Sixfold

The removal test leaves nothing behind. The product is an underwriting agent: it reads the submission, evaluates appetite and portfolio fit, recalls what it has learned about the broker and the risk class, and produces a reasoned recommendation. Take the models out and there is no workbench, no policy administration system and no rules engine underneath, because Sixfold deliberately does not supply any of those and integrates into the carrier's existing ones instead. The company describes itself as an underwriting brain rather than a system of record, and the architecture matches the description.

The oversight architecture is real but sits at the governance layer rather than in the execution path. A named and staffed ethics and responsible artificial intelligence function reviews new functionality quarterly and its explicit remit is deciding where humans should stay in the loop and evaluating human oversight design. That is more than most vendors offer.

In normal operation the agent produces a recommendation, explains its rationale and suggests a next action, leaving the underwriter to decide. What holds this off the top grade is the configurable straight through path: the agent can be set to carry cases all the way to quote ready and bind ready materials, and nothing public describes which cases qualify, what threshold or confidence measure governs the switch, or what sampling control checks the automated decisions afterwards. Federato sets the bar on this axis by publishing three named automation modes with a random audit on the fully automated tier, and no equivalent is published here.

The standard this A clears is the one the index has been asking about since the first cohort: does the vendor help the institution validate the model rather than simply pass someone else's test. Sixfold runs a formal AI governance programme anchored to the NIST AI Risk Management Framework version 1.0 and to the high risk provider requirements at articles 9 through 17 of the EU AI Act, cited by article.

The programme has five named focus areas, one of which is data and model governance covering data provenance, ethical use, and being able to explain how system behaviour relates to the inputs that shaped it. Owners are named, one of them publicly. The review cadence is published: programme owners meet fortnightly, risk and compliance runs quarterly reviews, the ethics function runs quarterly reviews of new functionality, and there is an annual retrospective.

Three successive annual Responsible AI reports have been issued, the 2024 edition available as an ungated document, which makes this a maintained and versioned artifact rather than a one off. Most decisive for this axis, Sixfold states that it gives customers documented answers to vendor artificial intelligence compliance questionnaires and shares its framework so carriers can build their own deployer position. Recorded as forward looking rather than delivered: the company says it is building toward a complete annex four technical documentation package and has not yet published one.

Named carriers across three continents and several buyer shapes: Zurich North America, New York Life, Guardian, Generali Global Corporate and Commercial, AXIS, Skyward Specialty and Mosaic. Customers are stated to represent roughly 270 billion dollars in gross written premium. More than 1.5 million submissions processed across more than 50 lines of business since launch in 2023, with more than 100 billion tokens processed.

Quantified outcomes are published and are unusually specific: processing times improved by between 50 and 97 percent, hit ratios up by 15 percent or more, gross written premium per underwriter up by as much as 30 percent, and Zurich underwriters saving up to two hours per submission. Adoption is reported at 89 to 90 percent measured as actual users against expected users, which is the figure enterprise artificial intelligence deployments usually avoid publishing.

Sixfold answers in one sentence the question this index has put to Clearwater, Feathery, Cytora and every other multi tenant platform and almost never had answered. Each insurer's book is walled off, and one carrier's data is never used to train another carrier's system.

That commitment matters more here than in most categories because the institutional memory feature is built precisely on feeding every underwriting decision back into the deployment, so without the boundary the product would be pooling competing carriers' risk selection judgement.

The governance programme carries a separate ethics and responsible artificial intelligence function whose stated remit includes deciding where the company's position is on where the technology should and should not operate, which is a rarer commitment than a fairness statement.

A privacy policy, a terms of use and a security commitments page are published, and security and privacy is a named governance focus area. Beyond that the public record is thin: no data processing agreement, subprocessor list, retention schedule or lawful basis statement was located.

That gap carries more weight here than for most vendors in this index because the life and health product line places medical and health information inside the processing path, which is the most sensitive data class any insurance vendor handles, and nothing public describes how it is treated differently from a commercial property schedule.

The commitment that each carrier's book is walled off is a real architectural fact and is credited on the stewardship axis, but it addresses competitive separation rather than the rights of the individual whose health information is being read.

A live trust centre operates at a dedicated subdomain and is described as continuously monitored, and a separate security commitments page is published and linked from the site footer. Security and privacy is one of the five named focus areas in the governance programme, with responsibility for the underlying infrastructure meeting the requirements of what is built on it.

Graded at B rather than higher because the specific certifications were not enumerated in the material reviewed in this pass, and the index rule is that a credential must be seen in the vendor's own material before it is credited. This follows the Socure precedent, where a hosted trust centre that does not render its framework list publicly earned the same grade. Worth confirming on a later pass, since a vendor selling to Zurich, New York Life and Generali would ordinarily hold a service organisation control report.

Sixfold is a technology supplier to carriers and holds no insurance licence, which is the correct posture and is not penalised on this axis. What earns the grade is the clarity of the stated position, which is unusual in this index. The company classifies itself explicitly as a provider under the EU AI Act and its customers as deployers, cites the high risk provider articles by number, and says it supports customers with the deployer obligations that fall on them.

It also states an intention to work with the National Association of Insurance Commissioners. Held off the top grade by the standing index ruling that engagement and self classification are not admission: no supervised regulator test of the product, no sandbox participation and no formal programme enrolment was located, which is what separates a B from an A here.

Well ahead of an axis where most of the index sits at C. Bias and fairness testing is a named component of the governance programme with methodology tailored to each line of business, and a dedicated public page addresses it. The company identifies algorithmic discrimination requirements as bearing hardest on its life and health customers, which is the correct read: that is where protected class exposure in insurance is sharpest.

Its chief executive has appeared publicly alongside the deputy insurance commissioner for property and casualty in Colorado, the state with the most developed quantitative testing regime for insurer use of external consumer data. Held off the top grade because methodology is published and results are not.

No demographic outcome figures, no differential rates and no independent audit summary appear anywhere located, so a carrier inheriting testing duties under the state regulator model bulletin still cannot see how the system performs across groups.

Public material stops short of what happens when the agent is wrong. No accuracy or error rate for the recommendations is published, and there is no remediation commitment, liability position or correction path. The gap is sharpest on the configurable straight through path, where quote ready and bind ready materials are produced without a described human checkpoint, and sharper still on the life and health side, where a wrongly assessed application affects an individual's ability to obtain cover.

The governance programme addresses whether harm could occur, which is a control before the fact rather than recourse after it. The applicant never contracts with Sixfold and has no stated route to challenge an assessment.

No model provider, model family, version or country of processing is disclosed anywhere located. The company reports processing more than 100 billion tokens, which indicates large language models are central, and the data and model governance function is stated to cover where data comes from, but that provenance commitment is described as internal practice and no supplier is named publicly.

Salesforce Ventures and Bessemer are investors rather than suppliers and should not be read as a disclosure. For a product that reads broker submissions, loss runs, financials and, on the life and health side, medical information, a buyer cannot determine from public material whether that content reaches a third party model provider.

The integration posture is a deliberate strategy rather than an afterthought. Sixfold does not ask carriers to replace anything and instead runs inside the underwriter's existing workbench, policy administration system, workflow and email inbox, which the company credits for its speed of adoption. Guidewire took a strategic investment position in the January 2026 Series B, which is a meaningful signal about policy administration reach given Guidewire's position in property and casualty.

Adnovum leads technical deployment across Germany, Austria and Switzerland. Held off the top grade because the policy administration and workbench systems are described as a category rather than enumerated. Federato holds the bar on this axis by naming Guidewire, Duck Creek and Sapiens as native integrations.

Customers are live across North America, South America, the United Kingdom, Europe and Australia, and a regional implementation partner covers the German speaking markets, so the platform is demonstrably multinational. What is missing is any statement of where data sits. No region list, residency commitment, single tenant option or self hosted path was located.

The walled off book commitment implies logical separation between carriers but says nothing about geography, and several of the markets served impose insurance data localisation requirements that a buyer would need answered before signing.

No pricing, tier structure, billing basis or indicative range is published, and the only route to a number is a contact form. This is the index norm and is measured against Sumsub, which publishes per verification rates on a public page.

Benefit claims are published in detail while cost context is absent: processing time improvements of 50 to 97 percent and gross written premium per underwriter rising by up to 30 percent are stated with no indication of what the platform costs to obtain them, so a carrier cannot construct a return calculation from public material.

Both major insurance halves are served with dedicated product lines: property and casualty, and life and health. More than 50 lines of business are covered. Buyers include carriers, managing general agents and reinsurers, and the named client list spans global composites, specialty writers and life carriers. Live across North America, South America, the United Kingdom, Europe and Australia, with a dedicated implementation partnership through Adnovum covering Germany, Austria and Switzerland.