The New Front Door: How AI Is Rewriting Insurance Distribution — And How Incumbents Win

First published on Substack by Paul Prendergast - March 14, 2026

‍

On February 9th, 2026, two AI-powered insurance apps went live inside ChatGPT.

The market’s reaction was immediate. WTW dropped 12%. Aon fell 9.3%. Arthur J. Gallagher lost 9.9%. The MarshBerry Broker Composite Index was down 8.9% in a single session. Billions wiped from broker valuations before lunch.

Then BofA put a number on the fear: $15 billion in low-complexity insurance commissions at risk from AI disintermediation.

The consensus response was swift: overreaction. Too early. Brokers aren’t going anywhere. Goldman Sachs called it “overdone.” TD Cowen said near-term commercial broker disintermediation from AI was unlikely. McKinsey concluded AI would “reshape existing models rather than disintermediate them.”

They may be right about the timeline.

They are wrong about the direction.

Because this is not a debate about whether AI replaces brokers. It is a debate about who owns the distribution infrastructure when AI becomes the front end — and whether incumbents move fast enough to be part of it.

______________________________________________

What Actually Changed on February 9th?

The apps that triggered the sell-off were not sophisticated. Insurify launched a car insurance comparison tool inside ChatGPT. Tuio, a Spanish digital insurer, launched a home insurance quoting app. Neither was going to put Marsh McLennan out of business by Thursday.

But what changed was not the products. It was the mechanics.

For the first time, an insurance provider could distribute its products and offer quotes directly inside an AI platform where hundreds of millions of buyers already perform their research. Until that day, AI could only provide generic answers drawn from static web content. It could not quote a real price for a real person or business.

That changed on February 9th. And it has not changed back.

Through OpenAI’s App Directory — effectively an app store inside ChatGPT — third parties can embed real products and workflows directly into the conversation. Taken together, these are clear signals that conversational AI is shifting from an information layer to an action layer.

Distribution economics are shaped by whoever controls the customer’s starting point. AI assistants are now delivering the first explanation of value, replacing the carrier, its agents, and distribution partners as the initial voice that shapes consumer perception.

When the first interaction happens inside an AI interface, the traditional pathway — website, form, comparison journey, broker call — becomes less central. Value migrates toward the firms that control, integrate with, or are discoverable within that new front door.

And the front door is moving fast. The technology is not limited to OpenAI. AI apps built on the same infrastructure and standards have also been adopted by Anthropic’s Claude, and Google’s Gemini is expected to publish its own standards for third-party apps in the coming months. The shift toward agent-to-agent distribution is becoming an industry-wide reality.

As of this week, the list of insurance apps in ChatGPT has grown to include Neptune Flood, Steadily (landlord insurance), and Jerry.ai (auto), joining Insurify and Tuio. Neptune’s chief engineer explained that they “architected our proprietary underwriting system as a modular, API-first underwriting system specifically so it could integrate into new digital environments like ChatGPT.”

This is not a wave coming. It is already here.

______________________________________________

Two Distribution Fronts — Not One

Here is what most of the coverage has missed.

The AI distribution shift is happening on two fronts simultaneously, and most incumbents are only watching one of them.

Front One: AI Chat Platforms. ChatGPT, Claude, Gemini. Consumer and SMB buyers asking insurance questions in natural language and getting real-time quotes from carriers that have built the API connectivity to respond. Tuio’s co-founder of WaniWani said: “For the first time, AI can access real offers, quote on behalf of the buyer, and compare coverage in real time. Every insurer will be impacted, whether they’ve built an AI app or not.”

Consumers and businesses are already uploading commercial offers and policy documents into ChatGPT to get independent analysis and advice. AI voice agents are calling call centers on behalf of buyers to collect and compare quotes. Procurement teams are using AI to evaluate coverage terms, exclusions, and pricing across multiple carriers simultaneously.

Front Two: Vertical SaaS Platforms. The operational software where SMBs run their businesses every day — ServiceTitan for field services, Toast for restaurants, Procore for construction, franchise management platforms for franchise operators. These platforms are now embedding AI agents that handle procurement on behalf of their users. When that AI agent surfaces an insurance need — a contractor scaling their crew, a restaurant adding a location, a franchisee coming up for renewal — it will fulfill that need through whatever insurance infrastructure is connected to its platform.

These two fronts are converging. The AI agent in a business’s operational software will query insurance products through the same API-first, MCP-compatible infrastructure that ChatGPT apps use. The question for every broker and carrier is whether their products are accessible via that infrastructure — or invisible to it.

______________________________________________

The Protocol That Changes Everything

To understand why this is moving so fast, you need to understand MCP.

Model Context Protocol is an open standard, originally developed by Anthropic and now governed by the Linux Foundation, that defines how AI agents connect to external tools and data sources in real time. Think of it as USB-C for AI — a standardised interface that lets any AI model query any compatible system, regardless of who built either one.

With MCP in place, AI assistants can respond to a prompt like “How much would it cost to insure my business?” by understanding the user’s intent, gathering necessary context from connected systems, and returning an accurate, personalised quote — all in the flow of a natural-language conversation.

Neptune Flood’s ChatGPT app is built on MCP. Their chief engineer explained: “Using the Model Context Protocol, a lightweight API layer securely orchestrates data retrieval, risk modelling, and rating in real time on top of our existing underwriting infrastructure. Because our underwriting stack is fully automated and cloud-native, we can extend instant quoting into conversational AI without changing our core workflow.”

That is the key sentence: without changing our core workflow. The carriers that move fast in this channel are not rebuilding their systems. They are exposing them through a standardised API layer that AI agents can query.

For a broker or carrier that is API-ready, connecting to the AI distribution layer is not a multi-year technology programme. It is a configuration exercise. For one that is not API-ready, it is a multi-year technology programme — and the market will not wait.

______________________________________________

What the Incumbent Advantage Actually Is

Here is where the narrative gets more nuanced — and more useful.

The carriers and brokers that are panicking about AI disintermediation are asking the wrong question. The right question is not “will AI replace us?” It is “what assets do we have that AI distribution actually needs?”

The answer is substantial.

Capacity and compliance. An AI agent can surface a quote. It cannot underwrite the risk, hold the regulatory authorisation, or carry the balance sheet. Every AI distribution channel, whether it is a ChatGPT app or an embedded insurance offer in a SaaS platform, needs a licensed, regulated, capitalised carrier behind it. Berenberg analysts pointed out that the regulatory burden and liability exposure of selling insurance directly are significant hurdles that OpenAI and others may not want to manage independently. The incumbent’s regulatory infrastructure is a moat, not a liability.

Product breadth and market relationships. The AI agents quoting inside ChatGPT today are doing simple, single-product personal lines. The SMB that needs a BOP, commercial auto, workers’ comp, and an umbrella needs a broker with multi-carrier access and placement expertise. AI accelerates the front of the journey. It does not replace the depth of what a well-positioned broker or MGA brings to complex commercial placement.

Customer data and relationship history. The broker that has a five-year relationship with a growing contractor business has renewal data, claims history, and risk context that no AI agent querying a cold lead can match. The retention economics in an embedded, data-rich context — where the broker is present in the platform the customer uses every day — are structurally superior to cold acquisition.

The distribution network. The brokers and MGAs that thrive will not be those who panic. They will be the ones whose infrastructure lets them plug into every new distribution channel. The broker with 50 carrier relationships and a well-managed delegated authority framework can deploy across AI channels, SaaS platforms, and traditional routes simultaneously. The challenger with one carrier and a ChatGPT app cannot.

The incumbent’s problem is not a lack of assets. It is a lack of the connectivity layer that makes those assets accessible to the channels where SMBs are moving.

______________________________________________

The Infrastructure Gap — And How to Close It Fast

The gap between where incumbents are today and where the AI distribution channel requires them to be is primarily a technology infrastructure gap. It has three components.

API-first rating and binding. For an AI agent to quote and bind your product, whether inside ChatGPT, Claude, Gemini, or a vertical SaaS platform’s embedded AI — your rating engine must respond to real-time API calls. Get API-ready. Can your systems deliver a real-time quote to a third-party platform today? If not, that is the first priority. Not because ChatGPT is coming for your book tomorrow, but because every emerging distribution channel requires this capability. Embedded insurance, partner integrations, comparison platforms, and AI agents all depend on the same thing: open, real-time API access to your rating and binding engines.

Data orchestration across channels. The SMB vertical SaaS channel holds operational data — revenue, headcount, transaction volume, job types — that makes for dramatically better underwriting than a static ACORD form. Real-time underwriting driven by live platform data improves policy accuracy by up to 40%. A broker or carrier that can ingest that data through an API layer and price against it has a structural advantage over one quoting blind. The challenge is connecting those data flows compliantly across multiple platforms without building bespoke integrations for each one.

Compliance architecture at scale. Distributing insurance through third-party AI platforms or SaaS channels is a regulated activity. In the US, that means surplus lines compliance, state-by-state authorisation, and varying requirements for affinity-style distribution across 50 states. In the UK, it means FCA authorisation, ICOBS, and Consumer Duty obligations. The AI chat platforms are not going to carry this. The SaaS platforms are not going to carry this. The broker or carrier must — and the ones who have pre-built this infrastructure will move in weeks where others move in years.

The purpose-built embedded insurance infrastructure layer — connecting rating engines, data orchestration, and compliance across both AI chat channels and vertical SaaS platforms — is the asset that allows an incumbent to move at the speed the market now requires.

This is what Kayna provides. Not a distribution channel of our own. Not a competing product. The API infrastructure and compliance architecture that allows a broker or carrier to plug into ChatGPT, Claude, ServiceTitan, Toast, and the next ten platforms that emerge — through a single integration, with a single compliance framework, without rebuilding their core systems.

______________________________________________

The Playbook for Incumbents

The market has already moved past the point where watching is a strategy. The ChatGPT app store is live and growing. Vertical SaaS platforms are embedding AI agents. MCP is becoming the standard interface through which AI accesses everything.

The playbook for incumbents who want to win has three moves, executed in parallel.

Move 1: Get your rating engine API-ready. This is the table-stakes requirement for any channel that matters in the next five years — AI chat, SaaS-embedded, or otherwise. If your products cannot be quoted in real time through an API call, they will not be quoted at all in the channels where SMBs are moving.

Move 2: Partner with the infrastructure layer, not the distribution channels directly. The mistake that slow-moving carriers and brokers will make is to try to build point-to-point integrations with individual AI platforms or SaaS tools. That does not scale. Each integration becomes its own project. The right move is to connect once to an infrastructure layer that handles the distribution mechanics across all channels simultaneously — and focus your resources on product, capacity, and the customer relationships that AI cannot replicate.

Move 3: Reframe your value proposition for the AI-front-end world. Your margin is not in quoting. It never was, really — quoting is about to be free. Your margin is in the depth of coverage, the accuracy of risk assessment, the quality of claims handling, and the retention economics of a customer who never leaves the platform where you are embedded. Position there.

Tuio’s co-founder said this about the February 9th launch: “Today is day zero of that transformation.”

Day zero was five weeks ago. The incumbents who move in the next quarter will have a structural head start. The ones who wait for more evidence will be building against competitors who already own the channel.

______________________________________________

The Bottom Line

$15 billion in low-complexity SMB insurance commissions at risk from AI disintermediation. The AI chat platforms are live and growing. The vertical SaaS AI agents are being deployed. MCP is standardising the protocol through which AI accesses insurance products.

The distribution infrastructure is being rebuilt. The question is not whether incumbents are part of that rebuild. The question is whether they are part of it on their terms — or someone else’s.

The brokers and carriers that connect to the AI distribution layer now, through infrastructure built for the purpose, will not just defend their SMB book. They will grow it — into channels with better data quality, lower acquisition costs, and retention economics that traditional distribution has never been able to match.

The new front door is open. The incumbents who walk through it first will own what is behind it.

______________________________________________

Paul Prendergast is CEO and Co-Founder of Kayna, the infrastructure layer connecting carriers, brokers, and vertical SaaS platforms for SMB insurance distribution. Kayna operates live programmes in the UK and US in partnership with WTW.