Most AI in underwriting has only sped up admin. Here's why the systems underneath aren't ready for agents yet, and what arms underwriters with real decision context.
Underwriters can only analyze a risk based the data they have at their fingertips. More often than not, that's not the full picture of a risk.
Underwriting data analysis uses submission data, loss runs, exposure schedules, third-party risk data, and portfolio performance signals to support a single underwriting decision, what to quote, what to charge, whether it fits the book, assembled and surfaced at the point an underwriter needs it rather than something they have to go find themselves.
The data underwriting data analysis draws on already exists inside most carriers. What's usually missing is the structure and context that turns it into something usable at the moment of decision, and that's exactly where Underwriting Edge 2026, hx's survey of 350 senior commercial P&C underwriters across the UK and US, finds the real gap. Two-thirds of underwriters name a context problem, inconsistent submission data, no visibility into similar prior risks, limited book-level visibility, among the three biggest drags on their decision quality. Given a straight choice between a setup that runs faster and one that surfaces more context and reasoning, underwriters chose the context-rich option every time, by a margin of at least 60%, rising to 73% for a live signal on book-level drift over a report that just refreshes itself.
Why underwriting data analysis is mostly admin today, not analysis
Manual work consumes 49% of a typical underwriter's week, submission review, re-keying, system navigation, reporting, almost equal to the judgment-led work of pricing, broker conversations, and portfolio review combined. Given the choice, underwriters would cut that to 36%, with re-keying and admin alone dropping from 17% of the week to 8%.
Asked what actually drives the inefficiency, underwriters point to fragmentation, not volume: manual data entry across systems (42%), switching between disconnected systems (39%), and unstructured submission data from brokers (36%) top the list. The data underwriters need to analyze a risk properly exists. It's just scattered across spreadsheets, siloed systems, and documents that were never standardized, with no shared layer connecting any of it to the decision in front of them. A related hx survey of the same population, covered independently by Carrier Management, found 85% of actuaries still rely on spreadsheets as a core part of their pricing stack, with underwriters separately citing pricing models that are inaccurate or out of date among their top barriers to good decisions.
What arming underwriters with full risk context would change
Arming underwriters with context means putting the risk research, prior-case context, and third-party data they need directly in front of them at the point of decision, not just another dashboard, so they can form an accurate view of the risk, make a call they can stand behind, and steer the book in the direction the carrier wants.
Underwriters describe this future in their own words: an AI that "unifies messy broker data and pulls all historical similar risk cases instantly," one where "every submission arrives with relevant market and portfolio context," or simply "less time spent searching for data and prior decisions." One underwriter drew the distinction sharply: "AI delivers context, not just information."
The hurdles between underwriters and that context
Several barriers compound rather than acting alone. Inconsistent, unstructured submission data from brokers is the single biggest drag on decision quality (44% of underwriters name it in their top three). Pressure to bind quickly leaves little room for deep analysis (38%). A quarter of underwriters say they have no insight into how prior decisions actually performed, and 27% say they spend real time just hunting for context on prior risks. Limited visibility into book-level patterns (33%) means even underwriters who want to steer the portfolio often can't see where it's actually drifting.
These barriers compound into a chain: fragmented data forces underwriters to work without shared context, which leaves them deciding under pressure anyway.
Why most systems aren't ready for agents yet
Where AI has landed in underwriting so far is uncomfortable reading. Among underwriters already using AI, 76% say its value has shown up as either saving time on manual admin (51%) or automating workflow actions (25%). Only 21% say it's improved the quality of their decisions. Underwriters aren't skeptical that AI can take work off their desk, they believe that readily, but they're far more doubtful it can lift the standard of a decision, carry senior judgment from one person to another, or be trusted to price on its own, and that skepticism barely moves when asked to look three to five years ahead.
AI can only act as well as the context layer underneath it, and in most underwriting environments today that layer doesn't exist: data is fragmented across systems, submissions arrive unstructured, and prior-decision context lives in individual people's heads rather than anywhere an agent, or a person, could reliably reach it. Underwriters' own top concerns for the future bear this out directly: AI deciding too early, before a submission is properly understood (40%), and AI deciding inside a black box with no auditability (33%). Put an agent on top of fragmented systems and ungoverned context, and the result is faster mistakes, not better decisions. Most carriers aren't ready to hand meaningful work to agents yet because that governed, shared layer doesn't exist, not because the technology falls short.
A shared context layer is what closes the gap
The fix is an underlying layer that connects submission data, pricing logic, prior decisions, and portfolio signals into one place an underwriter, or eventually an agent operating under that underwriter's authority, can draw on, not a faster version of the same fragmented setup.
That's the distinction between a traditional underwriting workbench, a place underwriters went to do tedious work, and what's emerging as the agentic underwriting workbench: a governed environment where AI agents can monitor, prepare, triage, and coordinate work continuously, handling the assembly work so a prepared file reaches the underwriter ready for the judgment call that requires their judgment. Calculation Engines capture a carrier's own bespoke pricing and rating logic. Workflow Builder lets a carrier tune how work moves through their own operations and lines. Portfolio Intelligence brings book-level context into every live decision, the same context that's currently invisible to 33% of underwriters trying to steer their book. Together they form the environment, tools, and context an agent needs to act safely, and the same thing an underwriter needs to decide well.
Carriers don't need to wait for a fully agentic future to start closing this gap. The foundation, specifically Calculation Engines and Portfolio Intelligence, is something to build now, so that whatever agentic orchestration layer comes next has something solid to run on, rather than compounding the same fragmentation faster.
A more profitable book, not just a faster one
A softening market changes what underwriting excellence is worth to a carrier. As profitability increasingly hinges on which risks get written and which get declined, a carrier is paying for the quality of the decision, not the volume of them. Underwriters themselves rank risk selection (48%), technical depth on complex risks (43%), and commercial instinct on price (38%) as what makes someone excellent, well above the administrative work that currently eats their week.
Arming underwriters with real context doesn't just save them time. It gives them what they say they're missing: the ability to judge a risk accurately, price it with confidence, and see how it fits the book they're building, which is what turns faster underwriting into a better-performing one.
How the hx platform builds that foundation
The hx platform gives carriers and MGAs a shared context layer across submission triage, pricing and rating, and portfolio intelligence, so the data an underwriter needs to assess a risk and the data that shows how it fits the book are drawn from the same place, not reassembled by hand each time.
Actuarial teams build and deploy pricing logic in native Python through Calculation Engines, with governed workflows replacing ad hoc spreadsheet logic. Portfolio Intelligence brings live book-level signals into the point of decision, the same visibility 33% of underwriters currently say they lack. API ingestion puts relevant submission and third-party data in front of underwriters at the point of pricing, cutting the re-keying that currently eats close to a fifth of their week. As hx's newest layer, hyperoperator, begins orchestrating more of that work directly, carriers who've already built this foundation are the ones positioned to put it to use safely and quickly.
Book a demo to explore how hyperexponential helps underwriting and actuarial teams build the shared context layer their decisions, and eventually their agents, will run on.
Frequently asked questions
Does more data actually improve underwriting decisions, or just slow underwriters down?
The evidence points to more context, not less, improving decisions, provided it's relevant and surfaced at the point of decision rather than requiring a separate search. In Underwriting Edge 2026, underwriters chose a context-rich option over a faster one by at least 60% every time they were asked, and inconsistent or missing context was the single most commonly cited driver of poor decision quality.
Why hasn't AI improved underwriting decision quality yet?
Most AI deployed in underwriting so far has targeted admin, summarizing documents, automating workflow steps, not the decision itself. Among underwriters already using AI, only 21% say it's improved their decision quality, versus 76% who say its value has been saving time or automating workflow actions. The constraint is that the systems underneath mostly lack a shared, governed context layer for either an agent or a person to reliably draw on, not the AI itself.
What's the difference between a traditional underwriting workbench and an agentic one?
A traditional workbench is a place underwriters go to do the work themselves, often manually, across disconnected systems. An agentic underwriting workbench lets AI agents handle the preparation, monitoring, and coordination work continuously, submission intake, triage, assembling context, so what reaches the underwriter is a prepared file ready for the judgment call, not raw, scattered inputs.
What should carriers build before trying to adopt agentic underwriting?
A shared context layer, connecting submission data, pricing logic, and portfolio signals, needs to exist before an agent can act safely inside it. In practice that means underlying capabilities like governed pricing logic and live portfolio visibility, since both determine whether an agent, or an underwriter, is working from an accurate, current view of the risk and the book, rather than fragmented or outdated information.
Will AI replace underwriting judgment?
Underwriters themselves don't think so, and don't want it to. Fear of being replaced by AI has fallen year over year across hyperexponential's underwriter surveys, and no part of the underwriting workflow draws majority support for full AI autonomy, even among tasks underwriters are most comfortable delegating. What underwriters consistently ask for instead is an AI that prepares and surfaces context, while they keep the complex risks and the final call.
