Artificial Labs

Machine learning does the ingestion and extraction work that makes the rest possible: broker submissions arrive as email, spreadsheets and slips in no consistent shape, and models turn them into structured risk data that can be triaged and matched against appetite. Smart Follow then applies model driven decisioning to commit capacity.

The grade is B on the removal test because Contract Builder, the placement workflow and the integration into the market's digital placement rails are a substantial product without any model in them, and digitising the London market's paper process is much of what the buyer is purchasing. Strip the extraction and the platform degrades to manual entry rather than ceasing to exist.

This is the most consequential autonomous act recorded in the index. Smart Follow reads slip data from the market's digital placement platform and decides whether to commit follow capacity to a risk without a human reviewing that individual submission, which means a machine creates a legally binding insurance contract and puts a syndicate's capital behind it.

The boundary is real and it is named: the agent operates within pre-set appetite and authority parameters, and a human underwriting director sets those rules. That is the same structure as the second of Federato's three published modes, where the system fully executes a defined subset of risks under predefined rules and guardrails. What is missing is Federato's third mode, the random audit of automated decisions by underwriters.

No sampling regime, no post bind review, no exception reporting and no described process for detecting appetite drift appear in public material, for a system whose failure mode is capacity committed to risks nobody looked at.

No extraction accuracy figure, no validation evidence, no error rate by document type or line of business, and no model documentation were located. The gap matters more here than for most extraction products because of what sits downstream: if a model misreads a limit, a deductible, a territory or a vessel class out of a slip, and the follow agent then binds against that misreading, the syndicate is on risk for terms it never agreed.

Structured data feeding an autonomous bind decision is the point at which extraction accuracy stops being an efficiency question and becomes an exposure question. Adherence to the market's Core Data Record standard constrains the shape of the data but says nothing about whether the values extracted into it are right.

The named roster is unusually strong for a company of 50 to 100 people and it covers both sides of the market: Apollo, Convex, Chaucer and AXIS on the carrier and managing agent side, Aon and Lockton among brokers, plus integration with the market's central placement platform.

The sharpest single fact is a named production deployment rather than a logo, with Apollo Syndicate Management running Smart Follow across Marine Hull, General Aviation and Marine Cargo, though that detail reaches the reader through third party market commentary rather than a vendor case study.

A 45 million dollar Series B closed in February 2026 with CommerzVentures, Move Capital and Augmentum Fintech, and MS and AD Ventures, a carrier venture arm, is on the register from an earlier round. What holds this at B rather than A is the complete absence of measurement: no quantified outcome exists anywhere, no submission volume, no bind rate, no cycle time reduction, from any named deployment.

The stewardship question here has a shape not seen elsewhere in this index and nothing addresses it. Artificial serves brokers and carriers simultaneously, and in a placement those two parties sit on opposite sides of a negotiation over terms, price and capacity. Named customers include two of the largest global brokers alongside several Lloyd's managing agents who receive their submissions.

Nothing published describes how information from one side is separated from the other, whether submission data from a broker informs anything served to a carrier, or whether appetite and pricing behaviour observed at one syndicate reaches a competitor on the same platform. This is the Rogo finding recurring in a different market, where one platform serves institutions routinely on opposite sides of the same transaction.

The privacy chain here is shorter than most of this index and that is structural rather than earned. Specialty and commercial risk data concerns corporate insureds, vessels, aircraft and cargo rather than individuals, so the subjects are largely businesses, the same mitigating property recorded for Nammu21's corporate borrowers.

Against that, no data protection agreement, retention schedule, subprocessor list or deletion commitment was located, and a United Kingdom company holding submission data for global brokers sits inside the United Kingdom data protection regime by default. Some specialty lines carry personal data regardless, including personal accident, crew and directors information inside submissions.

No attestation, certification, trust centre or dedicated security page was located, and no service organisation control report or international information security standard certificate is announced or offered on request. That is a visible gap for a platform holding live submission data from global brokers and bound contract records for Lloyd's syndicates, and it is the item a managing agent's own oversight function would raise first. Compare Akur8 in the same lane, which announces its attestation with a named chief information security officer and offers certificates on request.

The product is built to a named market wide data mandate rather than to generic compliance language. Smart Follow consumes slip data compliant with the Core Data Record, the standard imposed on the Lloyd's market under its Blueprint Two modernisation programme, and the platform integrates with the market's central digital placement infrastructure.

Building to a mandatory market standard that every participant must meet is a real admission process of the same class as Oscilar's Nacha preferred partner status. What is absent is any engagement with the conduct layer: no reference to the United Kingdom regulators supervising these firms, no discussion of delegated authority governance, and nothing on how algorithmically bound business is evidenced to a syndicate's own oversight function. Artificial holds no licence and needs none, which is the correct posture for a technology supplier under the index convention.

Specialty and commercial lines weaken the protected class analysis without removing the question, the same adaptation recorded for Federato. What replaces it is a market conduct exposure specific to algorithmic capacity: a follow agent applying an appetite model at scale will decline systematically, and if the declination pattern tracks industry, territory, vessel flag, operator size or domicile, the effect is that whole classes of insured lose access to capacity through a mechanism nobody reviews case by case.

Nothing published addresses whether appetite models are tested for unintended systematic exclusion, and the United Kingdom conduct regime places fair value and product governance duties on the firms deploying it. No fairness testing, no declination pattern analysis and no monitoring commitment were located.

No guarantee, indemnity or falsifiable commitment on extraction or decision quality was located, so nothing commercially binds the vendor. What lifts this off the floor is that accountability for an automated bind is unusually clear even though it does not sit with Artificial: the underwriting director who set the appetite and authority parameters owns the decision, the syndicate carries the capital risk, and a bound line is a legally enforceable contract the insured can rely on regardless of how it was written.

That is allocation of responsibility rather than recourse against the vendor. The party with no route at all is the insured whose risk an appetite model silently declined, who is never told a machine made that call and has nothing to appeal to.

The architecture is described in general terms as machine learning for data ingestion and extraction with algorithmic decisioning on top, and no model provider, hosting arrangement, third party data source or subprocessor is named anywhere.

Whether submission documents pass to an external model provider during extraction is unstated, and that question carries weight in this market because a slip contains commercially sensitive terms, pricing and capacity information belonging to parties who compete with each other. Consuming the market's Core Data Record standard discloses the shape of the data but nothing about whose models touch it.

Integration is with the market's own rails rather than with individual counterparties, which is a stronger position than a list of connectors. The platform reads from the central digital placement platform used across the London market, consumes data in the Core Data Record standard mandated market wide, and is described as automating flows across the legacy systems a syndicate already runs rather than replacing them.

Contract Builder produces the contractual artifact itself, so the platform spans submission, triage, decision and document rather than sitting at one step. Working with the shared infrastructure every participant already uses means adoption does not require a counterparty to adopt anything, which is the practical reason a managing agent can run this in production across named lines.

The platform is described as cloud based and no hosting provider, region selection, private deployment option or residency commitment was located. For a London market platform this is a smaller gap than for a vendor spanning localisation regimes, since the counterparties and the data are largely concentrated in one jurisdiction, but a global broker consuming the platform for submissions originating worldwide would still need terms that are not published. No on premise or customer environment option appears anywhere.

No pricing, packaging, basis of charge or trial route is published, and every path into the product is a contact form. Nothing indicates whether the platform is priced per user, per submission, per bound risk or as a platform fee, which matters in a market where a syndicate and a global broker would consume it at very different volumes. Category norm for London market technology rather than a specific failing, but it sits behind Akur8 in the same lane, which discloses its billing basis and offers a free pilot.

Coverage is deep rather than wide and that is deliberate. The platform serves Lloyd's managing agents and syndicates, global carriers and wholesale brokers, which is both sides of a placement, across specialty and commercial lines including marine hull, marine cargo and general aviation. The concentration is geographic: this is built for the Lloyd's and London market and its data standards, with entry into the United States stated as a plan rather than a footprint. That focus is the reason the product fits the market's rails so precisely, and it is also the limit, since a carrier outside the London placement architecture buys a much less complete product.