AI Insurance Distribution: Build vs Buy
Insurers distributing through ChatGPT, Claude and Gemini have two real paths: build the integration in-house, or buy compliant middleware from a specialist. Most Tier 1 and Tier 2 insurers land on buy, because compliance and multi-platform maintenance are harder than the initial build. Below is a neutral breakdown of both, plus the AI-native companies worth knowing as context, including where Marrow, WaniWani, Corgi, Kinro and boost.ai fit.
What are the AI-native approaches to insurance distribution?
Four distinct models exist in the market today, and they are not interchangeable.
Distribution middleware.
A specialist connects an insurer's existing pricing, underwriting and policy admin systems to AI platforms, and carries the compliance burden of doing so. Marrow and WaniWani both sit here, with different scope: Marrow builds for insurance only, WaniWani builds across insurance, budgets, HR and travel. Either way the insurer keeps its own brand, underwriting and regulatory status; the specialist handles the plumbing and the guardrails.
AI-native carriers and MGAs.
A company is built from the ground up to underwrite and sell insurance itself, using AI to compress the quote-to-bind cycle from days to minutes. Corgi sits here. This is not something an existing insurer can buy: Corgi is the seller, not a vendor.
AI-native agencies and brokers.
A company builds a chat-first, always-on layer for shopping and buying insurance, routing to a panel of carriers rather than underwriting itself. Kinro sits here. Also not a plug-in vendor for an existing insurer, but useful as a reference for what an AI-native front door looks like.
In-house virtual agent platforms.
A vendor gives an insurer the tooling to build its own conversational AI, deployed on the insurer's own website, app or contact centre. boost.ai sits here. This is the "build" path in practice for most insurers: rather than coding a chatbot from scratch, they license a platform like boost.ai and build on top of it. It's a genuine alternative to Marrow for first-party customer service automation, but it doesn't get an insurer into ChatGPT, Claude or Gemini's app ecosystems; that's a separate integration Marrow is built specifically to handle.
The first and fourth models are the real build-vs-buy decision for an insurer, and increasingly overlap: an insurer that already has boost.ai for its own site and contact centre still needs something like Marrow to show up inside ChatGPT itself. The second and third are more useful as landscape context: they show what "AI-native" looks like when a company has no legacy systems to work around.
Should an insurer build its own ChatGPT app?
Building in-house means engineering the integration into OpenAI's Apps SDK, and separately into Claude and Gemini's own app frameworks, since none of the three share a standard yet. It means maintaining that integration as each platform updates its rules, and building the compliance layer yourself: grounding every quote in real pricing data, surfacing disclosures, enforcing conduct rules as the conversation happens, and keeping an audit trail of what the AI said and why.
This tends to make sense when an insurer already has a strong in-house AI and platform engineering function, wants to own the compliance decisions itself instead of delegating them, and has the scale to justify dedicated, ongoing headcount across three, and growing, separate app ecosystems.
It tends not to make sense when speed to market matters, when engineering capacity is already stretched by core system modernisation, or when the compliance risk of an under-resourced in-house build is higher than the cost of a specialist. As AI platforms evolve, in the case of GPT-5, Claude and Gemini's newer releases, in unpredictable and frequent ways, insurers building in-house are also signing up to absorb every one of those changes themselves. Read more on what a compliant AI distribution architecture actually requires.
Worth flagging: many insurers already have a "build" tool in place without realising it changes the calculus here. Platforms like boost.ai give insurers the ability to build their own virtual agent for their website, app or contact centre, and a meaningful number of Tier 1 and Tier 2 insurers already run one. That's valuable infrastructure, but it's a separate job from showing up inside ChatGPT, Claude or Gemini's own app ecosystems. Having an in-house virtual agent doesn't answer the question of what happens when a customer asks ChatGPT directly; that requires either extending the in-house build into each platform's own app framework, or buying that layer separately.
What does buying AI insurance distribution middleware involve?
The insurer keeps its pricing, underwriting and policy admin systems exactly as they are, no re-platforming. The middleware layer translates conversational requests from an AI agent into validated transactions against those systems, catching hallucinated values before they ever reach the insurer, and maintains the compliance and platform-integration work centrally, once, across ChatGPT, Claude and Gemini, instead of the insurer building and rebuilding it three times over.
The trade-off is dependency: a third party now sits in a customer-facing channel, and if that party is doing the guardrail work, the insurer needs confidence in how it is done, and in the underlying data handling. Marrow, for instance, doesn't store personal data and generates a pre-filled, insurer-branded checkout so the customer completes the purchase on the insurer's own site, not a third-party one.
This is also a new category, and standards are still being established. That's part of why Marrow stewards the open AMI (Agent-Mediated Insurance) Standards rather than a closed, proprietary format: a canonical request and response for how an agent quotes, compares, discloses and binds a policy across motor, home, SMB and travel, open to read and build on. See the full breakdown of what the Marrow platform does under the hood.
What does a fully AI-native insurance company look like?
Corgi and Kinro are the clearest live examples. Neither is an infrastructure vendor an insurer could buy: Corgi underwrites its own policies for tech startups, and Kinro is a licensed agency quoting small business cover through chat, phone, text and WhatsApp. What they demonstrate is the ceiling: what quote-to-bind speed and conversational UX look like when a company has no legacy core system to integrate around. Useful for insurers thinking about a greenfield digital brand or MGA, not directly comparable to the build-vs-buy decision above.
How do Marrow, WaniWani, Corgi, Kinro and boost.ai compare at a glance?
| Marrow | WaniWani | Corgi Insurance | Kinro | boost.ai | |
|---|---|---|---|---|---|
| Category | AI distribution middleware, insurance only | AI distribution middleware, multi-vertical | AI-native carrier / MGA | AI-native agency | In-house virtual agent platform |
| Market | UK-born, onboarding insurers globally | US-founded; customers across Europe, Latin America, the Middle East, South Korea, Australia, US | US (California-licensed) | US | Norway-founded; customers across Europe and North America |
| Who it sells to | Insurers and brokers | Insurance, budgets, HR and travel companies, per WaniWani's own site | Venture-backed tech and AI startups, direct | Small businesses (trades, hospitality, retail, consultants), direct | Enterprises across financial services, insurance and telecoms, including insurers |
| Status | Live with Aviva, onboarding new Tier 1 and Tier 2 insurers now | Live with Tuio (Spain), first insurer-built app approved on ChatGPT; insurer logos on its site include AXA, Nationwide, Progressive, Oscar Health | Live, thousands of customers reported | Live, agency model across 50+ carriers | Live with 600+ deployed virtual agents across industries, insurance customers reported include Tryg |
| Founding team | Tim Graham (CEO): Head of Product at Kudo, a D2C motor insurtech, then Head of Product at Ki Insurance, the Lloyd's algorithmic underwriter. Adam Mesout (co-founder): lead algorithmic and analytics actuary at Ki Insurance alongside Tim, and senior actuary at Hadron Insurance | Co-founder Raphael Vullierme has said he spent close to a decade running an insurer before founding WaniWani | Not publicly disclosed | Not publicly disclosed | Founded 2016 by Lars Ropeid Selsås, Hadle Ropeid Selsås and Henry Vaage Iversen |
| Regulatory model | Not an insurer or broker; products underwritten by regulated carriers, developing FCA-authorisation as the channel matures | Infrastructure provider; underlying products underwritten by regulated carriers, not confirmed to hold FCA authorisation | Coverage underwritten by Technology Risk Retention Group, Inc.; Corgi Insurance Services is a licensed producer (CA #6012791) | Kinro Insurance Services LLC, licensed agency (NPN 22233799) | Not an insurer; software vendor to regulated industries, SOC 2 Type II certified |
| AI platforms | ChatGPT, Claude, Gemini | ChatGPT, Claude, Gemini, WhatsApp | None, direct web/app quoting | None, own chat tool plus phone/text/WhatsApp | None, deploys on the insurer's own website, app and contact centre, not third-party AI platforms |
| Standard | Stewards the open AMI Standards, published and free to build on | Proprietary SDK | None | None | None |
| Depth of integration | Connects to insurer core systems including legacy SOAP/XML; quote to bound policy on the insurer's own branded checkout | AI storefront: personalised pricing and lead capture, no app or form required | Underwrites and binds its own policies | Places business with its carrier panel | Deploys a virtual agent on the insurer's own channels |
FAQ
Is buying AI insurance distribution middleware more expensive than building in-house?
Not usually, once ongoing maintenance is counted. Build costs include the initial integration and keeping pace with three separate, evolving app platforms, plus the compliance work behind each. Buy costs are typically a platform fee plus outcome-based pricing, with the maintenance burden carried by the vendor. Marrow's engagement models range from a low-commitment agent build to a pay-on-bound-policy commission structure with zero software maintenance cost.
Can an insurer switch from build to buy, or the other way round, later?
Yes. Neither is a permanent architectural decision. Insurers that build in-house typically keep ownership of the underwriting and policy admin layer regardless, so switching the distribution layer on top of it is a vendor change, not a rebuild.
Do Corgi and Kinro compete with insurers, or with Marrow?
Neither. Corgi and Kinro compete with traditional brokers and carriers in their own markets, US tech-startup insurance and US small-business insurance respectively. They are not distribution infrastructure and do not compete with Marrow or WaniWani for an insurer's business.
Is Marrow only for UK insurers?
No. Marrow was built with FCA compliance in mind, so the compliance rail is properly rigorous instead of bolted on, and Aviva is its lighthouse customer, but the platform is onboarding new Tier 1 and Tier 2 insurers globally, in the UK and beyond.
We already use boost.ai for customer service, do we still need Marrow?
Most likely, yes, if showing up inside ChatGPT, Claude or Gemini matters to you. boost.ai and similar platforms build a virtual agent for an insurer's own website, app or contact centre. That's a genuinely different job from being quotable and bindable inside a third-party AI platform's own app ecosystem, which is what Marrow is built for. The two are complementary rather than overlapping: one is where your customers come to you, the other is where they're increasingly asking AI assistants instead.
This comparison reflects publicly available information as of July 2026. Statements characterising other companies' products or business models, including WaniWani, Corgi, Kinro and boost.ai, reflect Marrow's own understanding of public sources and are provided for informational purposes only, not as a substitute for verifying directly with each provider. Company names are trademarks of their respective owners. Details change; if something here looks out of date, let us know.