Avantos

The architecture is a knowledge graph linking client data, products, service teams, workflows and service expectations, with agents described as participating directly in workflows to prepare onboarding, coordinate across teams, identify opportunities, monitor progress and execute work across systems. The removal test returns a product because the graph is not itself a model.

Strip the agents and a unified client context layer remains, which is precisely what the company says the market lacks and what frees staff from navigating systems and moving data between platforms. The models make that context act; the context is the asset. Same shape as JIFFYAI and Nevis, where the unification is the substance and the agents are the interface to it.

One explicit boundary statement carries this grade: humans stay in control, with staff freed from navigating systems and moving data between platforms rather than freed from the decisions themselves. That is a coherent and defensible line, and the described agent duties sit on the correct side of it, preparing onboarding, coordinating handoffs, monitoring progress and moving work between systems.

Against it, agents are also described as executing work across systems and identifying opportunities, and no approval gate, confidence threshold, escalation path, audit trail or sampling regime is described anywhere. The distinction between an agent that moves data and an agent that commits an action in a custodial or policy administration system is the one that matters most, and public material does not draw it.

No accuracy figure, validation evidence, error analysis, model documentation or confidence exposure was located. The knowledge graph is described as establishing context, and the correctness question for a context layer is whether the relationships it asserts between clients, products, teams and obligations are right, because every downstream agent action inherits them.

A wrongly linked product, a stale service expectation or a mis attributed team ownership propagates into onboarding preparation, progress monitoring and opportunity identification at once. Nothing published describes how the graph is validated, how conflicts between source systems are resolved, or how a firm would detect that context had drifted from reality.

No financial institution is named as a customer anywhere and no volume, outcome or deployment figure exists. What is named is the capital: 35 million dollars total, a 25 million dollar Series A led by Bessemer Venture Partners and a 10 million dollar seed led by the MIT affiliated E14 Fund. The Norm Ai principle governs the reading and it matters more here than almost anywhere in the index, because the register is unusually buyer side.

Guardian Life, SEI and Vanguard came in as strategic investors and Mercer Advisors participated in the seed, which is an insurer, a wealth platform, an asset manager and a registered investment adviser all backing the company. Read carefully, though: SEI's chief product officer is quoted saying Avantos will be a catalyst for how the firm serves clients, in the future tense, which is an investment statement rather than a production reference. Claims that strategic partners report improved productivity are unattributed and unquantified. Prestige on the cap table describes who is interested, not whether the product works.

Nothing published defines a data boundary. The platform is described as a continuously updated system of context, which implies the graph learns and enriches over time, and no statement establishes whether that enrichment is contained to one institution's tenant, whether client data is used to train or tune any model, or whether service patterns observed at one firm inform anything served to another.

That question has weight for a product whose named strategic backers include firms that compete with each other in wealth and retirement. The benchmark answers to grade against are Rulebase, which forecloses training on customer data in one line, and DwellFi, which states that each agent's learning is contained to the customer's tenant.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The payload deserves an account that does not exist, because the whole design is to consolidate what is currently fragmented.

A knowledge graph spanning wealth, protection and retirement for the same client holds portfolio holdings, beneficiary designations, retirement balances and, where the platform reaches into underwriting systems as planned, the health and lifestyle information life and disability underwriting depends on.

Assembling all of that into one continuously updated context object about a named individual creates a concentration of sensitivity that none of the underlying systems carried alone, and nothing published addresses how it is governed, how long context persists or what a client can ask to be removed.

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. For a platform proposing to hold consolidated client context for wealth, protection and retirement relationships, this is the first item any institution's third party risk programme would request, and the absence is more conspicuous given the calibre of the strategic investors already engaged. Compare Zeplyn in the same lane, which names an independent attestation alongside its encryption and access controls.

Three distinct regulatory regimes are implied by the stated scope and not one is named. Wealth servicing sits under investment adviser and broker dealer supervision, protection sits under state insurance regulation with its own suitability and best interest standards, and retirement carries fiduciary duties under federal pension law.

A platform proposing to unify onboarding and servicing across all three is operating where those obligations meet, which is the hardest regulatory seam in the wealth market and the one a buyer would most want addressed. No supervisor, statute, rule or guidance instrument appears anywhere, and no account is given of how servicing actions taken by an agent enter a firm's supervised books and records. Avantos holds no licence and needs none, which is the correct posture for a technology supplier under the index convention.

The client health module is the exposure and it is not addressed. Scoring relationships in real time to surface risks and identify upsell opportunities means a model decides which clients receive proactive attention and which products are put in front of them. Two consequences follow. Attention is allocated by score, so a client the model rates as low health or low opportunity receives less human service, and the client never learns a score exists.

Product identification in protection and retirement runs directly into suitability and best interest obligations, where a machine surfaced opportunity that anchors an adviser toward a particular product is the same anchoring risk recorded for Elysian in claims. No testing, no analysis of how health and opportunity scores distribute across client segments, and no statement of what inputs drive them.

No guarantee, indemnity, falsifiable commitment or correction path was located. The single mechanism keeping this off the floor is the stated posture that humans remain in control, which places an identifiable person between an agent's work and any consequence to a client, but no audit trail, reasoning disclosure or confidence exposure is described to make that oversight effective.

The party with no route is the client, who is scored for relationship health and opportunity, whose service priority is set by that score, and who is never told an assessment exists or given any way to see or contest it. This is the recurring shape in the index, and it is quieter here than in credit or fraud because the harm is service a client never receives rather than an action taken against them.

The architecture is described and the parties are not. A knowledge graph with agents executing work across systems necessarily involves both external model providers and connected third party platforms, and neither is named anywhere. No foundation model provider, hosting arrangement, subprocessor or data source is disclosed, and nothing states whether client context assembled in the graph passes to an external model provider when an agent reasons over it. That is the fourth party question an examiner asks, and for a platform whose whole purpose is concentrating client data it is the version of the question that matters most.

This is the sharpest gap in the profile because integration is the entire thesis. The product exists to replace fragmented legacy platforms with one context layer, and the company states that Series A capital will be used to fund deeper integrations with custodians, customer relationship systems, portfolio management tools, underwriting systems and policy administration platforms. That is future tense.

Not one custodian, customer relationship system, portfolio platform, underwriting system or policy administration platform is named as currently connected, and no developer documentation or application programming interface reference was located. A product whose value proposition is unifying systems, whose connections to those systems are a funded roadmap item, is at the stage where a buyer is purchasing the thesis. Contrast Nevis and UPTIQ, which both name specific platforms on both sides of the wealth and banking stacks.

No hosting model, cloud provider, region selection, residency commitment or private deployment option was located. The consolidation design makes the question sharper rather than routine, because a continuously updated knowledge graph is by construction a persistent copy of client context assembled from systems of record the institution already controls, and where that copy lives is a question a wealth firm's vendor review asks before anything else. Nothing published answers it.

No pricing, packaging, basis of charge, pilot route or trial is published and every path into the product is a contact request. Nothing indicates whether charging is per seat, per client relationship, per module or as a platform fee, which is a live question for a product positioned as an operating system spanning four business lines rather than a point tool. Category norm rather than a specific failing.

Four business lines are named as in scope, wealth, protection, retirement and banking, and the buyer types described are asset managers, insurers, advisory firms and financial institutions. The strategic investor set corroborates that the segments are genuine targets rather than aspiration, since an insurer, a wealth platform, an asset manager and a registered investment adviser all took positions.

What holds this at B is that no institution of any type is named as a deployment, so the breadth is evidenced by who is interested rather than by who is running it. Spanning wealth, protection and retirement in one platform is a real ambition, since those three sit in different systems, different service models and different regulatory regimes inside the same firm.