AI for insurance underwriting helps commercial carriers process more submissions, quote faster, and reserve underwriter judgment for complex risks. Intelligent automation combines simple, rules-based automation with AI so underwriting work moves through the pipeline without a person touching every step. The rules-based layer is the older half of that pairing: software bots that follow a fixed, pre-programmed sequence of steps to move data between systems the same way every time, such as copying a field from a submission email into a rating spreadsheet. AI is what handles everything a fixed script can't: judging whether a submission fits appetite, interpreting a messy PDF, or deciding what to do with an exception. For commercial P&C carriers, the payoff can be measurable: fewer manual touchpoints, faster quote-to-bind, and underwriters who spend more time assessing risk instead of rekeying it.

Three factors decide whether that payoff shows up. Workflow determines how work moves between systems and people. Data quality determines whether the inputs feeding that workflow are usable. Decision rules determine which cases a machine can resolve and which ones need an underwriter's judgment.

How underwriting workflow automation works

Celent describes intelligent automation as this kind of rules-based automation paired with AI, positioning it as the successor to standalone rules-based bots (Celent). The distinction matters in practice. Rules-based automation handles well-defined, repeatable tasks. AI-augmented automation can handle workflows that change as conditions change and processes that require reasoning.

McKinsey's AI progression groups these capabilities into three tiers. Analytical AI finds patterns in data that help carriers identify appetite fit. Generative AI interprets unstructured formats and cuts the time needed to process emails and PDFs. Agentic AI plans and executes multiple steps within governed workflows, increases submission capacity, and routes exceptions to underwriters.

Over 80% of insurers already have this kind of rules-based automation in-house. Most carriers now need to decide how far up this stack to move and what controls to apply.

McKinsey estimates that even in large commercial lines, underwriters spend 30 to 40% of their time on administrative tasks such as rekeying data. That leaves underwriters with less time for the risk assessment work that actually requires their judgment.

Where submission automation creates value

The combined submission triage workflow is a major source of administrative work. Manual submission processing spans four sequential steps:

1. Receive and triage the submission.

2. Extract the data.

3. Validate the extracted fields.

4. Assess the risk against appetite.

Submission volume regularly exceeds underwriting teams' capacity for thorough review at each step. Automation compresses the sequence: extraction and validation tools can process submissions across multiple lines of business while supporting underwriter adoption rather than replacing underwriter judgment.

How much that compression helps depends on the underwriter's workflow. High-volume, low-touch lines see the biggest gains, since compressing four manual steps into one matters most when an underwriter is working through dozens of submissions a day.

A lead underwriter pricing a large, complex risk, or one who already carries most of the relevant knowledge without needing to consult external systems, gets less from faster extraction and validation. The same logic applies to renewals versus new business: a heavy-renewals book with familiar accounts has less friction to compress than a book weighted toward new submissions.

One carrier reached quoting and binding in under five minutes with no human involvement for risks under $50,000, according to Datos Insights, by codifying its underwriting rules and decisions in full. Carriers using extraction and validation automation can reduce manual handling and increase the number of submissions a commercial lines underwriter processes each day, though the size of that gain tracks the variables above rather than applying uniformly across commercial underwriting.

Data: the layer that decides whether automation scales

Automation scales only when carriers can convert fragmented, unstructured submission data into accurate, standardized inputs. Commercial submissions commonly arrive as:

• PDFs

• Emails

• Spreadsheets

• ACORD forms

• Loss runs

• Broker notes

In commercial and specialty lines, underwriters spend a large share of their initial review time manually sifting unstructured PDF packets. Standards close part of the gap. Lloyd's Core Data Record aligns with ACORD technical standards and defines mandatory and conditional mandatory data fields.

AI extraction can then structure the source material, subject to validation and exception handling, which shortens intake time and increases quote capacity. The same extraction approach applies to loss runs across formats, though the accuracy gains still depend on the quality of the underlying workflow and data feeding it.

Decisioning: who, or what, makes the call

A decisioning framework separates three modes. Under decision support, the person decides. Under decision augmentation, the person and the machine decide together. Under decision automation, a machine decides.

Carriers apply these modes through two underwriting automation types: complete straight-through processing (STP) and partial process automation. A routing model sends work straight through when it clears quality thresholds and sits within defined appetite. It can also prepare and refer, where AI assembles the file and recommends an action but an underwriter decides, or send complex risks and incomplete data to specialist review.

Carriers should reserve full automation for lower-risk, standardized processes where speed matters most. STP adoption is furthest along in personal lines and standard commercial lines, while adoption in complex product lines remains low. Lloyd's syndicates, MGAs, and reinsurers can apply the same routing model while adapting authority thresholds and audit requirements to their own delegated arrangements.

Why implementations stall

AutoRek's 2026 Insurance Report found three recurring barriers to scaling AI in carrier operations: legacy system integration challenges, cited by 42% of firms; fragmented data environments, cited by 39%; and a shortage of in-house AI expertise, cited by 40%. Over half of surveyed firms described their data governance frameworks as early-stage or developing.

These gaps compound each other. A carrier with fragmented data cannot fully trust straight-through processing, and a carrier without in-house AI expertise struggles to build the integration layer that fragmented data requires.

Underwriter skepticism toward automation is often treated as a change-management problem, something to overcome through training and communication. hyperexponential's 2026 Underwriting Edge report, a survey of 350 CUOs, heads of underwriting, and senior underwriters conducted by Coleman Parkes, backs that skepticism with data: where AI has already reached underwriters, 51% say its biggest contribution so far is time saved on admin, but only 21% say it has improved decision quality. Part of the skepticism is a product gap, not just a training gap. Many automation tools accelerate individual submissions but don't give underwriters a persistent, single view of account status and history across a book.

Without that view, the tool reads to senior underwriters as a faster calculator, not something they'd trust to run the account. Involving underwriters in model design and governance from the outset addresses the process side of that skepticism. Building the account-level visibility they're missing addresses the product side, and both matter.

Legacy integration should be assessed before any agent deployment, since it typically adds more complexity and cost than the automation project itself.

What underwriting teams can do next

McKinsey's Insurance Productivity 2030 research expects manual pricing and underwriting to largely disappear for personal and small commercial products by the end of the decade, while complex commercial and specialty lines keep meaningful underwriter involvement.

Carriers that treat workflow, data quality, and decision authority as one governed system get more from their automation investment than carriers that fund the three separately. Start with a single high-volume workflow, set quality and referral thresholds before deployment, and measure quote-to-bind, overrides, and portfolio outcomes before expanding scope.

How the hx platform executes work around underwriting decisions

Traditional underwriting interfaces manage and route underwriting work but do not execute it, which puts a ceiling on efficiency. The hx platform is an agentic platform for commercial insurance underwriting. It executes the work around decisions inside the carrier's own governed underwriting intelligence, and captured feedback supports continued refinement under carrier governance.

• Applies the carrier's own governed underwriting intelligence through agent-led action while the underwriter stays in command.

• Provides a native Python environment for pricing and rating. Aviva built 20 pricing models within nine months of implementation, one of several results detailed in hx's customer stories.

• Ingests relevant data through APIs and applies in-model underwriting safeguards at the point of pricing, so underwriters stop rekeying it.

• Records every action taken in the platform, producing a decision trace that shows exactly what happened and why, rather than one reconstructed after the fact.

• Uses captured underwriting actions as portfolio intelligence inputs, letting CUOs monitor appetite execution, spot referral patterns, and refine risk selection.

• Replaces fragmented spreadsheets and legacy rating tools with a single connected workflow, without carriers needing to replatform their core systems.

Book a demo to explore how hyperexponential helps commercial underwriting teams cut rekeying and shorten quote-to-bind with governed, agent-led pricing and rating.

FAQs about intelligent automation in insurance

How should carriers measure intelligent automation performance?

Measure each stage separately. For intake, track cycle time, extraction accuracy, exceptions, and submission volume. For underwriting, monitor referrals, quote turnaround, quote-to-bind ratio, and adoption. For portfolio outcomes, compare GWP, loss ratio, and combined ratio across automated and referred cohorts. Pair speed metrics with override rates and decision-trace completeness so faster processing doesn't mask weaker risk selection.

How should carriers validate third-party AI vendors?

Require vendors to explain how their components contribute to a decision rather than treating the vendor itself as the explanation. Test vendor-supported recommendations and exceptions against carrier appetite and actual outcomes, and confirm every vendor-supported action appears in the decision trace so carriers can investigate its role in individual decisions.

How should actuaries and underwriters divide responsibilities in automated workflows?

Actuaries build pricing models, rate structures, and deployment controls. Underwriters apply those models and keep bind or decline authority where judgment is required, and they feed market and risk-selection intelligence back to actuaries for model refinement. Technology leaders support the architecture, APIs, and deployment underneath both roles. This division keeps model integrity and binding authority with the people who own them.

How often should automation rules be reviewed?

Review automation rules on a regular cadence using override rates, referral patterns, portfolio outcomes, and underwriter feedback. Changing market conditions may require carriers to adjust appetite, authority, data-completeness, or model-confidence thresholds. Every update should stay subject to carrier governance, with rule and model versions recorded so teams can trace what changed, who approved it, and how earlier decisions were made.

What should an underwriting decision trace contain?

A decision trace should log when new information arrives mid-decision, such as an updated loss run or a refreshed third-party data source, and whether the relevant underwriter or broker was notified at that point. Without this, a recommendation can rely on stale data even if every earlier step was recorded correctly. Effective traces treat notification timing as part of the audit record, not just the data itself.