Korint

B with the tension stated, and the vendor states it more plainly than most: its own position is that AI cannot be effective without a modern platform underneath, and that insurers should invest in the platform first rather than deploy AI on outdated systems. That is an accurate description of what is being sold.

Strip the agents and a modern cloud core insurance system remains, covering product configuration, rating, policy administration, distribution and claims, and that is a saleable product in its own right, which is what the incumbent policy administration vendors sell. The AI is substantial and shipped rather than announced, roughly 50 agents covering core management actions plus an AI Companion, so the centrality floor is not in question, but the platform is the asset. Checked against three insurance platform peers in the index and this is in line rather than harsh: Federato, Akur8 and Artificial Labs all sit at B on this axis.

Roughly 50 agents are described as covering core management actions across a platform that administers policies and processes claims, and no oversight model is published for any of them. That is the gap: nothing states which agent actions execute directly against a policy or claim versus draft for approval, what value or complexity threshold routes to a human, how an agent decision is reviewed or reversed, or what a claims handler sees when an agent has already acted.

The consequence class is higher than for the research and analytics vendors because the party affected is a policyholder at claim time, when an incorrect automated action denies or delays payment on a live loss. The MCP servers extend the surface further by letting a client run its own agents against the platform, and no boundary is described for those either.

Nothing published on how any of the roughly 50 agents were evaluated, what accuracy they achieve on the actions they take, how the AI Companion is grounded, or what happens when an agent misreads a policy document or claim submission. This is the shape Aiera was credited for addressing and almost nobody else is: publishing how a generative output was evaluated, including the limits of the metric, is the achievable standard here and it is absent.

The EU AI Act dimension is worth naming given the European buyer base, since insurance pricing and claims handling attract obligations that fall differently on a provider of a system than on the deployer using it, and the provider and deployer split is not addressed anywhere.

Multiple named customers, no quantified outcome attributed to any of them, which is the B bar exactly. WTW, Aon and Marsh are named as clients alongside local MGAs, with 25 plus insurance products launched or migrated across them. Three of the largest global insurance brokers is a strong reference set for a company of about twenty people.

A named customer testimonial exists from Yeet VTC, a French ride hailing platform running embedded insurance, and an unattributed client quote describes centralising underwriting, accounting, reporting and distributor oversight across ten business verticals. What stops this from being an A is that the headline numbers are the vendor's own: four week product launches benchmarked against a Deloitte study putting the industry average at 12 to 18 months, and a ten times faster claim. Those compare the vendor to a published market figure rather than reporting what a named client measured.

B on architecture rather than policy, and the question is real because of who shares the platform. An insurer, the MGAs holding its delegated authority, and competing wholesale brokers can all transact on the same system, and the vendor's stated answer is structural: exposed features are compartmentalised by partner, access is controlled, and full traceability of all exchanges is maintained.

Distributor oversight is presented as a feature, which means the platform is deliberately the place where one party watches another, making the separation boundary load bearing. Thinner than VantedgeAI's B, which named isolated workspaces, role based access and deployment inside the client's own cloud, and nothing here states whether portfolio or pricing data informs anything served to another partner.

GDPR compliance is asserted and some structure is described, specifically controlled access, features compartmentalised by partner and full traceability of exchanges. That is more than a bare claim but less than a posture: no data processing agreement, retention schedule, subprocessor list, impact assessment or breach commitment is published.

The sensitivity is high and specific to the lines it serves, since motor and fleet policy and claims records carry driving history, incident detail, location data and in bodily injury claims medical information about identified individuals who are the insurer's customers rather than the platform's.

Two named attestations, ISO 27001 and SOC 2 TYPE 2, plus GDPR alignment. The type distinction matters and is worth recording, following the point raised on VantedgeAI: a type 2 report tests whether controls operated effectively over a period, where a type 1 is only an auditor's opinion that controls were suitably designed at a point in time. This is the stronger one.

Supporting detail is described rather than only asserted, specifically controlled access, features compartmentalised by partner and full traceability of exchanges. Short of A because there is no trust centre publishing report scope, audit period, penetration testing cadence or subprocessor list, so a buyer must request the reports to learn what was actually in scope.

No licence, regulated entity or supervisory relationship named, and the perimeter here is unusually concrete rather than theoretical. The vendor states that its platform quality is what allows wholesale brokers and MGAs to obtain full delegations from their insurers, which means the system is being offered as the control evidence supporting a delegated underwriting authority, and delegated authority is a supervised arrangement in which the insurer remains accountable for what the delegate does.

Insurance distribution duties under the IDD, product oversight and governance requirements covering product approval and target market, and French market conduct rules all sit directly on the activities the platform automates. No public material engages with any of it.

No governance framework, model inventory, review process or fairness testing published. The exposure is concentrated in the rating engine rather than the agents: pricing management is a named core module, and property and casualty pricing sits directly inside anti discrimination law, including the European prohibition on gender as a rating factor and the general problem that proxies such as postcode, vehicle type and occupation correlate with protected characteristics.

A platform whose selling point is letting a client configure and launch complex products in weeks is a platform that accelerates the deployment of rating structures nobody outside the client has reviewed. Nothing describes constraints on what a configured rating model may use, or any testing of the resulting price dispersion.

No warranty, service level, accuracy commitment or remedy published, on a platform where the agents act inside claims. The recourse question has a distinctive shape here because of the delegation chain the vendor is built to enable: a policyholder whose claim is wrongly handled complains to the broker or MGA, which holds delegated authority from the insurer, which carries the regulatory accountability, while the automated action originated in a platform none of them built.

Nothing states where responsibility sits when an agent rather than a delegate makes the error, and the policyholder is three parties removed from the system that acted. Full traceability of exchanges is published, and as recorded for Aiera, tracing an error is not the same as having recourse for it.

Roughly 50 agents and an AI Companion are shipped and no underlying model, provider or version is named anywhere. That gap is larger than usual for two reasons. First, the sovereignty positioning: the platform is sold on European hosting and GDPR alignment against non European competitors, and if the agents call a United States foundation model provider then the sovereignty claim covers the data at rest but not necessarily the inference path, which is exactly the distinction a French insurance buyer is being invited to care about.

Second, the MCP servers invert the usual direction of the question, since a client's own agents, built on models the vendor did not choose or assess, are being connected directly to policy and claims data, and nothing states what those agents may reach or what the client is told about the boundary.

The strongest axis and it is in line with insurance platform peers, since Federato and Artificial Labs both hold A here. Three things earn it. First, the deliberate no rip and replace design: modular components plug into an existing back office via API with no back office modification, so the vendor can be either the core system or a layer on one, and stated deployment averages a month with no integration fees.

Second, an event based architecture with an insurance API positioned explicitly so a client can run its own AI tools against platform portfolio data. Third, and the part that is genuinely ahead of the index, MCP servers that let a client connect its own internally developed agents directly to the platform.

That makes the core system machine addressable to the buyer's own agents rather than only to its people, which is the same distribution instinct Aiera showed on the content side and which almost no core platform vendor in this index has built.

B on a stated residency commitment, which 271 of 308 vendors do not make. Hosting is stated as Europe, paired with GDPR alignment and French origin, and sovereignty is used as an explicit selling point against non European core platforms in the vendor's own competitive comparison. Deployment shape is also described rather than implied: all in one as the core system or modular alongside an incumbent, cloud delivered, roughly one month to live, unlimited seats. Short of A because no specific region, cloud provider, single tenant option or in country residency guarantee is named, which matters for buyers outside France in the same European market.

Unusually forthcoming for this index, where 263 of 308 vendors sit at C. No rate is published, but the pricing model and three specific commercial terms are: pricing is success based on volume processed through the platform, there are no per user licences and seats are unlimited once live, and there are no integration fees. Average deployment is stated at about one month.

That tells a buyer the shape of the bill and how it scales, which is more than the disclosure of a bare unit of consumption that left Cohesion at C. It stops short of A because no rate, band, minimum or worked example appears, so a prospective MGA still cannot estimate annual cost.

Several distinct roles served, all inside one vertical. Buyers are insurers, managing general agents, wholesale and delegated authority brokers, and the product reaches embedded insurance distributors such as the ride hailing platform named in its own testimonial. Lines are property and casualty, with a stated concentration in business to business mobility including motor and bicycle fleets, plus specialty and retail embedded lines.

Geography is European with French market fit presented as a selling point. No life, health, reinsurance or non insurance financial services surface, which is what keeps this off the A held by broader insurance vendors such as Akur8.