Feathery

The load bearing model work is extraction: pulling structured values out of documents, emails and voice recordings so a submission or an onboarding packet can be prefilled rather than rekeyed, which is the actual bottleneck in insurance and advisory operations and cannot be built as rules. A plain language assistant that composes and deploys workflows sits above that.

Underneath, though, is a configurable forms, logic and integration engine that is the company's origin and would still function without any of it. Apply the removal test and a no code intake and workflow platform survives, which places this with the workflow vendors rather than the model native ones.

Human control over the business process is real and built in, with approvals, routing and collaborative review as named capabilities, so a submission or onboarding packet still passes the people who are accountable for it. The unexamined layer is one level up, in how the platform itself changes.

The assistant is described as taking a plain language description and both building and deploying the workflow, which means production intake processes in regulated operations can be created by an agent with no described review, testing or sign off step between the request and live traffic. Exception flagging and next best action recommendations are advisory, which is the right posture for them.

Extraction accuracy is the number that matters most here and it is not published anywhere. A misread limit, premium, date or loss description does not stay in the intake layer; it flows into a bound policy, a claim reserve or a client account and is discovered later.

Nothing public offers extraction accuracy or exception rates by document type, confidence scoring behaviour, human verification thresholds, model documentation, evaluation methodology or a stated position on supporting a customer's own validation.

The customer roster is unusually strong for a company at this stage and is named across all three of its buyer types: global and specialty carriers alongside a life insurer, several established registered investment advisers, and three sizeable insurance brokerages. Adoption is stated at more than 300 firms processing tens of millions of submissions a month. Two insurance carriers' venture arms are investors, which means underwriters ran their own diligence.

What is entirely absent is measurement. No extraction accuracy, no straight through processing rate, no time saved, no error reduction, no case study reporting a before and after at any of those named firms.

This vendor states the cross client data question more openly than any other in the index, and that candour is exactly why it needs flagging. The company describes its next phase as building products that leverage its data network across clients and evolving its models using insights gathered from an expanding client base. Those clients include carriers that compete with one another for the same risks and advisers competing for the same households.

No public material defines what enters that network, whether it is anonymised, whether participation can be declined, or what a firm's data contributes to capabilities sold to its rivals. Buyers signing today are signing ahead of terms that have not been published.

The intake surface carries some of the most sensitive material either industry handles, including adviser onboarding packets with full personal financial circumstances, insurance applications, and first notice of loss submissions that often contain medical and incident detail. Extraction from voice recordings and email adds channels most peers do not touch.

A third party listing references enterprise grade security and compliance, but this pass located no published privacy framework, retention schedule, subprocessor list or statement of service provider obligations on the vendor's own material.

A third party directory listing references enterprise grade security and compliance including a service organisation control attestation, and that is the only assurance signal this pass surfaced. No trust centre, enumerated certification list, attestation scope or audit period was located on the vendor's own site.

Carriers and advisers of the size named as customers would require attestations contractually, so the control environment is likely stronger than the published record, and the grade reflects verifiable evidence rather than a judgement on the controls.

Feathery supplies software and holds no licence, the expected posture. Its domain grounding is better than most at this stage and shows in the vocabulary: naming book roll ingestion and delegated authority reporting intake as workflows indicates the product was built against how insurance distribution actually operates rather than against a generic intake problem.

The workflows themselves sit inside regulated processes, covering account opening and adviser transitions under securities rules and submission, binding and claims notification under state insurance supervision, with the obligation remaining the firm's.

The decisioning layer does not merely move data, it surfaces insights across a book of business and makes next best action recommendations that feed underwriting and advisory work. In insurance that engages state supervisory expectations for insurers using artificial intelligence and unfair discrimination duties that land on the carrier; in advice it touches suitability, where a recommendation engine steering an adviser toward particular actions has consequences for the client. This pass located no bias or fairness testing, no demographic analysis, no accuracy evidence for the recommendations and no description of how a firm would evidence oversight of them to an examiner.

No accuracy commitment, no remediation term and no correction route were located. The exposure is concrete rather than theoretical: extraction accuracy is unpublished, a misread limit, premium, date or loss description flows into a bound policy or a client account, and the resulting error surfaces months later at a claim. Nothing states who carries that loss, and nothing gives the adviser or insured a route to have an extracted value challenged before it becomes the record.

No model providers are named for the extraction, voice transcription or assistant layers, no subprocessor list is published, and nothing describes where documents, emails and voice recordings are processed. The stated intention to build products on a data network spanning clients makes the omission sharper, since the same undisclosed chain would carry material from carriers, brokers and advisers who compete with one another and have no published account of what separates them.

Integration targets are the right ones and reflect real familiarity with these operations, spanning customer relationship systems, custodians, policy administration platforms, agency management systems and internal databases, with custom interface connections available where a workflow needs something specific.

Input coverage is broader than peers, taking documents, email and voice recordings rather than forms alone, and output extends to document generation and electronic signature so a workflow closes end to end. Public developer documentation, a status page, a changelog and a named partner directory were not located in this pass.

Delivery is cloud hosted software as a service. Customers include a global carrier and a specialty insurer with international operations, which makes residency a live question rather than a theoretical one, and no public material identifies hosting regions, residency options, transfer mechanisms, tenancy separation or subprocessors. Tenancy separation deserves specific attention here given the stated intention to build products on a cross client data network.

No rates, tier structure, billing unit or minimum was located in this pass. The pricing dimension a buyer would most want is implied by the product and unstated: the platform is sold on customers running dozens of workflows each and processing submissions at volume, so whether charging follows workflows, seats, submissions or extracted documents changes the economics entirely, and none of it is published.

Segmentation is genuine rather than cosmetic, with distinct workflow sets enumerated for three different buyers: advisers and broker dealers get client onboarding, account opening, proposal generation and adviser transitions; carriers get submission intake, book roll and delegated authority reporting ingestion, first notice of loss and portfolio analysis; brokers get agency system data entry, proposals, policy checking and benefits documentation. That is domain knowledge, not a vertical page. The boundary is equally clear: insurance and wealth management only, with no material for banks, credit unions, lenders, payments or capital markets.