hyperexponential

hx Renew would survive the loss of its AI intact. The pricing itself is performed by models the carrier's own actuaries write in Python, and no model learns the price. What the company is bought for, rating complex specialty risk, would keep working with the machine learning stripped out. The AI line is real and shipped rather than a roadmap.

It includes a data ingestion library and a virtual actuarial assistant, plus Portfolio Intelligence and Triage, which reads submissions and prioritizes against appetite with agentic reasoning. That layer sits at ingestion, triage and authoring assistance around a rating core that predates it. The company has repositioned publicly as an agentic underwriting platform, and the picture changes if the agentic products become the reason carriers buy.

The word governed is doing the work a mechanism should be doing. The company states that audit trails and governed logic make agent led underwriting explainable, and that agents are called as governed tools inside every workflow. Both assert that constraints exist, and neither describes what the constraint is. No confidence threshold, escalation trigger or review queue is named. Nor is any approval gate or sampling control on automated output.

The audit trail is a reconstruction capability after the fact rather than a control in the path. At least one competitor publishes three named automation modes, with a random audit on the fully automated tier.

The change control is strong, and it covers only the models the customer wrote. Every decision, every rule change and every model update is captured end to end, framed explicitly as a path for regulatory review. The carrier's own actuaries author and own the models rather than receiving them as a black box, and customers hold structured access to all data, configuration and code at any time.

A model risk function could reconstruct how any given price was reached and what changed since the last version, which is what supervisory model risk guidance asks for. The AI layer hyperexponential itself supplies is a different matter. Described as foundation models, agent frameworks and inference governed for insurance, it carries no published validation evidence or accuracy figure. There is no account of how the agentic triage reasoning is tested.

Named carriers and published aggregate outcomes sit side by side. The carriers span three market shapes: Aviva, Markel, Sompo, Convex, Conduit Re, HDI and the specialty managing general agent Novacore. More than $45 billion of gross written premium is contracted through the platform annually. Published outcomes include quote turnaround reduced by up to 50 percent and an average loss ratio improvement of one to two percentage points across clients.

Pricing models deploy up to ten times faster. The company also reports growing sales tenfold since its 2021 Series A while remaining profitable, a harder claim to make loosely than a customer count. No per client figure is attached to a named carrier.

Competing carriers share the platform, and nothing states the boundary between them. Aviva, Markel, Sompo, Convex, Conduit Re and HDI all price risk on it, and pricing models are the most closely held intellectual property a specialty carrier owns. Whether one carrier's models, submission flow or pricing outcomes inform anything served to another is unstated. The portability commitment gives customers structured access to their own data, configuration and code.

That is a real stewardship positive, but it governs a customer's access to its own assets rather than the boundary between customers. At least one competing insurance platform with an overlapping client list states that boundary in public.

Privacy exposure is limited, because commercial property and casualty submissions concern businesses, their assets and their loss history rather than individuals. Consumer financial privacy regimes bear lightly on the processing path. The material is still commercially sensitive to a degree that would ordinarily attract a published handling statement, and none exists outside the gated trust center. In public there is no data processing agreement or subprocessor list, and no retention schedule or lawful basis statement.

Both certifications enterprise insurance buyers ask for are held and named on the vendor's own product pages: SOC 2 Type II and ISO 27001:2022. A dedicated trust center operates at its own subdomain on a continuously monitored platform.

A single badge row puts two audited certifications beside four supervisory bodies. It reads SOC 2 Type II, ISO 27001, Prudential Regulation Authority, National Association of Insurance Commissioners, Bermuda Monetary Authority and Lloyd's, and the company states elsewhere that the platform is designed for those regimes. The four supervisory bodies have certified, approved and admitted nothing.

Designed for is a statement of build intent, not a credential, and placing four supervisor names beside two real certifications invites a buyer to read approvals that do not exist. hyperexponential is a technology supplier to carriers and holds no insurance license, which is the correct posture. Set those four names aside and there is no formal admission, supervised regulator test or program enrollment. The underlying claim is not false, and the regimes named are those of its buyers, so this is a presentation finding rather than an integrity one.

The exposure is lower than in consumer credit or life and health underwriting, but it is not absent. Commercial property and casualty specialty lines weaken the protected class analysis, since the insured is usually a business. Appetite based triage still decides which submissions receive attention first. A prioritization model trained on historical writing patterns will reproduce whatever the book has historically favored, including by geography and industry class. Nothing published addresses fairness, differential outcomes or bias testing, or gives any account of that risk.

Reconstruction is not recourse. The audit trail lets a carrier reconstruct after the fact how a price or a triage decision was reached. The consequence of a misprioritized submission falls on the broker whose client waits and on the insured whose risk is deprioritized, on an appetite judgment neither of them sees. Neither has any stated route to challenge it. For the AI layer, no error rate or remediation commitment is published, and there is no liability position or correction path.

The platform layer is described as foundation models, agent frameworks and inference, governed for insurance. That names the class and identifies no supplier or model family, and no version or country of processing. The gap is narrower in one respect: the pricing models themselves are written by the carrier's own actuaries and are fully known to the buyer. Third party models sit in the ingestion, triage and assistant layer. There a buyer cannot determine whether broker submission content, loss runs and financials reach an external provider.

The integration posture is explicit and the partner list is named. The platform runs inside underwriting workbenches, policy administration systems and broker portals, as well as APIs and scheduled jobs. The company states that the carrier's policy administration system remains the system of record. Named integration partnerships exist with Send for its underwriting workbench, with Cytora for risk digitization, and with Supercede.

Pre built integrations and code snippets let customers plug into alternative data sources and algorithms at the point of pricing. Portability is written down too: customers retain structured access to all data, configuration and code at any time, stated as a commitment to transparency and operational continuity.

A carrier can tell how to leave and cannot tell where its data sits while it stays. Customers hold structured access to all data, configuration and code at any time, stated as a commitment to portability and operational continuity. No region list or residency commitment is published, and there is no single tenant option or self hosted path. The platform serves customers across the United Kingdom, Europe, Bermuda and the United States, several of which impose insurance data requirements.

Outcome claims are stated with precision while cost is absent. An average loss ratio improvement of one to two percentage points and a halving of quote turnaround are published with nothing indicating what the platform costs. A carrier cannot construct the return calculation the outcome figures are plainly designed to support. No pricing or tier structure is published, nor any billing basis or indicative range, and the route to a number is a demo request.

Depth in one segment rather than breadth across several. Commercial property and casualty is served thoroughly, including specialty and excess and surplus lines. Reinsurance runs through a dedicated reinsurance pricing model, and managing general agents are covered across admitted, non admitted and reinsurance business on one foundation. Geographic reach spans the United Kingdom, Europe, Bermuda and the United States. Life and health is absent entirely, and some competitors serve both halves of the insurance market.