The problem, in the words we hear on calls

"Our underwriters are typing, not underwriting." That is the most common version. A new group submission arrives as an email, a twelve page application PDF, a provider roster in a spreadsheet, five years of loss runs from the prior carrier and a few pages on board actions or prior claims. Someone keys all of it. Only then does anyone find out the group includes a specialty you do not write in that state.

Accenture's underwriting research found that more than a third of an underwriter's time still goes to non-core work such as data collection and administration (Accenture, 2024 survey, published 2025). In MPL the cost of that time is harder to absorb than it used to be. AM Best reported that insurers focused on MPL posted a collective underwriting loss of $712 million in 2025, up from $546 million in 2024, while direct premiums written grew only 3.6% to $9.4 billion (AM Best, 2026). The NAIC puts the 2025 direct loss and defense and cost containment (DCC) ratio for the line at 75.6%, up from 2024 (NAIC, 2026).

When losses and DCC eat three quarters of premium, the expense side is the lever management actually controls. And most of the expense is people reading documents.

The second version we hear: "We know the answer is in the file somewhere." Prior acts dates, tail endorsements, the claims-made retroactive date on a lateral hire, the consent-to-settle language on a specific form. The facts exist. They sit in PDFs and scanned letters instead of fields, so nobody can query them.

What AI-native MPL carriers have already shown

The market has a proof point. Indigo, an MPL carrier founded in 2023 and built around its own AI platform, announced in November 2025 that 20% of its submissions were being underwritten automatically with no human involvement, and that five underwriters were on pace to handle more than 7,000 submissions that year. Human underwriters still handle exceptions and edge cases (Indigo via Insurance-Canada.ca, 2025). That is a real result.

Most carriers reading this page cannot rebuild their company around a new platform. They have a policy admin system that has run for a decade, a claims system, underwriters with twenty years of judgment, and a book they cannot put at risk. The question for an incumbent is how to get the same kind of throughput on top of what already runs. That is the problem OC is built for.

What OutcomeCatalyst does for an MPL insurer, step by step

OC is delivered and run by our team. You do not buy a license and staff a data project.

1. Name the unit. Before anything is built, we agree on the countable thing the agent will move: hours of keying per submission, submissions quoted per underwriter per week, days from receipt to quote, or open claim files re-reviewed per month. If we cannot name it, we do not start.

2. Connect the systems you already run. We read from the policy admin system, the broker and agent inbox, rating worksheets, the claims system and the document store where they sit. Those connections feed a knowledge graph we call the brain, which links a physician to their policy, their group, their specialty class, their claims and every document that mentions them.

3. Write down how your veterans decide. We sit with your senior underwriters and claims managers and capture the rules they apply without thinking: which specialties need a referral, how a prior paid claim changes a schedule credit, when a gap in coverage needs a letter. Those rules become the agent's instructions, and your team signs off on them.

4. Deploy the agents on real volume. The agents work live submissions and files, show their reasoning and the source page for every fact, and put each item in a queue for a person to approve, change or reject.

5. Measure the unit and tune. We report the unit we named in step one against the baseline, and we adjust rules where your people overrode the agent.

Submission intake and triage

The submission and triage agent opens every new application as it lands. It pulls the named insured, the specialty mix, the states, limits requested, the current carrier and the retroactive dates out of the application and the roster. It reads the loss runs and lists each prior claim with status, paid and reserved amounts. Then it checks the whole submission against your appetite and flags what is outside it, missing or inconsistent, such as a roster count that does not match the application or a loss run that stops a year short.

Out-of-appetite submissions get a draft decline for the underwriter to send. In-appetite submissions arrive in the policy admin system already keyed, with a short summary of what a senior underwriter would look at first. The underwriter still prices and binds. They just start at the judgment step instead of the typing step.

Policy file integrity

MPL policies change shape constantly. Physicians join and leave groups, take leave, retire and need tail, or move from claims-made to occurrence. The policy file integrity agent compares what the policy admin system says against what the endorsements, applications and correspondence say: limits, retroactive dates, named providers, extended reporting periods. Where they disagree, it opens an item with both sources side by side. This matters most before a reinsurance renewal or a market conduct exam, when somebody has to prove the book matches the system.

Claims file review

MPL claims run for years and the file grows the whole time: notice of incident, the complaint, expert reviews, defense counsel reports, invoices, settlement correspondence. The claims agent re-reads open files on a schedule and when new documents arrive. It flags reports that suggest the reserve no longer fits, invoices that do not follow your litigation guidelines, coverage questions that were never answered in writing, and, where they exist, recovery opportunities against another party. In MPL that last item is less frequent than in auto or property, so for most carriers the bigger value is reserve accuracy and DCC control. Your claims professionals decide every one.

The systems it reads

• Your policy admin system: policies, endorsements, insureds, rating and billing records.

• The broker and agent inbox: new business and renewal submissions with their attachments.

• Applications, provider rosters, loss runs and supplemental questionnaires, in PDF, scanned image or spreadsheet form.

• Rating worksheets and underwriting guidelines, including the spreadsheets underwriters actually use.

• Your claims system and the claim documents: incident reports, counsel reports, expert reviews, invoices.

• Your document management system, where the history of each insured usually lives.

During scoping we confirm how each system exposes data (API, database read, scheduled export or document folder) before any build starts.

What stays human

Pricing, binding, declines, reserve changes, coverage positions and settlement decisions stay with your licensed people. The agents prepare, check and recommend. Every recommendation carries the page and line it came from, so an underwriter can verify a fact in seconds instead of trusting a score. Overrides are logged and become the next round of rules. We think this is the right split for MPL in particular: a medical malpractice policy is a long relationship with a physician, and the judgment calls are where your carrier earns its loss ratio.

What to measure

Underwriting expense ratio is the board-level number, but it moves slowly and has many inputs. Track the units underneath it:

• Minutes of keying per submission, before and after.

• Submissions quoted per underwriter per week, which is where capacity shows up.

• Days from receipt to quote, which brokers notice first.

• Share of out-of-appetite submissions declined before keying.

• Policy file discrepancies open and closed per month.

• Open claim files re-reviewed per month, and reserve changes that came from a flagged document.

Pick one lead unit. Projects that chase all six rarely prove any. For a longer walkthrough of the underwriting side, read our insurance underwriting automation guide.

How a project starts

It starts with a strategy call where you describe the bottleneck and we ask what you would count to prove it moved. If there is a fit, we scope one workflow, usually submission triage because the volume is visible and the baseline is easy to measure. We confirm system access, write the rules with your senior people, and run the agent on live submissions with your team approving every output. Once the first unit moves, the same brain supports the next agent, because policy files and claims draw on the same connected data.

Frequently asked questions

Do we need a data team to run this?

No. OC builds, runs and maintains the connections, the knowledge graph and the agents. Your team's job is to approve agent work and tell us when a rule is wrong. An IT contact for system access during setup is enough.

Do you copy our data into a new warehouse?

No. The brain reads from your policy admin, claims and document systems where they already sit and keeps the links between records. Your systems stay the record of truth.

Can it work with our policy admin system?

That is the first thing we check in scoping. We connect to policy admin and claims systems through whatever access they offer, whether an API, a database read, a scheduled export or a document folder, and we confirm the route for your specific system before anything is built.

Will the AI price or bind policies on its own?

Not unless you decide it should, and we would start with it off. Out of the box the agent prepares the submission, checks it against appetite and recommends. A licensed underwriter prices and binds. Some carriers later choose to let narrow, low-risk renewals pass with lighter review. That is your call, made on evidence from the approval logs.

Why not build this ourselves?

Some carriers can. The hard part is not the model. It is connecting a decade of policy admin data to the documents around it, writing down how your best underwriters decide, and keeping it running after go-live while forms, rules and staff change. We wrote up the trade-offs in buy vs build an AI context layer.

What happens after go-live if people don't use it?

Then the project failed, and we count that as our miss. We watch approval rates and overrides, sit with the underwriters who skip the queue, and change the agent until it saves them time. Adoption stays inside the engagement after go-live.

What are your security and compliance standards?

OutcomeCatalyst is HIPAA-aligned and SOC 2 Type 2 aligned, with the formal audit underway and expected to complete before year-end. MPL claim files often contain protected health information, and we scope access, logging and data handling with that in mind from the first call.

Sources

• Accenture, "AI underwriting: Beyond the hype," Insurance Blog, September 2025, citing Accenture's 2024 underwriting survey: insuranceblog.accenture.com

• AM Best, Best's Market Segment Report on US medical professional liability 2025 results, as reported by Captive.com, May 2026: captive.com

• NAIC, Medical Malpractice Insurance topic page, last updated April 1, 2026: content.naic.org

• Indigo, "Indigo Reaches Milestone: 20% of Submissions Now Fully Underwritten by AI," November 3, 2025, as published by Insurance-Canada.ca: insurance-canada.ca; original release on Business Wire

Book a strategy call to see this on your own book.