Koios Intelligence

The assistant is the product. Olivo is a conversational agent handling quote intake, customer service, document extraction and compliance validation across voice, chat, text and email, and the later architecture is described as multi agent orchestration combined with voice AI infrastructure and intelligent document processing.

Remove the language models and nothing purchasable remains, since the integrations into broker management systems are plumbing that carries the assistant's output rather than a product in their own right. The founder's own background is actuarial science and computational finance rather than software distribution, and the company has been building this single capability since 2017.

A described division of labor rather than an unbounded agent, and both customer announcements state it in the same terms from the client side. Univesta's framing is that the agents handle quote intake and data transfer so brokers can dedicate themselves to the advisory role. Humania's is that automated compliance validation lets human reviewers concentrate on complex cases. That is triage with people retained for judgment and exceptions, which is the right shape.

The complication sits in the company's own earlier description of the assistant as a virtual intermediary that lets a consumer shop coverage options and even complete insurance purchases. Completing a purchase is a different act from preparing one. Nothing published describes what the assistant may bind unassisted, at what value, or how a consumer reaches a licensed person mid conversation.

No accuracy, error rate or evaluation methodology published for any component, and the compliance module is where that matters most. It is described as validating information consistency across multiple sources, evaluating applications against underwriting and regulatory requirements and generating risk profiling summaries, which makes it a control rather than a convenience, and an unmeasured control gives an insurer false assurance in precisely the function where assurance is the point.

The vendor's own language is that this allows audits at a scale previously impossible, and scale without a measured error rate means errors are also produced at a scale previously impossible. The unreconciled 400 percent and 37 percent conversion claims are a second reason to treat published figures here with care.

Customer evidence is strong, with several named customers and several named executives quoted. Named clients include PMA Insurance, Tardif Group Insurance, partners of Intact within the Virage Group, and Rogers Insurance, among the largest brokers in Canada. Two deployments are announced with named executives quoted.

At Univesta Insurance, one of Quebec's largest property and casualty firms, the partner and vice president for the client contact center describes AI agents processing home and auto quote requests into existing management systems. Humania Assurance, a Quebec life and health insurer, is deploying the compliance module across life insurance workflows. No quantified result is attributed to any of them. The figures that exist are vendor aggregated, and they do not agree with each other. Material from 2023 claims a potential 400 percent increase in sales conversion, while 2025 material states clients typically see 37 percent. Those are not the same claim about the same metric.

Nothing published on boundaries between clients, and the platform's position makes the question a real one. Competing brokerages sell overlapping products in the same regional markets and several sit on this platform simultaneously, while insurers using it are also the counterparties those brokers place business with. A system that sees quote requests, coverage comparisons and conversion outcomes across both sides gains a view of local market pricing and demand that neither side sees. Nothing states whether conversational data, quote patterns or conversion signals from one client inform models, prompts or defaults serving another.

Three regimes are named specifically rather than privacy asserted generally: GDPR, HIPAA and the Autorité des marchés financiers (AMF), the Quebec financial sector regulator. Deployment options across multiple cloud regions support the residency side of those obligations.

Naming HIPAA is the substantive part rather than boilerplate, because the life and health workflows mean the assistant handles medical disclosures gathered in application conversations, the most sensitive data any insurance intake touches. These are compliance claims without supporting documentation. There is no retention schedule, no processing agreement and no subprocessor list. Nothing describes how recorded conversations, which the product creates by design, are stored, for how long, or who inside the brokerage can replay them.

The compliance claims that do appear, GDPR, HIPAA and AMF alignment, are regulatory positions rather than independent attestations about the security of the environment, and the two are routinely conflated. No SOC 2, ISO 27001, penetration testing program or trust center is published. Insurers of the size named among the clients run vendor security assessments as standard, so the evidence very likely exists privately. Nothing is published for a prospective buyer to read.

The perimeter question turns on what the assistant does rather than what it stores, and no license or registration of its own is stated. In Quebec, where the company is based and most of its named deployments sit, the Autorité des marchés financiers supervises insurance distribution. Distribution conducted without a licensed representative carries specific obligations around disclosure and a consumer's ability to reach a natural person.

An assistant that presents coverage options and can carry a consumer through to purchase is operating inside that regime. Alignment with the AMF is claimed at a high level. Nothing states how the product draws the boundary between assistance and distribution, or which party holds the license for the act being performed.

There are two distinct exposures, and no governance framework, evaluation or fairness testing is published for either. The first is linguistic. The product advertises service 24/7 in the customer's language of choice across voice channels, and speech recognition and language understanding error rates vary by accent, dialect and language.

A caller in accented French or English can get a materially different quality of service from one the model handles well, in a market where language rights are a live matter. The second is advisory. An assistant presenting coverage options shapes which products a consumer considers, and nothing describes what governs that ordering or whether conversion performance influences it.

The harmed party is a consumer who may not discover the harm for years, and no warranty, service level or remedy is published. Insurance is the rare purchase where a defect stays invisible until the moment of loss. A consumer guided to the wrong coverage, whose disclosure was mis captured from a conversation, or whose application was mis validated learns about it at claim time, when the remedy question is a denied claim rather than a refund.

Complete traceability of conversations and disclosures is described, but traceability tells you what happened without telling anyone what they are owed. Nothing states where responsibility sits between the vendor, the brokerage and the insurer when the automated intake is the source of the error.

The company describes natural language processing, state of the art AI, voice AI infrastructure and multi agent orchestration without identifying a single component. No model, provider, version or speech recognition vendor is named.

The voice layer is the notable omission, because speech recognition is rarely built in house and is the component whose accuracy varies most across the accents and languages this product explicitly serves, so a buyer cannot assess the exposure or the residency of voice data without knowing whose engine transcribes it.

Integration is named, specific and strategically chosen, which is more than a claim of API availability. Olivo is interconnected with Applied, the dominant broker management system in North American insurance distribution, and with the Centre for Study of Insurance Operations (CSIO), the Canadian industry data standards body. The company built into the sector's own interchange standard rather than around it.

The Univesta deployment describes extracted data transferring automatically into the client's existing management systems and explicitly eliminating double entry, the concrete sign that an integration is real. The same assistant is reachable over web, phone, text and email. For a broker, work that does not land in the management system does not exist, and this lands there.

Deployment options across multiple cloud regions are stated. That is a residency commitment, and it is substantive for this buyer set because Quebec and Canadian financial sector expectations around where personal information sits are specific and enforced. No region, provider, tenancy model or on premise option is named. Nothing describes where recorded voice conversations in particular are held, the largest and most sensitive data the product generates.

No pricing, model, unit or minimum published anywhere. The route in is a demonstration booking. Nothing indicates whether the assistant is priced per broker seat, per conversation, per quote completed or per policy sold, and the last of those would matter considerably given the product's stated effect on conversion, since a vendor paid on conversion has an interest in the assistant's persuasive success that a vendor paid per seat does not.

Both sides of the distribution relationship and several lines. Buyers are insurance brokers and agencies as well as insurers themselves, and the stated scope covers personal and commercial property and casualty, life and health, and financial services. The company reports locations in Canada, the United States and North Africa with expansion into France. Every substantiated deployment is Canadian and predominantly Quebec based, so the multi line breadth is evidenced while the multi market breadth is asserted.