AI agents will buy insurance. Is your pricing ready?

Customers are starting to ask AI assistants to do tasks for them, not only answer questions. Booking travel, comparing energy tariffs and filling in forms are already common requests. Buying insurance is a natural next step, and in some cases it is already happening: an assistant gathers a customer's details, requests quotes through an API and presents the options.

For insurers and MGAs, this raises a practical question. When the party asking for a quote is software acting for a customer, what does your pricing need to provide?

How agents request quotes

An AI agent cannot fill in a web form the way a person does, at least not reliably. It works best with structured interfaces:

APIs that accept a defined set of inputs and return a price.
MCP servers, which let an AI client discover what a product needs, request a price and read the result in a standard way.
Embedded journeys with partners, where the agent works inside a platform that already has a pricing integration.

In each case, the agent needs to understand the product well enough to ask the customer the right questions and to explain what comes back.

What pricing needs to provide

A clear input schema

The agent has to know which inputs are required, what type each one is, which values are allowed and what each one means. Field names such as region or vet_fee_limit are not enough on their own. Each input needs a plain-English description an agent can follow when asking the customer, for example "Annual vet-fee limit: the maximum the policy pays for vet treatment each year".

Ptally models carry this information for every input, and the MCP server and API expose it, so an agent can build a valid request without guessing.

Consistent results

The same inputs should always produce the same price, whichever channel the request comes through. If an agent gets one price through an API and the customer sees another on your website, trust is lost quickly. A single model serving every channel avoids this.

Clear decisions

An agent needs to know whether a quote can be bought. Every response should say whether the quote is valid, declined or referred to an underwriter, with the reason. An agent that presents a declined quote as available, or misses a referral, creates problems for the customer and for you.

Explanations

Customers will ask the agent why a price is what it is, or why it changed from last year. The agent can only answer accurately if the pricing engine explains how the price was built. Ptally returns a debug trace with every input, lookup, formula and rule, which an agent can read and summarise.

Speed

Agents often request several quotes in one conversation. Slow responses make the conversation slow, and an agent may move on to faster providers. Response times that work for price comparison websites also work for agents.

Consumer protection still applies

An AI agent acting for a customer does not change the insurer's obligations. Firms still need to make sure customers understand what they are buying, that products meet their needs and that pricing is fair. Some questions to think through now:

Disclosure. How will key information, such as exclusions, excesses and limits, reach the customer when an agent sits in the middle?
Accuracy of inputs. If an agent fills in an answer on the customer's behalf, how do you confirm the information is correct before cover starts?
Vulnerable customers. How will you identify and support customers who need extra help when you are not speaking to them directly?
Records. Can you show what was quoted, with which inputs, and which model version priced it?

Clear input descriptions, explicit decisions and full traces help with each of these. They give the agent what it needs to present the product correctly and give you a record of what happened.

What to do now

You do not need to wait for agents to become mainstream to prepare. The same steps make pricing better for every channel:

Move pricing into a dedicated rating engine that serves APIs, forms and AI clients from one model.
Write plain-English descriptions for every input, as if explaining the question to a customer.
Make every decision explicit, including the reasons for declines and referrals.
Return explanations with every quote.
Test the model on every change, so agents and customers always get the price you intended.

Our article on AI for insurance distribution covers the wider changes to distribution channels in more detail.

Getting started

Every Ptally model is available through the API and our MCP server at https://ptally.ai/mcp from launch, so AI clients can quote your products as soon as they go live. There is no implementation fee and nothing to pay upfront. Talk to our team about preparing your products for AI-led distribution.