How to Scale MGA Claims Operations Without Adding Headcount
Where agentic AI takes work off the claims desk, what the handler keeps, and how to phase it in few weeks.
11 min read
US MGA direct premium written reached $102.6 billion in 2025, up 12% on the prior year, while the broader P&C market grew about 5%. The claims desk absorbing that growth has not expanded at anything close to the same rate.
You scale an MGA claims operation by moving the administrative work around each claim onto AI automation, so each handler carries more files, standardising how decisions get made so quality holds as volume climbs, and letting the same automation produce the oversight evidence your capacity providers now demand. Capacity comes from the hours you take back. Consistency comes from encoding your SOP once instead of teaching it 40 times. Both arrive in the same deployment.
This applies whether you’re a US program administrator, a UK MGA writing on delegated authority, or a Lloyd’s syndicate running its own service company. The pressure arrives in the same place: the claims file.
What you will learn:
- Where an MGA claims team’s hours actually go, and which of them need a person
- What capacity providers and Lloyd’s now expect you to evidence about claims handling
- What agentic AI does at each point in the claim lifecycle, and what the handler keeps
- Why an MGA can push automation further and faster than a multi-line carrier
- A sequence for putting it in place, with the guardrails each stage needs
Why MGA claims capacity runs out before the growth does
Premium arrives in one step. Claims volume arrives later, unevenly, and lands on a team sized for last year’s book. Hiring is the slowest lever available to close that gap.
The US Bureau of Labor Statistics projects employment of claims adjusters, appraisers, examiners, and investigators to decline 6% between 2025 and 2035, with roughly 21,600 annual openings driven almost entirely by replacement rather than growth. The pool you’re recruiting from is shrinking while your book grows.
The work filling that pool’s time is mostly not claims work. Rising Medical Solutions’ 12th Annual Workers’ Compensation Benchmarking Study found 46% of claims professionals spend 30% to 40% or more of their time on routine tasks. That study covers workers’ comp and doesn’t publish its sample size, and the shape of it holds on every claims floor we’ve worked on.
For an MGA the ratio is worse, because delegated authority adds a second job on top of the first. Alongside handling the claim, your team produces the evidence that the claim was handled properly: bordereaux, service-standard reporting, audit responses, reserve movements explained back to a capacity provider who was never in the file.
Look at a single new claim and sort the work by whether it needs a person.
Work on the claim file | What it needs from a person |
|---|---|
Read inbound mail, decide what it relates to | Nothing. It’s classification. |
Index and file attachments to the right exposure | Nothing. It’s routing. |
Identify what’s missing and request it | Nothing until an exception appears. |
Check whether an expected action happened | Nothing. It’s monitoring. |
Assemble the monthly bordereau and reconcile it | Nothing. It’s aggregation. |
Decide coverage, liability, quantum, and strategy | Everything. This is the job. |
Five rows of that table are the capacity you’re missing. We’ve written before about why you can’t staff your way out of broken claim execution.
What capacity providers now grade you on
Capacity providers have moved from grading MGAs on premium volume to grading them on evidence of consistent claims handling. Two changes in the last 18 months put that in writing.
In the Clyde & Co and MGAA MGA Opinion Report 2025, 77% of MGAs said the claims process with carriers needs improvement, up from 59% in 2023, and 91% of carriers agreed claims handling needs to be clearer, faster, and more customer-focused. A near-unanimous carrier view in a market survey tells you where the next round of contract conditions is coming from.
In London the bar moved by rule. On 1 January 2026, Claims Management became the fifth fundamental “hurdle” Principle at Lloyd’s, meaning a syndicate’s overall category rating cannot exceed its lowest hurdle rating. The same document requires active monitoring against service standards and “regular data feeds for timely reserve and loss ratio assessment” from delegated arrangements. The FCA is expanding its own review of outsourced claims oversight to other delegated authority models from Q2 2026.
Read that from the MGA side. Your syndicate partner’s Lloyd’s rating is now capped by its claims score, and part of that score depends on how well it can see into your operation. Your reporting quality became their rating problem, which makes it your renewal problem.
Consistency used to be a quality aspiration. It’s now a commercial asset that decides what capacity you can hold, and on what terms.
How agentic AI adds claims capacity without adding handlers
Agentic AI works the claim rather than assisting the person working it. It reads what arrives, applies your SOP to decide the next action, executes the actions that fall inside your rules and boundaries, and brings a handler in at the points where judgment is required.
An AI assistant waits to be asked. It summarises a document when a handler opens it and pastes the result back into the file. The handler still holds the whole file in their head, and the time saved is measured in minutes of reading.
Agentic automation runs against the file continuously. When mail lands, it’s already classified, indexed to the right exposure, and checked against what the claim should contain by now. The handler opens a file that has moved since they last saw it. The capacity comes from cutting the number of occasions a person has to touch the file at all. Bain estimates generative AI in P&C claims handling can deliver a 20% to 25% decrease in loss adjusting expenses and a 30% to 50% decrease in total leakage.
Here’s what that looks like at each point in an MGA claim.
Point in the claim | The manual loop today | What the AI automation does | What the handler keeps |
|---|---|---|---|
FNOL intake | Read the notification email, open the claim, key the fields, open exposures, assign | Parse the email and attachments, pre-fill the FNOL, open the right exposures on your rules, route by licence and workload | Reviewing anything the rules flagged as ambiguous |
Document handling | Download attachments, rename, file to the right exposure, read, note the file | Ingest, categorise, attach to the correct exposure, summarise, re-check on every new document | Reading the documents that change the outcome |
Missing information | Notice a gap, remember the standard list, draft the request, diary the chase | Compare the file against the SOP checklist, issue the standard request, track the response, chase on schedule | Deciding when a gap is material enough to reserve for |
Coverage and liability | Read the policy, the endorsements, the loss facts, form a view | Surface the applicable wording, flag coverage gaps and document inconsistencies for attention | The determination. Always. |
Reserving | Set at FNOL on limited facts, revisit when someone remembers | Recommend on your rules when new facts land, alert on indemnity creep | Approving or overriding the movement |
Monitoring | Supervisor spot-checks files and diaries | Watch every file against expected actions and service standards, escalate the ones sitting idle | Working the escalations |
Reporting | Export, map, cleanse, reconcile in a spreadsheet | Generate from the structured record of what actually happened | Reviewing before it goes out |
Four things make this land harder at an MGA than at a multi-line carrier.
- Your SOP is knowable. A carrier writing 40 lines cannot encode one procedure. An MGA writing one specialism can write its handling standard down and have it followed on every file. The narrower the book, the higher the share of the work that can run on rules.
- Your judgment content is concentrated. On a specialist book the hard calls cluster in predictable places. Everything before them is preparation, which is exactly what automation is good at.
- You have to produce the evidence anyway. A carrier automating claims gets efficiency. An MGA automating claims gets efficiency and the delegated-authority reporting it was going to build by hand regardless. The same work pays twice.
- Supervisor span is a constraint too. Most MGA capacity plans count handlers and forget that each one needs oversight. When completion monitoring runs automatically, a supervisor stops checking whether routine steps happened and starts working exceptions, which raises how many handlers one person can carry. That’s capacity you gain without hiring in either row.
See Clive AI in Action.
See what this looks like on your own claim types and bordereau reporting.
How the same AI automation produces your oversight evidence
Every action an AI agent takes is a structured event with a timestamp, an input, and a rule behind it. That turns oversight reporting into a by-product of handling instead of a monthly reconstruction.
Most delegated claims reporting is still assembled after the fact, out of spreadsheets. Reiss Gavin of Charles Taylor InsureTech named the consequence: “Spreadsheets are flexible but lack control, resulting in data quality issues that require manual intervention.” Every manual intervention is a person-hour you pay for twice, once to produce the number and once to defend it.
Record each action on the claim as it happens and the reports build themselves. Take the five things a capacity provider asks for:
- Reserve adequacy and movement rationale. Today it gets reconstructed from handler notes at the end of each month. It should be a record written to the file as the moment the reserve moves, with the trigger that caused it captured alongside.
- Service standard performance. Usually it gets sampled by hand ahead of an audit. When every action on the file carries a timestamp, performance against your service standards becomes a live number you can read at any point in the month.
- Claims handled outside authority. Today you find them during the audit. If authority limits sit inside the workflow, the claim stops or escalates at the decision point, so there is nothing to discover months later.
- Bordereaux in the Lloyd’s core data standard. Today it means mapping and cleansing spreadsheets every month. When the claim record already holds the required fields in the required format, the bordereau comes straight out of it.
- Evidence of consistent handling across handlers. Usually it rests on anecdote and QA sampling. Every file should be measured against the same encoded SOP, so consistency becomes something you can show across the whole book.
That last row compounds. When the SOP lives in the workflow instead of in a handler’s head, a new joiner’s first file follows the same path as a 20-year veteran’s. Consistency stops depending on tenure, which is what makes growth survivable on a thin hiring pipeline.
The guardrail that makes this defensible: the automation triggers the rule, the handler makes the call. An auditor asking who decided what, and why, gets a traceable answer with the source documents attached. We’ve set out the autonomy levels and the guardrails each one needs separately.
How to sequence the rollout for adding claims capacity
Order the work by volume and judgment content. Automate the highest-volume, lowest-judgment work first, prove the accuracy holds, then raise autonomy a step at a time.
Capgemini’s World Property and Casualty Insurance Report 2026 found only 10% of P&C insurers have successfully scaled AI, 60% remain in exploration or proof-of-concept, and investment splits 72% to technology and infrastructure against 28% to change management and training. Programmes stall on the 28%.
MGAs have a structural advantage here: fewer legacy systems, tighter decision-making, and a book specific enough to configure against precisely. Tego, an Australian medical malpractice specialist, went live in nine weeks. Qover, a pan-European embedded insurance enabler working across 32 countries, went live in 16 weeks and cut the time to launch a new line of business from around two months to one or two days.
How long your own rollout takes depends on your integrations, the state of your claim data, and how well documented your handling standard already is. The order matters more than the calendar, because each step gives the next one cleaner inputs.
Two things decide whether this works. Confirm your capacity providers’ reporting requirements in writing on day 1, because retrofitting a data field on day 80 costs ten times what it costs on day 20. And put a claims person in charge of encoding the SOP, because the encoding is the product and only a handler knows what the procedure actually is.
Step | What to automate | Where autonomy sits | Guardrail to set first |
|---|---|---|---|
Baseline | Nothing yet. Measure touches per claim, active work time, and inbound mail volume per handler | Handler does everything | Agree the metrics with your capacity providers before you change anything |
Intake and documents | Classify inbound mail, index attachments to the right exposure, pre-fill FNOL | Automation acts, handler reviews | A confidence threshold below which a file routes to a person |
Missing information | Issue and chase the standard information requests on your rules | Automation acts, handler is notified | A named exception owner for every request type |
Authority and escalation | Enforce authority limits at the decision point | Automation stops the file, handler decides | Hard authority ceilings, with every override logged |
Reporting | Generate bordereaux and service standard reporting from the claim record | Automation produces, handler signs off | Run it alongside your existing process for one full reporting cycle |
Extend | Take the next-highest-volume task and repeat | Raise autonomy only where accuracy has held | A regular accuracy review owned by your claims lead |
Two things decide whether this works, whatever pace you set. Confirm your capacity providers’ reporting requirements in writing before you configure anything, because adding a data field late costs many times what it costs at the start. And put a claims person in charge of writing the SOP into the system, because getting that procedure right is the whole job, and only someone who has handled files knows what the procedure actually is.
How Odie cut reimbursement time by almost 75%
Odie, an American pet insurance MGA, automated the administrative layer around its claims and cut the time a member waits for reimbursement by almost 75%, without moving the judgment calls away from its claims team.
The case study reports a reduction in reimbursement time of almost 75%. Read that as a capacity number as much as a service number, because every hour that comes out of the cycle is an hour a handler would have spent keying, chasing, or assembling, and it comes out of every claim in the book at once.
“With Five Sigma, we are able to automate many processes and utilize the latest AI advancements for useful recommendations, without losing the reassuring human touch that our claims teams bring to our members.”
Miles Thorson, Co-Founder and CEO, Odie Pet Insurance
Where Clive AI fits an MGA claims desk
Everything above is an operating-model argument, and you can run at it with whatever tooling gets you the result. What an MGA needs specifically is the administrative work removed, the SOP applied identically on every file, and the evidence produced as the claim moves.
That’s what Clive™, Five Sigma’s agentic AI claims expert, is built to do. Clive reads and classifies inbound mail, indexes documents to the right exposure, flags what’s missing and what looks inconsistent, applies your SOP to advance the file, and brings a handler in where judgment is required. It runs inside the Five Sigma CMS or on top of the claims system you already have, which matters when a capacity provider’s platform is not yours to replace.
The design constraint is deliberate: Clive triggers the rule, the adjuster makes the call.
Key takeaways
- Capacity comes from removing the work around the decision. Five of the six activities on a new claim need no human judgment.
- Agentic AI differs from an assistant by acting on the file continuously instead of responding when asked. The saving is counted in handler touches removed.
- MGAs can automate further and faster than multi-line carriers because a single specialism has a knowable, encodable SOP.
- Supervisor span is capacity too. Automated claim completion monitoring lets one supervisor carry more handlers, so you avoid hiring in a second row.
- Record each action as it happens and bordereaux become a by-product of handling. Lloyd’s made Claims Management a hurdle Principle in January 2026.
- Sequence beats speed. Automate the highest-volume, lowest-judgment work first, and set each step’s guardrail before you start it.
Want to see what agentic automation does to your capacity math?
Frequently asked questions
How can an MGA handle more claims without hiring more adjusters?
By moving intake, document indexing, information chasing, and reporting onto agentic automation, so handlers spend their time on coverage, liability, and quantum. Bain estimates generative AI in P&C claims can cut loss adjusting expenses by 20% to 25%.
What is the difference between an AI assistant and agentic AI in claims?
An assistant responds when a handler asks it something. Agentic AI works the file continuously, applying your SOP to decide and execute the next action, and escalating to a handler where judgment is required. The capacity gain comes from fewer handler touches.
What should an MGA automate first in claims?
Intake and document handling, the highest-volume and lowest-judgment work on the desk. Once inbound mail is classified and attachments are indexed to the right exposure automatically, handler time moves to coverage and liability with no change in headcount.
Does claims automation weaken oversight of delegated authority?
It strengthens it when built correctly. Every automated action is a timestamped event with a recorded rule and source document, so audit evidence and bordereaux generate from the claim record instead of being reconstructed from handler notes.
How long does it take an MGA to implement agentic AI in claims?
Who makes the decision when agentic AI handles a claim?
The adjuster. The automation triggers the rule and prepares the file; coverage, liability, quantum, and settlement stay with a licensed handler, with authority limits enforced at the decision point and every override logged.
