What Is an MGA, and How Do MGAs Use AI in Underwriting?

A managing general agent (MGA) is an insurance intermediary that a carrier grants delegated underwriting authority: the power to select risks, price them, bind coverage, and in some cases administer policies or handle claims on the insurer's behalf. In market shorthand, the carrier hands the MGA its "pen." That single fact, delegated authority, is what separates an MGA from an ordinary broker, and it is also what makes modern AI so useful to the many MGAs that want to grow without building heavy technology of their own.

What is an MGA managing general agent?

In one line, a managing general agent (MGA) is a firm that underwrites on behalf of one or more insurers under a binding authority agreement. The insurer, often called the carrier or the capacity provider, keeps the balance-sheet risk and the regulatory license. The MGA does the frontline work the carrier would otherwise do itself: defining appetite, reviewing submissions, rating and pricing, issuing quotes, binding policies, and frequently managing the book once it is on risk. In the United States this arrangement is recognized in the National Association of Insurance Commissioners (NAIC) Managing General Agents Act (Model #225), a model law adopted, in varying forms, by many states. At Lloyd's of London, the same delegated model runs through firms called coverholders.

The line to remember is simple: if a carrier has given you authority to bind its paper, you are acting as an MGA, not merely reselling someone else's product.

MGAs tend to concentrate where standard carriers move slowly: specialty and niche lines, program business, hard-to-place risks, new or volatile classes, and narrow verticals where deep expertise beats broad coverage. The carrier gains distribution and specialized underwriting talent without staffing it internally. The MGA gains the ability to write business without holding its own capital. It is a model built for focus and speed, and it has become a large part of the market. Conning, a widely cited reference for the segment, reported in its 2024 Managing General Agents study that US managing general agent premium surpassed $100 billion, with the MGA channel growing faster than the overall property and casualty market.

How an MGA differs from a broker and a carrier

A retail broker represents the client and shops the market, but it cannot commit an insurer to a risk. A carrier holds the capital, the license, and the ultimate liability. The MGA sits between them with something neither a pure broker nor a pure distributor has: the authority to say yes on the carrier's behalf, within agreed limits. Those limits, appetite, line sizes, pricing bands, and referral triggers, are written into the delegated authority contract, and staying inside them is the core discipline of running an MGA. A related term, managing general underwriter (MGU), is often used for MGAs whose work is weighted heavily toward technical underwriting.

Why MGAs run lean on technology

Most MGAs are built to be nimble. They are frequently founded by underwriters rather than engineers, and they win on speed, specialization, and service rather than on infrastructure. That means the typical MGA runs "thin-IT": a small technology footprint, a policy-admin or issuance system provided by the carrier or a third party, and a great deal of work still moving through email, spreadsheets, and PDF documents. Submissions arrive as broker emails with attachments, ACORD forms, loss runs, and schedules of values. Someone has to read them, key the data, check them against appetite, and turn them into a quote fast enough to win the deal before a competitor does.

This is exactly the pressure point where AI earns its place, which is why the external-layer model has become attractive for MGAs.

How MGAs use AI in underwriting

MGAs use AI to automate the highest-volume, most repetitive parts of the underwriting funnel: reading submissions, triaging them, and getting to a quote. The goal is not to remove the underwriter's judgment. It is to remove the clerical drag that surrounds it. Three uses matter most.

Submission intake

Every MGA lives or dies on submission flow, and most of that flow is unstructured. Document AI and large language models can now read a broker email, open the attached ACORD form or spreadsheet, extract the fields that matter, and drop clean, structured data into the workflow without an underwriter rekeying anything. Done well, submission intake automation turns a pile of inconsistent PDFs into a normalized, ready-to-underwrite record in minutes, and it captures data the team would otherwise miss under time pressure.

Triage and appetite matching

Not every submission deserves the same attention. AI can score each incoming risk against the MGA's defined appetite and the carrier's guidelines, flag the ones that fit, and route or decline the rest before an underwriter spends an hour on a risk that was never going to bind. Faster, more consistent triage means underwriters spend their time on the submissions most likely to convert and most likely to be priced correctly, which protects both the loss ratio and the relationship with the carrier.

Speed-to-quote and decisioning

Once the data is structured and the risk is confirmed in appetite, the same layer can enrich it with third-party data, run a consistent risk score, and draft a quote for the underwriter to review and release. Because MGAs compete heavily on turnaround, the ability to reduce quote turnaround time with AI is often the difference between winning and losing a piece of business. Speed here is an automation problem, not a headcount problem, and treating it that way is what lets a lean team punch above its weight.

Why the external-layer model fits MGAs

Here is the part that matters most for a thin-IT MGA: none of this requires ripping out or rebuilding a core system. The most practical way to add AI to an MGA is as an external layer that sits on top of whatever policy-admin or issuance system the carrier already mandates. It automates intake, triage, enrichment, scoring, and quote drafting, then writes the results back into that system of record. This is the model WIR is built on. It is AI underwriting without replacing the core system, an approach that keeps the carrier's system of record intact, the delegated-authority controls in place, and the integration burden low.

That external-layer approach suits MGAs for three reasons. It respects the carrier relationship, because the carrier's core stays the system of record. It matches how MGAs are staffed, because it automates the clerical layer instead of demanding an in-house engineering team. And it preserves the delegated-authority guardrails, because appetite rules, line sizes, and referral triggers can be encoded and enforced consistently rather than living only in a senior underwriter's head.

Brazil, SUSEP, and LGPD

WIR was born in Brazil, where the MGA model is younger and less codified than in the United States or at Lloyd's, though delegated underwriting and program-style arrangements are growing under the oversight of SUSEP, the Superintendência de Seguros Privados. Two rules shape how any AI layer should behave in that market. First, underwriting decisions have to stay explainable and auditable, so that a regulator or a carrier can see why a risk was accepted, priced, or declined. Second, because submissions are full of personal and business data, every step has to comply with the LGPD, Brazil's general data protection law. An external AI layer that logs its inputs, its scores, and its outputs is well suited to both requirements, which is why auditability belongs in the design from the start rather than bolted on later.

The takeaway

An MGA is a carrier's delegated underwriter, trusted to hold the pen but rarely built like a technology company. AI closes that gap by automating the submission intake, triage, and speed-to-quote work that used to demand more people, and the cleanest way to adopt it is as an external layer that leaves the core system exactly where it is. For an MGA, that is how you underwrite faster and more consistently without turning into an IT shop.