Send Technology

The product orchestrates other people's models, yet models still sit in its own operating path. The company frames AI as a natural evolution of what it already does at scale. What it describes orchestrates the carrier's raters, data providers, core platform and agents, with a framework for customers and integrators to build agents Send did not write. Without the models, a working underwriting workbench remains, which is what the company shipped from 2017 onward. The models are in the path rather than merely around it because Send runs its own ingestion, extraction, triage and enrichment on every submission that enters the platform.

Oversight is audited but not described. The company states that its agents are governed across their entire lifecycle with clear accountability, transparency and human oversight built in, and describes the goal as autonomy without loss of control. An accredited auditor examined the management system behind that claim under ISO 42001, so a human oversight design demonstrably exists. The design itself is not public.

No confidence threshold, escalation trigger or review queue is named, and there is no stated approval checkpoint or sampling control on automated output. An audited assertion that oversight exists is stronger than an unaudited one and still weaker than a named mechanism with a stated position in the execution path.

The ISO 42001 certification is audited rather than asserted. An accredited body verified that the AI management system extends compliance and audit processes to cover governance and risk. It also verified that tools are monitored and continually improved rather than deployed once and left. The stated purpose is customer facing, giving insurers the governance, auditability and control their risk and compliance teams expect.

The limit is that a management system certification attests to the process around the models. No evidence about the models themselves is published. There is no accuracy figure or error rate, and no validation summary or documentation package. At least one competitor also commits in public to answering customers' own AI compliance questionnaires and shares its framework so carriers can build their deployer position.

Two independent analyst firms have evaluated the platform. Datos Insights named Send a major provider in its Underwriting Workbench Market Navigator. A senior principal there is quoted on the finding that prebuilt product models cut new product launch timelines by 65 percent, from three months to four weeks. Celent placed the platform in the top quadrant of its Underwriting Workbench reports for North America.

Argenta Group is named with a published case study as the first Lloyd's managing agent to deploy the workbench across all classes of business. More than $26 billion of gross written premium runs through the platform. The strongest evidence is structural. Duck Creek, a core insurance platform serving more than half of the top 20 global property and casualty carriers, bought the company in July 2026, which puts an independent party with money at stake behind the diligence.

Send makes a data boundary commitment in public: customer data stays in the customer's own ecosystem. The ISO 42001 scope is stated to include enhanced data privacy protections within the underwriting products, so the commitment sits inside an audited management system rather than standing alone. Still, the commitment is a slogan rather than an architecture.

Nothing states whether one carrier's submissions, enrichment history or underwriting outcomes inform models or agents serving another carrier on the same platform. That is the question that matters where competing specialty writers share an orchestration layer. At least one competitor states flatly that one carrier's data is never used to train another's system.

Exposure is reduced because commercial and specialty submissions concern businesses, their assets and their loss history rather than individuals. Three certifications touching information security and AI governance provide indirect assurance, and the ISO 42001 scope is stated to include data privacy protections inside the underwriting products. Direct disclosure is absent.

No data processing agreement or subprocessor list is public, and there is no retention schedule or lawful basis statement. Delegated authority and London Market bordereaux nonetheless carry personal data in some classes, and nothing addresses how it is treated.

Three audited certifications are held and named by the vendor with dates. ISO 27001 was first accredited in May 2022 and has since been renewed. A SOC 2 Type II report was obtained in May 2025. ISO/IEC 42001:2023, the AI management system standard, was obtained in early 2026. The chief operating officer and co founder is named as the person who led the ISO 42001 audit, which puts an individual's name behind it.

The company's own framing is that fewer than 15 percent of insurance technology providers hold the security pair. That is a marketing claim, but the certifications behind it are real and independently issued.

Send is a technology supplier and has no insurance license, which is the correct posture. The platform operates inside heavily supervised distribution, particularly delegated authority and the London Market, where binding authority arrangements carry their own regulatory obligations. Compliance management is named as a product capability. Formal standing is absent. No supervised regulator test of the product, sandbox participation or admission to a program is published. The three certifications are technical conformity assessments and do not amount to regulatory standing.

ISO 42001 requires a documented impact assessment, so the certification implies one exists. Beyond that inference nothing is published. There is no fairness testing or differential outcome monitoring, and no impact assessment summary. Commercial and specialty property and casualty weakens the protected class analysis considerably, because the insured is usually a business rather than a person. The residual exposure sits in triage. Prioritizing submissions by appetite on historical patterns reproduces whatever the book has previously favored by geography, industry class and broker relationship, and no account of that is given.

The orchestration position raises a question of its own. When a carrier deploys a third party agent inside the Send framework and that agent errs, nothing states where responsibility sits between the carrier that chose it, the agent's builder and the platform that ran it. For the extraction, enrichment and triage the platform performs itself, no accuracy figure or error rate is published. There is no remediation commitment, liability position or correction path either.

No model provider, family, version or country of processing is disclosed for the ingestion, extraction and enrichment that Send itself performs. The structure here inverts the usual problem in an interesting way and does not solve it. Because the agentic framework is built for carriers and system integrators to bring and deploy their own agents, a large part of the model supply chain is chosen by the buyer and therefore fully known to it. That transparency covers the agents the customer brought. It says nothing about the models Send runs underneath, which are the ones reading every broker submission that enters the platform.

Integration is not a feature here, it is the entire proposition. The architecture is application programming interface first, data is captured once and stays available across the underwriting lifecycle to power downstream systems, and the company states plainly that it fits an existing technology landscape without a rip and replace. It orchestrates the carrier's own raters, data providers, core platform and third party artificial intelligence agents.

A named integration exists with the hyperexponential rating platform, and following the Duck Creek acquisition the engine remains available standalone with integration to all core systems while also connecting into the Intelligent Insurance Cloud. Low code microservice construction with prebuilt product templates supports rapid configuration. The agentic framework goes further than integration by letting customers and system integrators build and deploy their own agents inside the workflow.

The deployment story is clear as far as it goes. Send is software as a service on a low code microservice architecture, pre configured for rapid deployment and configurable afterwards. Where the data sits is not part of it. No region list or residency commitment is published, and there is no stated single tenant option or self hosted path. The statement that customer data stays in the customer's own ecosystem gestures at the question without answering it. The platform serves the London Market, wider Europe and North America, jurisdictions with materially different expectations about where insurance data may be processed.

No price, tier structure, billing basis or indicative range is published, and the route to a number is a demo request. One near miss stands out. The Datos Insights finding that prebuilt product models cut launch timelines from three months to four weeks is the kind of figure a buyer would price against, and it is published with no cost beside it.

Breadth of operating model is wide inside a single market. One platform serves six distinct shapes. Commercial carriers, specialty writers and reinsurers are covered, as are managing general agents, delegated authority arrangements and the London Market. The last two are separate regulatory and operational worlds. Multi carrier and multi program complexity is explicitly supported, and a dedicated North American product line sits alongside the United Kingdom and London Market business. Life, health and personal lines are absent entirely, so coverage is deep within one half of the insurance market rather than across it.