the record thins.

Two passes located no region list, no data residency options, no statement on where consortium data is held or processed, and no on premises or private cloud path for a carrier whose own policy requires one. For a business concentrated in United States property and casualty that gap is less binding than it would be for a cross border vendor, and it remains unpublished rather than answered.

Pricing is absent across the site, the press archive and third party coverage, in line with every core platform in this lane. The commercial argument is made comparatively instead of numerically, with the specialty suite positioned as delivering enterprise capability at a fraction of the cost of other core platforms, and the May 2026 market statement urging carriers to stop signing contracts, extensions and expansions on the grounds that artificial intelligence has added a vendor line item rather than removing one.

Making cost the centre of the public argument while publishing no cost of its own is the tension a buyer should name in the first meeting. One useful signal does appear: self service configuration is presented as removing the system integrator dependency, which speaks to the cost that usually dominates ownership here, and it is unquantified like the rest.

Coverage is broad on both axes that matter in this lane, institution type and line of business. On institution type it spans large carriers, specialty carriers, brokers and managing general agents, with a purpose built suite for the smaller specialty end rather than a stripped configuration of the enterprise product, and stated penetration of 22 of the top 25 United States property and casualty carriers alongside 7 of the top 10 managing general agents.

On lines of business the consortium and analytics estate is documented across homeowners, personal automobile, workers compensation, commercial automobile, commercial package, commercial property and business owner policies, with marine handled as its own operational area. The concentration is United States property and casualty, which is where the depth and the named customers sit, and international presence is asserted less specifically than the domestic position.