Not long ago, most insurance products had two or three routes to market: a broker, a direct line and perhaps a comparison site. A product launched today might be expected to quote on an aggregator, inside a partner's checkout, on a broker portal, in a chat window, over the phone with a voice agent and through an AI assistant acting for the customer.
That is a lot of places for a price to appear. And every one of them wants the same three things: the right price, quickly, with a clear account of what is and isn't covered.
Where quotes come from now
- Comparison sites send heavy, spiky traffic and expect an answer in a fraction of a second.
- Embedded insurance puts cover inside someone else's journey, such as buying a phone, hiring a van or booking a trip. The quote has to be one quick, simple step.
- Brokers and partners want your rates and rules over an API, usually in their own field names.
- Chat lets a customer ask for a quote in their own words and ask questions as they go.
- Voice does the same over the phone, for new business, renewals and changes.
- AI agents gather a customer's details, request quotes and compare them, all through software. We look at this in more depth in when the customer is an AI agent.
The old way to serve a new channel was a new integration, and often a new copy of the rating logic to go with it. Copies drift. Sooner or later the same risk gets two prices.
Pricing as a single service
The setup that holds up as channels multiply is simple to describe. There is one model, and every channel calls it:
- a real-time API for aggregators, partners and your own systems;
- embedded quote forms for sites and apps that don't want to build a journey;
- chat and voice agents that collect answers in conversation and price them against the model;
- an MCP server, so AI clients can read a product's inputs and price it.
Change a rate and every channel picks it up together, because there is only one place to change it.
What AI actually changes
Two things.
Quoting becomes a conversation. A customer can describe their situation however they like, and an assistant turns that into the product's inputs, requests a price and talks them through it. The assistant doesn't need to know how pricing works. It needs an engine that says what it wants to know and explains what it worked out.
Small channels become worth doing. A niche product, a small partner or a trial proposition rarely justified its own integration. When a channel can read a product's inputs and quote it through a standard interface, the cost of adding one drops a long way.
Keep the conversation and the calculation apart
For AI-led distribution to be safe, the AI should run the conversation and the rating engine should set the price. The engine behind it needs:
- Inputs an agent can follow, each with a type, allowed values and a plain description.
- One answer per risk, whichever channel asks.
- Explicit decisions: valid, declined or referred, with the rule that made the call.
- A trace of how the price was built, for customers, brokers and complaints teams.
- The speed to cope with aggregator peaks and live calls.
Keeping those two jobs separate is what lets you put AI in front of customers without losing control of the price.
How this works in Swallow
Every Swallow project gets its channels from the same model: the API, customisable embedded forms, chat AI and voice AI. When you publish a new version, they all price from it. Our voice agents can handle inbound and outbound calls to quote, renew and amend, and for renewals you set the boundaries they negotiate within,. Calls are recorded and transcribed, and a customer who asks for a person is passed to one.
Every quote, whichever channel it came from, is saved with the version that priced it, and can return a full trace of how the price was made. See the sales channels page for more, or talk to our team.
