Jack Kudale, Cowbell’s founder and chief executive, put a number on the shift on July 28, 2026: OMNI, the AI-native decision intelligence system now running beneath the carrier’s excess and surplus cyber book, compressed product deployment from roughly eight months to six weeks and contributed to a 53% increase in new business (Cowbell, July 2026).

That compression is the actual story, more than the launch itself. A carrier that can ship a new product in six weeks instead of eight months can also ship a mispriced one in six weeks, and the E&S segment OMNI runs on, high-volume, lower-premium small and midsize business, is precisely the part of the cyber book where loss history is thinnest relative to exposure growth. Cowbell says human underwriters retain final decision authority over every OMNI recommendation, so the question this piece pursues is not whether the system sets rates. It is whether the internal controls that review those rates can move as fast as the product does.

What OMNI Actually Runs Underneath the Book

OMNI is built on what Cowbell calls the Cowbell Platform, and the company describes it as combining decision intelligence, orchestration, and governance into a single AI-native operating model rather than a bolt-on assistant. Specialized bi-directional AI agents and small language models continuously gather submission information, analyze risk signals, assess appetite, and orchestrate underwriting workflows across intake, claims, and cyber services, generating what Cowbell calls decision-ready recommendations at each step. A governance layer called Bellwether provides observability into how those agents are performing, the closest thing in the release to an internal audit trail for the AI itself.

Rajeev Gupta, Cowbell’s co-founder and chief product officer, drew a sharp line between OMNI and the assistant-style tools that dominated carrier AI announcements through 2025: “Insurance doesn’t need another chatbot” (Cowbell, July 2026). His point is that OMNI orchestrates decisions across the full specialty insurance lifecycle rather than answering questions inside one workflow screen, which is a materially different governance object than a copilot an underwriter can choose to ignore. The company’s own language on human oversight is explicit and worth quoting directly: human underwriters retain final decision authority, so that expert judgment, transparency, and governance remain central to the process. That framing keeps the legal responsibility for a bound policy exactly where it has always sat, with the underwriter and the carrier’s actuarial function, not with the agent that drafted the recommendation.

The first major application, per the release, targets non-admitted quotes for small and midsize organizations, a segment where Cowbell says its risk pool now spans more than 55 million entities worldwide. Quotes that once took days or weeks are delivered in minutes. That speed is the product Cowbell is selling to brokers, and it is also the mechanism behind the 53% new-business figure: faster, more complete quotes win broker flow in a market where E&S placements move on responsiveness as much as price.

The Filing Gap That Makes Six-Week Deployment Possible

The reason Cowbell can compress a release cycle this far without tripping a regulatory gate is structural, not technological. Surplus lines carriers write on a nonadmitted basis, and nonadmitted property and casualty business is largely exempt from the prior-approval rate and form filing that governs admitted markets (NAIC, Surplus Lines). An admitted personal auto or homeowners carrier proposing a 53% volume shift in six weeks would be filing rate pages and actuarial memoranda with a state regulator well before launch. A surplus lines carrier faces no equivalent external checkpoint on rate adequacy before it binds. That is exactly what makes an eight-month-to-six-week compression possible in E&S in a way it would not be in most admitted lines, and it is also why the internal actuarial control, not a state rate analyst, becomes the only brake on a book that can now reprice and rebuild products at agent speed.

That control point matters more, not less, once decision-ready recommendations move at the pace OMNI describes. A rate indication behind an E&S cyber product still needs a credible loss-cost trend, a defensible ILF curve for excess layers, and a documented view of how the model’s learned appetite maps to the carrier’s actual bound experience. None of that changes because the deployment pipeline got faster. What changes is the cadence at which those checks have to run to keep up with a product cycle that used to give an actuarial team eight months of runway between concept and bind, and now gives it six weeks.

A Credibility Problem Growing Faster Than Loss History

The book OMNI targets is where the credibility problem is sharpest. Small and midsize enterprises account for roughly 98% of cyber claims by volume but only about 49% of total incident cost, according to NetDiligence’s fifteenth annual Cyber Claims Study, which analyzed more than 10,000 claims from 2020 through 2024 (NetDiligence, 2025). The five-year average incident cost across that dataset ran $246,000 for SMEs versus roughly $10.3 million for large enterprises, and SME severity climbed nearly 30% year over year in the most recent study. High claim frequency at low average severity is exactly the mix that should build credibility fastest in a loss-cost model, but only if the mix of business being written stays stable long enough for the triangles to mature. A book growing 53% in new business is, by definition, diluting its own seasoned experience with fresh, undeveloped exposure at the same time the underlying agents are learning what to write.

Pricing conditions in the wider cyber market compound the problem. Surplus lines carriers now write nearly two-thirds of all U.S. cyber insurance premium, and their incurred loss ratio reached almost 56 in 2025, worse than the roughly 50.2 posted by admitted carriers, according to AM Best’s market segment report on the U.S. cyber sector (Insurance Journal, July 2026). Total cyber premium was essentially flat at $7.5 billion in 2025, and the market has now logged eight consecutive quarters of pricing declines through the first quarter of 2026. Third-party claims, the slower-developing liability side of a cyber loss, grew roughly 30% and are adding tail uncertainty to loss patterns that were already hard to pin down. “As long as pricing continues to decline, insurers will have difficulty reversing the increasing loss ratio,” AM Best’s analysts wrote in the report. S&P Global Ratings, by contrast, expects premium increases of 15% to 20% in 2026 as claim severity and AI-driven attack costs push the cycle the other way. A carrier accelerating new-business volume into a segment where the pricing signal itself is unsettled, softening by one measure and expected to harden by another, is exactly the environment where rate-adequacy review needs to tighten, not relax, even as the product pipeline speeds up.

Governance on the Record

Regulators are not standing still on any of this. As of mid-2026, roughly 25 states have formally adopted the NAIC’s AI model bulletin, which requires insurers to maintain a written AI systems program covering governance, risk management, and internal controls over predictive models, and the NAIC’s Big Data and Artificial Intelligence Working Group has an AI Systems Evaluation Tool in pilot across a dozen participating states, with broader adoption anticipated at the 2026 Fall National Meeting (NAIC, Artificial Intelligence). Colorado has gone further on the mechanics of testing. Its Division of Insurance built a quantitative testing regime for algorithms and predictive models under SB21-169, and in 2025 it expanded the underlying governance regulation beyond life insurance to cover private passenger auto and health benefit plan insurers that rely on external consumer data (Colorado DOI, SB21-169). Cyber E&S is not yet inside Colorado’s scope, but the regulatory direction, board-level oversight, a documented risk management framework, and quantitative testing for unfair discrimination, is the same direction Cowbell’s own Bellwether layer gestures toward when it promises observability into agent performance. The gap between the two is that Bellwether is a proprietary internal tool built and controlled by the carrier that also profits from the volume it enables, while the NAIC and Colorado frameworks are external requirements a regulator can examine. An insurer’s own observability layer is a necessary first step, not a substitute for the kind of documented, third-party-auditable validation record that state AI systems programs increasingly expect, particularly for a book expanding new business by more than half in a single reporting period.

Build the Operating Model or Buy the Layer

Cowbell’s approach sits at one end of a spectrum carriers are choosing across in 2026. OMNI is a proprietary, end-to-end operating model built in-house and wrapped around the carrier’s own book, claims process, and cybersecurity services, with governance (Bellwether) built as part of the same stack. That is a different architecture from Sixfold’s AI Underwriter, which launched in June 2026 across six named carriers representing $270 billion in gross written premium and reported hit-ratio gains of 15% or more and up to 30% higher gross written premium per underwriter (The Insurer, June 2026); Sixfold sells a walled, carrier-specific fine-tuned layer that plugs into an existing book rather than replacing the carrier’s operating model wholesale, as actuary.info covered at launch. Duck Creek’s Agentic AI Platform takes a middle path, a five-layer orchestration system sold as separately priced applications that carriers can adopt incrementally rather than committing to a single vendor’s full stack, as detailed in actuary.info’s comparison of P&C AI vendor architectures.

SystemArchitectureReported outcomeGovernance layer
Cowbell OMNIProprietary, end-to-end, built in-house53% new-business growth; 8-month to 6-week deploymentBellwether (internal, carrier-controlled)
Sixfold AI UnderwriterWalled, carrier-specific fine-tuned layer15%+ hit-ratio gain; up to 30% GWP per underwriterCarrier-run validation against ASOP No. 56
Duck Creek Agentic AIFive-layer orchestration, separately priced modulesIncremental adoption across underwriting and claimsVendor-supplied, module-level

The build path Cowbell chose has an advantage the plug-in vendors cannot match: because OMNI is native to the carrier’s own platform rather than layered onto a legacy underwriting workbench, its governance instrumentation can, in principle, capture every decision at the point it is made rather than reconstructing an audit trail after the fact. The cost of that advantage is that Cowbell owns the entire validation burden itself, with no vendor to share the documentation, model-risk testing, or regulatory defense of the pricing logic underneath a 53%-larger book. A carrier that buys a layer like Sixfold’s or Duck Creek’s at least inherits some vendor-side testing infrastructure and a body of cross-carrier deployment experience; a carrier that builds an AI-native operating model in-house is answerable for every part of it, from the small language models parsing submissions to the recommendation logic an underwriter signs off on in minutes.

What the Actuarial Function Owns Now

None of this requires OMNI to be setting rates on its own to matter to the actuarial function, and Cowbell’s public language is careful not to claim that it does. What it requires is a recognition that when submission-to-bind time collapses and product-launch cadence quickens from months to weeks, the rate-adequacy review, the ILF and trend selections behind an E&S excess layer, and the periodic comparison of what the agents are recommending against what the carrier’s actual bound experience is producing all need to run on a cadence that matches the product, not the other way around. In an admitted line, a state rate analyst provides an external forcing function for that cadence, however imperfect. In surplus lines, there is no equivalent gate, which puts the full weight of rate-adequacy discipline on internal actuarial and model-risk functions precisely when the volume of business flowing through the model is growing fastest.

The near-term test is not whether OMNI’s recommendations look reasonable in isolation. It is whether Cowbell, or any E&S cyber carrier following the same build-fast playbook, can show a documented, dated record of rate-adequacy review that kept pace with a six-week product cycle rather than trailing it by a quarter or two once the loss triangles mature enough to reveal whether the SME growth was priced correctly the first time.

Further Reading

Sources