Akur8

Transparent Machine Learning is the flagship product and the company's namesake: it automates feature engineering, variable selection and geospatial smoothing to build risk models the vendor states are produced up to ten times faster than manual methods, while keeping every coefficient legible. Gradient boosted models were added to complement the generalised linear and additive model families without surrendering interpretability.

The grade is B rather than A on the removal test applied to the whole company. Strip the machine learning and Arius reserving survives intact, a product that ran for years under Milliman with 150 clients and more than 1,500 users, alongside a rate repository and a real time deployment engine. Graded separately the pricing product would earn A and the reserving product would sit at C, which is a split worth stating rather than averaging silently.

The design intent keeps the actuary in the chair: models are built with full transparency, auditability and control, and the platform positions itself as empowering actuarial experts rather than replacing them. The output is also gated by a professional obligation the vendor does not control, since a certified actuary signs the model before it is filed.

The company has now introduced transparent agents to automate manual steps in actuarial workflows and market intelligence that reads unstructured regulatory filings, and neither carries a described approval gate, sampling control or review threshold in public material. That is the gap between B and the A held by Federato, which publishes three named automation modes and states that automated decisions are randomly audited by underwriters.

The second A on this axis in the index after Bretton AI, and it is earned structurally rather than by documentation. Every model the platform produces passes two external checks the vendor does not control: a certified actuary signs it under professional obligation, and a regulator receives it as a rate filing with the power to reject it.

That is a mandatory validation loop performed by a party with authority over the outcome, stronger than the reinsurer scrutiny that earned Federato part of a B. The platform supports it with full transparency of every variable contribution, traceable and consistent workflows, a rate repository built for submission, and automated documentation and reporting features aimed at satisfying audit requirements for internal control.

The qualification recorded for Bretton applies here too: the framework is described and the artifacts it produces should be requested in diligence. No accuracy figure for the Transparent Machine Learning engine itself is published.

Roughly 330 customers across more than 40 countries with approximately 200 staff, and the named roster reaches the top of the market: AXA, Generali, Munich Re, MAPFRE, HDI, Tokio Marine and MS and AD among global carriers, Canopius, Bass Underwriters, TMNAS, FCCI, NEXT, HDVI and Canal in commercial and specialty lines, and Cypress, Madison Mutual, Western Reserve Group, Georgia Farm Bureau, Sura, Mutua Madrilena and Matmut in personal and mutual lines.

Recent named adoptions include Branch and Tune Protect with executives quoted on the record. Outcome figures are stated as ten times faster pricing model development and four times faster reserving. The Arius acquisition added 150 insurance and consulting clients of whom more than a third are top tier carriers in the United States and Canada. Milliman holds equity in the company, which is a customer adjacent party putting capital behind the product.

The stated position is direct and unusually plain: the company describes its AI as transparent and human controlled and states that humanity comes first within the team, with clients and in the solution. Built in guardrails and full model explainability are named as platform properties, and the marketing argument is explicitly against embedding an opaque black box at the centre of a heavily regulated process.

What is not addressed is the cross customer question, and it is sharper here than in most of the index because risk selection and pricing are precisely where carriers compete. Competing insurers model their own books on one platform, and nothing published states whether one carrier's exposure data, loss experience or model structures inform anything served to another. Same open question recorded for Federato and Clearwater Analytics.

The security page addresses infrastructure protection and confidentiality but no data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The data at issue is policyholder level exposure and claims history for entire books of business, which is among the most sensitive material any vendor in this index handles, and the individuals whose records are modelled have a relationship with the carrier and none with Akur8.

A French company selling across Europe sits inside the General Data Protection Regulation by default, and the absence of an explicit statement of lawful basis, transfer mechanism or retention period is what holds the grade at C rather than a claim that protections are absent.

Service organisation control type two attestation is announced publicly with its scope named as security, availability and confidentiality, and the announcement quotes a named chief information security officer, which is more accountability than most of the index offers.

A dedicated security page describes protection designed in from the start, and the company states that the attestation report, certificates for the international information security standard and documentation of controls are available on request. Available on request is what holds this at B rather than A: a buyer cannot verify the current scope or period without entering a sales conversation, and no trust centre or continuously updated compliance portal is published. Still ahead of the category norm, where a footer line is common and Persona asserts certifications without naming any.

The product exists inside the rate filing regime rather than alongside it. The rate repository is described as regulatory ready and centralises rate management for submission, model outputs are filed with regulators and signed off by certified professionals, and the company states plainly that unexplainable outputs carry legal and compliance consequences in this domain.

Published work engages named regulatory events rather than generic compliance language: the United Kingdom ban on price walking in home and motor insurance effective January 2022, and detariffing transitions in Asia Pacific markets where risk model insights cannot be reflected directly in price because of regulatory limits. A market intelligence product now reads unstructured regulatory filings themselves. Akur8 holds no licence and needs none, which is the correct posture for a technology supplier under the index convention.

Insurance pricing is the most bias exposed function in this index, and the standing rule set with Sardine applies: explainability describes how a model reasons, not whether outcomes fall unevenly. What lifts Akur8 above the explainability floor is that constraint is a product feature, not just visibility.

Published work addresses pricing where risk model insights cannot be reflected in price because regulators forbid it, and names health and mandatory lines as ethically sensitive products requiring exactly that treatment, which means the platform is built to exclude and cap variables and to evidence that exclusion to a supervisor. State insurance regulators have adopted a model bulletin on insurer artificial intelligence use and several states impose unfair discrimination testing duties. What keeps this off an A is that no fairness testing methodology, proxy discrimination analysis or disparate impact tooling is named anywhere, and the obligation still lands entirely on the carrier.

No guarantee, indemnity, service level commitment on model quality or published falsifiable accuracy claim was located, so there is no commercial risk transfer of the kind Glia offers. What exists instead is recourse inside the regulatory regime rather than from the vendor: a certified actuary signs the model, a supervisor reviews the filing, and a policyholder who believes a rate is unfairly discriminatory complains to a state insurance department.

Full variable level transparency makes a challenged rate genuinely examinable, which is more than most of the index provides. The limit is the recurring shape across this index, sharpened here because the affected party is a policyholder paying a premium set by a model, who has no relationship with Akur8, is not told which system priced them, and has no route to the vendor at all.

The core modelling engine is proprietary and built in house, which shortens the chain at the point that matters most for pricing, and Amazon Web Services is named as the sole infrastructure provider. Arius arrived from Milliman and Milliman retains equity, so that lineage is disclosed.

What is not disclosed is anything about the newer layers: the transparent agents automating actuarial steps and the market intelligence product reading unstructured regulatory filings both imply language models, and no provider, hosting arrangement or data boundary for them is named anywhere. Whether carrier data reaches an external model provider during those operations is the first question a buyer should ask, and the same unresolved tension recorded for Rowspace applies here.

Cloud based and no code with open APIs, described as compatible with existing information technology ecosystems, and the rate repository plus real time pricing engine are the mechanism by which an approved model reaches production without an information technology project. The company frames removal of the information technology bottleneck as a core benefit and several customers cite time to market as the reason for adoption.

Guidewire Software is an investor, which is the dominant policy administration core in this market, but no named integration to Guidewire, Duck Creek or Sapiens was located, which is what separates this from the A held by Federato for native connections to all three. Arius brings its own established footprint in United States and Canadian carrier reserving environments.

Amazon Web Services is named as the cloud provider and the security page leans on the assurance programmes Amazon complies with. No on premise, private cloud or customer environment option was located, and no region selection, residency commitment or data localisation statement appears in public material.

That matters more than usual for a vendor whose customer base spans more than 40 countries including Asia Pacific markets with localisation rules and European carriers subject to transfer restrictions, and whose payload is policyholder level exposure and claims data. Compare Flagright, the only fully on premise option in the index, and DwellFi, where deployment location is written into the contract before signing.

No rate card is published, but the pricing model is disclosed and it is an unusual one: charges are based on the volume of insurance premium modelled on the platform and explicitly not on a per user basis, which removes the penalty a carrier would otherwise pay for putting more of its actuarial team on the tool. A free two week pilot with no obligation is offered to prospective clients, which is a real evaluation route in a category where competitors gate everything behind a demo request.

This is the Glia shape of publishing the model rather than the rate. It falls short of Sumsub's published per verification rates, which remain the index benchmark. The pricing basis and pilot terms reach the reader through a third party software profile rather than the vendor's own pricing page, which states only that pricing adapts to teams of any size.

The span runs from global composite carriers and reinsurers down to regional mutuals and managing general agents, with a white paper written specifically for lean actuarial teams at managing general agents and regional carriers who face the same pricing problems without enterprise budgets. Lines covered are all non-life classes including motor, property, commercial, specialty and health.

Geographic reach is more than 40 countries with published material addressing tariffed and detariffing markets in Asia Pacific as a distinct regulatory context rather than an afterthought. Arius extends the buyer set to actuarial consulting firms and to reinsurers. Four buyer roles are addressed on the platform: actuarial, underwriting, product and finance.