Starting an MGA: from rating guide to first live quote

The to-do list for a new managing general agent is long. You need capacity, a binding authority, distribution, compliance, systems and people. Pricing runs through all of it, because until the product can return a correct price wherever you plan to sell it, there is nothing for anyone to buy.

This post follows the path from an agreed rating approach to the first real quote, and picks out the choices that matter most when you're starting from nothing.

Agree the rules with your capacity partner first

Whoever backs you, whether an insurer, a Lloyd's syndicate or a reinsurer, will have a view on how the product is priced. Settle these before you build anything:

  • Whose rates they are. Some partners hand you the rating guide. Others expect you to propose the rates and get them signed off.
  • How much you can move on your own. Binding authorities often allow small adjustments within set limits, with anything bigger going back to the partner.
  • What proof they want. Usually test results, a history of changes and regular reports on what you've quoted and written.
  • Where appetite ends. Which risks decline outright, which go to an underwriter, and how those rules are written down.

Each of these is far cheaper to agree now than to retrofit once the model is built.

Treat the rating guide as a specification

The rating guide is what the model will be built from, so it needs to be complete. That means every input and its allowed values, every rate table, the order factors apply in, rounding, minimum premiums, fees, tax and the underwriting rules.

Most importantly, it needs worked examples with exact premiums. They do two jobs. They settle the questions the written rules leave open, such as whether a discount applies before or after the minimum premium. And they become the tests that prove the model is right. A set covering your main risk types, both sides of each band boundary and every decline and refer rule is a solid place to start.

Decide how the model will be built

There are three broad routes:

  1. A spreadsheet. Fast to start, but hard to serve live, hard to audit and easy to break. See replacing a spreadsheet rater for why that catches up with you.
  2. Your own code. As flexible as you like, but it needs developers who understand pricing, and every rate change becomes a release.
  3. A rating engine. The model is configured, not coded, tested on every change and available over an API from the first version.

For a new MGA, the third route is usually the quickest to market. In Swallow you can describe the product in a prompt and upload supporting data from Excel to get a first draft, build on a drag-and-drop canvas, or have an AI client such as Claude write the model through the MCP server. However it's made, the result is a readable model with labelled inputs, plain rate tables and written-out formulas that your underwriters and your partner can review. Our developer guide shows a complete one. If you're writing in the US, you can also start from public rate filings, as we cover in building pricing models from US rate filings.

Prove it before you sell it

Every worked example should match to the penny before launch. Then add tests for:

  • values either side of each age, sum insured or other band;
  • each decline and refer rule, plus cases that sit just inside it;
  • combinations that are unusual but perfectly valid.

In Swallow, tests are saved on each version, so the whole suite can run again every time the model changes. Swallow's AI can also suggest a batch of tests from a prompt for you to trim. When a test fails, the quote trace shows the value of every input, lookup and formula, so you can find the step that went wrong. The Test page has more.

Start with one or two channels

Most MGAs launch through a small number of channels and add more once the product has settled. Typical first choices:

  • Brokers, who will usually want an API.
  • Embedded partners, with your quote sitting inside their journey.
  • Your own site, with an embedded quote form.
  • Comparison sites, which need fast answers at volume.

When one model serves every channel, adding a new one later is a setup task rather than a project. Swallow's API, embedded forms and chat and voice agents all price from the same model. See sales channels.

Watch the numbers once you're trading

The first few months show how your rates meet real demand. Keep an eye on:

  • quote volume and conversion, split by channel;
  • how many quotes decline or refer, and which rules are behind them;
  • average premium against what the rating guide expected;
  • response times, especially on busy channels.

Swallow's Data area shows live quote data, and it can produce bordereau reports for your capacity partners.

You will almost certainly change rates early. In Swallow, a change goes into a new draft copied from the live version. You run the tests, review the change and approve it, with your team's roles deciding who can approve. Then you publish it. Every step is logged with who did it and when, which is the kind of record a capacity partner asks to see. Our post on the insurance rate change process goes through it in detail.

Getting started

If you have a rating guide, a spreadsheet or a filing, even an early draft you're still working through with your partner, send it over and we'll talk you through getting it live on Swallow. Get in touch.

Continue reading

From filing to rater: building pricing models from US rate filings

What sits inside a US rate filing, how to rebuild it as a working, tested rating model, where AI speeds things up and which gaps to expect.

Swallow compared: enterprise pricing platforms, core system raters and AI newcomers

Where Swallow sits against established insurance pricing and rating platforms and newer AI-led options, on model format, testing, approvals, AI and distribution.

Building pricing models with Claude and ChatGPT: what works, what doesn't

A practical look at building insurance pricing models with Claude and ChatGPT over MCP: where AI clients help, where they trip up and how to keep people in charge.

AI in insurance pricing: closing the gap between rating guide and live model

Pricing design is rarely the bottleneck. Implementation is. How AI shortens the path from signed-off rating guide to a tested, live pricing model.

Fair value under Consumer Duty: the evidence your pricing should leave behind

What the FCA's Consumer Duty and pricing rules ask of insurance pricing teams, and the records that make fair value something you can show, not just claim.

When the customer is an AI agent: getting your pricing ready

AI assistants are starting to shop for insurance on people's behalf. Here's what your rating engine has to give them to quote your products properly.

How to connect the Swallow MCP server to Claude

Connect Claude to your Swallow pricing infrastructure using MCP to analyse models, inspect rating logic and interact with your tools using natural language

Regression testing insurance pricing: why every rate change needs a test suite

How to build a regression test suite for an insurance pricing model, from worked examples and band edges to declines, refers and real quotes.

A safer rate change: eight steps from analysis to monitoring

A repeatable process for insurance rate changes, from the change statement and draft version to tests, approval, release and monitoring, and where AI helps.

Readable by design: pricing models people can check, even when AI writes them

When AI drafts pricing logic, review becomes the slow step. Why a declarative, readable model format is faster to check and audit than generated code.

Moving off a spreadsheet rater without changing a single price

How to tell when an Excel rater has outgrown its job, what keeping it really costs, and a migration plan that uses the spreadsheet's own answers as the tests.

AI and insurance distribution: one price, every channel

Aggregators, embedded journeys, chat, voice and AI agents all need the same thing from pricing. Why one explainable rating engine should sit behind them all.

AI in regulated industries: why explainability has to come first

Insurers can get real gains from AI and still meet what the FCA, the EU and US regulators expect, provided every decision can be explained, reproduced and signed off.

Swallow vs Earnix

Swallow and Earnix Price-It compared for insurance pricing and rating: who each is for, how models are built, tested and released, how AI is used, and distribution.

Swallow vs Guidewire PricingCenter

Swallow and Guidewire PricingCenter compared: who each suits, how each relates to PolicyCenter, how models are built, tested and released, AI and distribution.

Swallow vs Duck Creek Rating

Swallow and Duck Creek Rating compared: who each suits, suite and standalone use, how rates are built, tested and released, deployment, AI and distribution.

Swallow vs Oracle Insbridge

Swallow and Oracle Insurance Insbridge Enterprise Rating compared: architecture, how rates are built and changed, testing, integration, AI, and how to migrate.

Swallow vs Akur8 Deploy

Swallow and Akur8 Deploy compared: where each rating engine fits, how rates reach production, testing and sign-off, how AI is used and when to pick which.

Swallow vs hyperexponential (hx Renew)

Swallow and hyperexponential's hx Renew compared: who each suits, Python models versus readable JSON, testing and approvals, AI and the Guidewire deal.

Swallow vs WTW Radar

Swallow and WTW Radar compared for pricing and real-time rating: who each suits, how models are built and changed, testing and approvals, and AI features.

Swallow vs building a rating engine in house

Build or buy? A fair look at writing your own insurance rating engine versus using Swallow: testing, versioning, audit, serving at volume and rate changes.

How to connect the Swallow MCP server to ChatGPT

Connect ChatGPT to your Swallow pricing infrastructure using MCP to analyse models, inspect rating logic and interact with your tools using natural language

Swallow Data: Live insight into your pricing performance

See how Swallow’s Data area gives you real-time visibility over quotes, performance, and customer behaviour across your entire pricing model

Swallow Tests: Validate your pricing with real-world scenarios

See how Swallow’s Tests area lets you simulate quotes, inspect calculations, and perfect your pricing model before publishing

Swallow Build: Bring your pricing model to life

A quick look at how Swallow’s Build area lets you edit, test, and publish pricing models with complete control and instant feedback

Swallow Overview: Core features at a glance

See how the AI actuary inside Swallow turns a simple prompt into a fully tested and deployable insurance pricing model for electric vehicles in just a few clicks

Why Modern Insurers Are Moving to AI-Powered Enterprise Pricing Engines

A look at how AI is transforming insurance pricing, from spreadsheet-based models to enterprise-grade rating engines built for speed, consistency, and control

Swallow / MGT

How MGT use Swallow to power AI-driven pricing, accelerate model updates and support fast, reliable commercial insurance quoting.

AI Agent for Insurance Products

Sam talks through our new AI Agent for Insurance products. Get a new Insurance product built and live within a couple of minutes.

Swallow / Ouch!

How Ouch! used Swallow to launch digital takaful products faster, with transparent pricing, audit readiness and seamless model updates.

Pricing Pain Point #4 - Slow deployment

Whilst the modelling process is where pricing strategy begins, getting changes in front of customers is where the rubber hits the road and new pricing is tested with real customers. Moving quickly to make changes to live pricing offers pricing teams an opportunity to respond to the market, test and iterate, and roll out new products.

Pricing Pain Point #3 - Testing concerns

With pricing critical to growth and profitability, businesses must understand the impact of any price changes, as well as ensure what goes live to customers is identical to that modelled. With pricing driven by complex financial models, this is particularly true - a percentage point error in the code deployed could easily lead to a significant financial impact.

Swallow / Open

How Open used Swallow to launch embedded insurance in the UK faster, with flexible pricing, rapid updates and lower operational cost.

Pricing Pain Point #2 - Lengthy process

In any fast moving business, having a clear record of decisions, changes and their impacts is key to operating efficiently and successfully. For regulated industries, a clear governance process is a requirement and critical to ensuring compliance with local regulations.

Pricing Pain Point #1 - Data issues

Underlying all pricing strategy is data, whether it be on the market and competition or a company’s own quote, sales and cost data. Collecting this data and using it to inform future pricing decisions is one of a pricing team’s key objectives.

Swallow / Rivr

How Rivr uses Swallow to deliver pricing models faster, with better governance, smoother client handovers and one-click deployment.

What we've learned about pricing

Over the past few months, we've talked to over 100 pricing and technology professionals from a range of industries. From our conversations, many pricing pain points have emerged, helping us to better understand the challenges businesses face and inform the direction of our platform.

Why I created Swallow

Financial pricing models (think insurance, credit, mortgages, financing, leasing, memberships) have, and continue to be, designed in spreadsheets and published through in-house custom code or legacy enterprise software. Pricing changes can take weeks instead of hours.

We feel you, pricing is hard...

Dynamic pricing is hard to create and serve to your customers. Because it is hard, it can become a technical concern rather than a business concern. Here be where the dragons lie.