Herald

Carrier connectivity came first and is still what the company is bought for: 35 carriers, 80 products and a data standard, none of which requires a model. Strip the models and a substantial working product remains. AI arrived by acquisition in July 2025, when Herald bought Babbix and turned it into an AI division. It is real rather than announced: an agent in the broker's inbox extracts client data, resolves gaps at intake and assembles submissions. Models sit in the operating path for brokers who use the agent workflow, and nowhere at all for those consuming the raw interface.

An agent reads a broker's inbox, extracts client information and sends submissions to carriers. Sending a submission is a market facing act performed on behalf of a client, and it is where a broker's professional duty attaches. Nothing public describes an approval step before dispatch, a confidence threshold on the extracted fields or a review queue. Nor does anything say what happens when the agent has read a figure wrongly and the submission has already gone. The automation is real and the control around it is undescribed. That is an absence of a stated mechanism, not evidence that the automation is unsafe.

The company says the capability removes the burden of human input. That states the objective without any account of how the output is checked once the human is gone. Nothing is published on validation, accuracy, monitoring or governance of the extraction and agent capability. There is no error rate for reading a PDF or a spreadsheet into a submission, and no benchmark, certification or published program. The gap weighs more than it would for a model that only recommends, because here the output leaves the building.

Herald says it works with eight of the 25 largest brokerages and names none of them. If accurate, that is significant penetration of the top of the market. One named customer voice exists: a national cyber practice leader at the wholesaler Wholesure, quoted on the Babbix acquisition. Counterparty scale is specific and checkable in a way the customer list is not, at more than 35 carrier partners and more than 80 products.

Investors include Lightspeed Venture Partners and Brewer Lane Ventures. Outcome evidence is absent. There is no quote turnaround figure, bind rate or extraction accuracy, and no time saved or case study with numbers.

Herald has a powerful data position and does not describe how it treats it. Sitting between eight of the largest 25 brokerages and more than 35 carriers gives it a view of submission flow on both sides of a market: which risks are shopped where, at what point in the cycle, and by whom. Nothing states whether that aggregate view is used. Nor does anything say whether one brokerage's flow informs what a competing brokerage is served, or whether carriers can see what reaches rivals. The stated ambition to set an open data standard makes the question larger, because a standard setter is also the party best placed to read everything that flows through it.

The privacy surface is wider than the product description suggests, because the agent works inside a broker's email inbox rather than a bounded application. An inbox holds the client correspondence the agent is there to read and everything else the broker receives. Nothing describes what is accessed or retained, or how material outside the scope of a placement is handled.

Commercial insurance data mostly concerns businesses rather than consumers, which lowers the exposure without answering the inbox question. No data processing agreement, subprocessor list, retention schedule or lawful basis statement is published.

The site says the platform carries enterprise security certified for the most sensitive and valuable data, and names no certification. Asserting a certification without naming it is weaker than saying nothing. It invites a buyer to assume a report exists that they cannot check, request or verify. No SOC report, ISO standard or trust center is published. Nor is there a penetration testing statement or a subprocessor list. Service level objectives are offered as tailored to each customer, a commercial commitment rather than a security one.

Herald is a technology supplier and has no insurance license, which is the correct posture for its shape. It publishes no statement of regulatory position in any form. There is no supervised test, program admission or supervisory engagement. Placement itself is a licensed activity, and the platform reaches surplus lines products where filing and diligence duties attach to the producer. An agent that assembles and dispatches submissions works inside that regulated act, and the vendor does not address it.

The placement system is marketed on its ability to move business toward a brokerage's top carrier partners, which sits close to a live conduct question. Steering placement by carrier relationship rather than client fit is the conflict that contingent commission rules and broker duty of care exist to police. A model that ranks or routes submissions with that objective encoded makes a distribution decision on the client's behalf.

Nothing public addresses how client interest and carrier preference are balanced, or whether the broker sees which one drives a recommendation. No fairness testing, governance program or impact assessment is published.

No error rate, remediation commitment, liability position or correction path is published, and the consequence chain is long. An agent extracts figures from client material and submits them to a carrier. Suppose an exposure, payroll or revenue figure is read wrongly and the policy is bound on it. The insured discovers the problem at claim time, when a misstatement in the application is exactly the ground on which cover is disputed.

The broker carries the professional liability, the carrier the underwriting consequence, and the insured the loss. Nothing states where responsibility sits or how a misread field is caught before bind.

No model provider, family, version or country of processing is disclosed for the extraction and agent capability. The artificial intelligence line was acquired rather than built, which normally makes the supply chain easier to describe, and no description is offered. A broker cannot determine whether client information pulled from an inbox and assembled into a submission passes through a third party model provider on the way to the carrier.

Integration is the company rather than a feature of it. More than 35 carrier connections carry more than 80 products across six named lines of business. The interface is designed to drop into whatever the brokerage already runs: an agency management system, a CRM or a platform built in house. A public API reference sits on its own documentation subdomain. The ambition goes past integration into standard setting.

Herald describes itself as building an open data standard for broker to carrier transactions, aimed at the duplicative entry and inconsistent formats that make commercial placement slow. An email inbox is treated as a first class integration surface alongside the systems, an accurate read of where brokers actually work.

Delivery is hosted infrastructure, with service level objectives described as tailored to each customer and tested at production scale. There is no published region list, residency commitment, single tenant option or self hosted path. The footprint appears to be United States only, which narrows the exposure and suggests domestic processing. Inference is not disclosure, and nothing states it.

Pricing is not published for either the connectivity interface or the agent: no tier structure, billing basis or indicative range. The route to a number is a contact request. The gap stands out for an infrastructure product sold to developers. Per call or per transaction pricing is the usual convention there, and a public reference already exists for everything except the cost.

Herald serves both sides of the distribution relationship. Brokerages, wholesalers and managing general agents sit on one side, carriers seeking digital distribution reach on the other, and software platforms embed the connection between them. Six named lines describe small and mid commercial business rather than the whole market. They are cyber, management liability and professional liability, plus business owners policy, general liability and workers compensation. The footprint is United States only. Line coverage stops short of property, marine, casualty towers and the specialty classes, and there is no life or health exposure at all.