An underwriter opens a new submission and spends the first hour on work that is not underwriting. Loss runs get pulled, values get re-keyed into another system, and a guideline gets hunted down before the first judgment call. The queue grows while the part of the job that needs an underwriter waits.
AI helps with that, but the evidence shows where. In the Underwriting Edge 2026 survey of 350 senior commercial P&C underwriters, 51 percent said AI's greatest contribution has been saving time on manual admin. Only 21 percent said it has improved the quality of their decisions.
The survey was published by an underwriting software vendor, so read it as one data point.
This article explains what an AI copilot for underwriters does, where underwriter time goes, how a copilot raises productivity, and how to measure the gain on your own book instead of borrowing a headline percentage.
What Is an AI Copilot for Underwriters?
An AI copilot for underwriters is software that assists with the preparation work around a decision while the underwriter keeps the decision. It reads submissions, pulls related data, summarizes what matters, and drafts next steps. A person reviews every output.
The definition matters because the word "copilot" is used loosely. A clear definition lets you judge any tool against the same test: who decides?
What a Copilot Does
A copilot takes on the work that sits between a submission arriving and an underwriter being ready to decide. The table below shows the common tasks and who owns the outcome.
Copilot task | What the underwriter receives | Who decides |
|---|---|---|
Summarize the submission | A short view of the risk and its open questions | Underwriter |
Pull related data | Policy, loss, and guideline information in one place | Underwriter |
Flag gaps | Missing or inconsistent fields before review starts | Underwriter |
Draft correspondence | A first version of the broker reply or referral note | Underwriter |
What a Copilot Does Not Do
A copilot does not bind, decline, or price a risk on its own. It suggests, and a person approves. That split matches what underwriters in the same survey described as their ideal setup: AI takes on data, triage, and drafting, while the underwriter keeps complex risks and the final call.
Where a Copilot Sits
A copilot works best inside the systems an underwriter already uses. A standalone tool adds one more screen to open, which adds to the time problem it was bought to solve.
Where Does Underwriter Time Actually Go?
Productivity gains come from hours a copilot returns, so the first job is to know where hours go. Most underwriting teams have never measured it. They know the queue is long, but they cannot say which part of the day produces the delay.
The Time-to-Decision Ledger
We use a four-bucket ledger to sort an underwriter's time on a single file. Only one bucket holds the judgment call.
Bucket | What it includes | Copilot role |
|---|---|---|
Gather | Finding submission documents, loss runs, and policy history | Assembles the file in one view |
Prepare | Re-keying, reconciling, and standardizing data | Hands over data in consistent structures |
Decide | Evaluating the risk, applying guidelines, setting terms | Supplies context; the underwriter decides |
Communicate | Broker replies, referrals, and file notes | Drafts for review |
Gather and Prepare are the buckets a copilot assists most directly. Decide stays with the underwriter in every case.
Why Admin Dominates
Manual data entry across systems tops the list of workflow blockers in the survey, named by 42 percent of underwriters. Submission data arrives as broker emails, PDFs, and spreadsheets, and policy and claims history sit in separate systems. Every switch between them costs time.
The same gap that slows a desk also splits decisions. Two underwriters working from different versions of the same facts reach different terms, and neither can point to the guideline that explains the difference.
Saved Time Is Not the Same as Better Decisions
The survey's 51 percent and 21 percent figures describe two different results. The first is throughput: more files handled in the same day. The second is quality: files decided with better information.
A copilot that only saves admin hours lifts throughput. A copilot that also puts consistent data and comparable context in front of the underwriter lifts decision quality. Productivity needs both.
How Does a Copilot Increase Underwriting Productivity?
A copilot raises productivity through the four buckets, and each bucket has its own mechanism and its own measure. Treating them separately keeps the gain honest.
Gather: Assemble the File Once
Gathering is search work. The underwriter opens the broker email, the attachments, the policy system, and the claims history to build a picture of the risk. A copilot assembles that picture once, in one place, so the underwriter starts from a complete file.
The measure here is time to a complete file. Count the minutes from submission arrival to the moment the underwriter has everything needed to begin.
Prepare: Hand Over Consistent Data
Prepare is the re-keying and reconciling that fills the gap between raw documents and usable data. When data reaches the underwriting system in the same structure every time, the underwriter stops fixing it and starts reading it.
The measure is touches per file. Count how many systems and screens an underwriter opens before deciding.
Decide: Put Context at the Point of Decision
Decide is where the underwriter applies judgment, and it is the bucket a copilot should never take over. A copilot assists here by surfacing the guideline that applies and the comparable prior risks, so the call rests on shared references instead of memory.
The measure is the override rate. When an underwriter changes a suggestion, that override records the judgment the tool did not have.
Communicate: Draft, Then Review
Communicate covers broker replies, referrals, and file notes. A copilot drafts the first version from the file, and the underwriter edits and sends it. The underwriter keeps the relationship and the tone.
The measure is time from decision to response. Faster replies matter to brokers, and drafting removes the blank-page delay.
How Do You Measure Copilot Productivity?
Measure on your own book, in your own workflow, against your own baseline. A percentage from someone else's deployment answers a question about their operation.
Why Public Percentages Do Not Transfer
Headline productivity figures measure different things. Premium per employee, hours per submission, and quote turnaround time share no common base, so two figures that look comparable are not.
A figure also reflects one book, one line of business, and one workflow. Before you quote any number, check its population, its timeframe, its exact metric, and its date.
The Scorecard
Use a short scorecard that maps to the ledger. Each metric names a bucket, so a gain in one place cannot hide a loss in another.
Metric | Ledger bucket | What it shows |
|---|---|---|
Time to a complete file | Gather | How fast the underwriter can start |
Touches per file | Prepare | How much switching remains |
Submission-to-quote time | All four | End-to-end cycle time |
Submissions reviewed per underwriter per week | All four | Throughput |
Override rate on suggestions | Decide | Where judgment adds what the tool lacks |
Referral rate | Decide | Whether consistency is improving |
Set a Baseline First
A baseline turns a vendor claim into your own result. The steps below take a single desk from no data to a first comparison.
Pick one line of business and one underwriting desk.
Time a sample of recent files across the four buckets.
Record how many systems each file touched.
Run the copilot on new files for a fixed review window.
Compare the same metrics on the same desk.
How Do You Roll Out a Copilot Without Weakening Underwriting Judgment?
Rollout decides whether the productivity gain holds. The failures are rarely technical. They come from scope, accountability, and data.
Start Narrow
Begin with one line of business and one desk. A narrow start lets you validate the scorecard, test the review process, and see where the data breaks before the copilot reaches the whole team.
Keep a Person Accountable
Every suggestion needs a named underwriter who reviews it. An underwriting decision has to be defensible to a broker, a reinsurer, and a regulator, and a suggestion from a tool is not a decision.
Fix the Data Path First
A copilot is only as consistent as the data it reads. If policy and submission data arrive in different shapes from different systems, the copilot inherits the inconsistency. Settle how source data reaches your core system before you judge the copilot's output.
Avoid the Two Common Failures
The first failure is a copilot that sits outside the workflow, so underwriters must leave their system to use it. The second is measuring only time saved. Pair every time metric with a decision-quality metric, such as override rate or referral rate.
How InsOps Helps
InsOps builds an insurance-trained AI that assists underwriting teams with the data behind each decision. LiLa, our insurance-trained LLM, runs inside your own environment, so PII and PHI never leave controlled infrastructure. A person reviews and validates every mapping before it is deployed.
Our Integration Gateway connects to Guidewire PolicyCenter, UnderwritingCenter, and Quoting Services, so policy and submission data from your source systems arrives in consistent Guidewire structures without custom engineering for each source. That addresses the Prepare bucket in the ledger. InsOps migrates legacy data into Guidewire and keeps it flowing in real time.
Summarizing submissions, surfacing comparable prior risks, and drafting broker correspondence are capabilities InsOps is building toward inside LiLa. They are not shipped features today, and none would replace an underwriter's judgment. The underwriter reviews every suggestion and makes the final call.
If you are evaluating how to raise underwriter productivity without taking judgment away from your underwriters, contact us to talk through what this could look like for your operation.
Frequently Asked Questions
What is an AI copilot for underwriters?
An AI copilot for underwriters is software that summarizes submissions, pulls related data, and drafts next steps while the underwriter makes every decision. It assists with preparation and does not bind, decline, or price a risk on its own.
Why does underwriter productivity matter?
Underwriter time is limited capacity, and every hour spent gathering and re-keying data is an hour not spent evaluating risk. Returning those hours lets underwriters spend more of the day on judgment and broker relationships.
How do you measure the productivity gain from an AI copilot?
Set a baseline on one desk, then compare the same metrics after a fixed review window. Track time to a complete file, touches per file, submission-to-quote time, submissions reviewed per week, and override rate on suggestions.
Will an AI copilot replace underwriters?
No. A copilot prepares information and suggests next steps, and the underwriter reviews and decides. Underwriters in the Underwriting Edge 2026 survey described the same split, with AI handling data, triage, and drafting while the underwriter keeps complex risks and the final call.
What are the common challenges when adopting an underwriting copilot?
The most common challenges are a copilot that sits outside the underwriter's workflow, inconsistent source data, and measuring only time saved. Address each one before expanding beyond the first desk.
How does InsOps assist underwriting teams?
InsOps assists with the data behind each underwriting decision. LiLa runs inside your own environment, and the Integration Gateway delivers policy and submission data into Guidewire in consistent structures, with a person validating every mapping before deployment. Submission summaries and comparable-risk context are capabilities InsOps is building toward, not shipped features today.
How long does a copilot pilot take, and what goes wrong?
Plan for a baseline period, then a fixed pilot window on one desk and one line of business, then a review. The first read shows where time goes, not a final number. The common pitfall is skipping the baseline, which leaves nothing to compare against.