What the Future of P&C Insurance Underwriting Might Look Like

The future of P&C insurance underwriting is opening up. Insurers can now reach more data than any previous generation of underwriters could: telematics feeds, satellite imagery, building-level hazard scores, supplier records, and the unstructured contents of every submission that lands in an inbox. Exposures are moving faster than the loss histories behind them, and for the first time underwriters have the evidence to keep pace.

What follows looks at where P&C insurance underwriting could go next, and at how underwriting decisions and the underwriter’s job may change.  It draws on what insurers are already building and what analysts and regulators have put in writing.

From Traditional Underwriting to Data-Driven Risk Assessment

Traditional underwriting rests on three things:

1. historical loss data
2. the underwriter’s experience
3. manual assessment of a submission’s documents

That combination still anchors most risk selection today, and its limitation is that it looks backwards. A five-year loss run describes what happened to a similar risk in a static world, and describes less once the exposure starts to move.

What is changing is the evidence supply. Structured data still matters, but underwriters increasingly have access to unstructured material such as loss narratives, engineering reports, broker emails, and photographs.

They also have access to alternative data from outside the insurance record:

• IoT sensor output
• building permits
• court filings
• supplier concentration data

Having the data and being able to use it are two different things. Capgemini’s World Property and Casualty Insurance Report 2026 found that only 12% of insurers rate their own data as highly ready to use, at a point when the industry is leaning harder than ever on unstructured sources.

As those inputs become usable, risk assessment is likely to shift from description towards prediction. Rather than asking what a risk has cost before, a forward-looking model asks what it is likely to cost given its characteristics and the direction its exposure is moving. Data-driven underwriting also brings consistency. When the same data reaches every underwriter in the same form, and appetite rules are applied the same way each time, the spread of outcomes narrows.

How AI Could Change the Underwriting Process

Adoption has moved quickly. Celent’s GenAI-oneers research, which surveyed 55 property and casualty and life insurers, recorded active generative AI deployments rising from 8% in 2023 to 44% by May 2025. Underwriting has absorbed much of that capacity.

Submission analysis is the obvious first application. A commercial submission arrives as a bundle of email text, spreadsheets, loss runs, and scanned schedules, and much of an underwriter’s day goes on turning that into something comparable. AI-assisted submission analysis can read it, pull out the fields that matter, and flag what is missing before anyone even opens the file.

Automated data extraction and enrichment extend that idea outwards. Once the risk is identified, external sources could be attached without anyone asking: geocoding, hazard scores, financial health indicators, prior claims. The underwriter can then start from a fully assembled picture.

Risk classification and segmentation become finer as a result: with more consistent inputs, an insurer could sort its book into narrower segments and apply different appetite, pricing, and referral rules to each. Predictive risk assessment sits on top, estimating likely frequency and severity for a given risk rather than leaning on class averages.

None of this takes judgement out of the process. The most plausible version of AI insurance underwriting has the technology handling volume, retrieval, and pattern detection, while the underwriter decides what the pattern means. Models without human involvement reason poorly about risks they have not seen before, and they cannot be held accountable for a decision.

The Changing Role of the P&C Underwriter

If submission handling, data entry, and document chasing move to software, an underwriter’s day changes shape. Time released from repetitive administration is likeliest to flow into the work that is hardest to automate.

That means more time on complex risk interpretation: the accounts where the data conflicts, the exposure is unusual, or the broker’s story does not match the schedule. It can also mean more portfolio work. An underwriter who can see aggregate exposure across a segment in close to real time can make decisions about appetite, limits, and reinsurance, rather than only pricing risks one at a time.

Human oversight of AI-supported decisions is likely to become a named part of the job. Someone must review the cases a model refers, check its recommendations hold up, and notice when its behaviour drifts.

The skills question is live, and insurers are not yet comfortable with their answer. Capgemini’s 2026 P&C research also found that 67% of insurers cite a shortage of AI skills, and that 43% of employees name job security as a top concern. Future underwriters may need a mix that looks different from the traditional one:

• Read model output critically: understand what a score is built from, and where it is likely to be wrong
• Question data quality: recognise when an enriched field is stale, mismatched, or absent
• Explain a decision in plain language, to a broker, a regulator, or an auditor
• Work with technical teams: describe an underwriting problem well enough for someone to model it
• Hold the line on judgement: know when the answer is to decline, refer, or ask for more

Underwriting Risks That Are Harder to Assess

Some exposures resist data-driven assessment, and they’re among the fastest growing.

Emerging risks arrive without a loss history. AI liability, PFAS (the ‘forever chemicals’ now appearing in liability claims), novel battery technologies such as the lithium-ion storage systems now installed in buildings and on grid sites, and biometric privacy claims all must be priced before anyone knows what they cost. Pricing models are calibrated on the past, so a risk with a short record leaves the underwriter working from scenario analysis, expert judgement, and analogy.

Climate-related exposure comes with plenty of history, but that history may no longer describe the present. Swiss Re Institute estimated insured natural catastrophe losses at USD 107 billion in 2025, the sixth consecutive year above USD 100 billion, with wildfire losses reaching a record USD 40 billion. When the frequency and the severity of a peril are both moving, a long average is a weaker guide than it looks.

Complex commercial risks compound the difficulty. A multinational manufacturer or logistics account carries exposure across property, liability, marine, cyber, and business interruption at once, and the interactions between those lines are rarely visible in any one of them.

Interconnected risk is hardest of all, because correlation defeats the assumption that losses occur independently. Lloyd’s modelled a hypothetical cyber-attack on a major financial services payments system and put potential global economic losses at USD 3.5 trillion over five years. Events shaped like that produce simultaneous claims across an entire portfolio.

The failure mode is the same in each case. Traditional underwriting models work well when the future resembles the past and risks behave independently. Where exposures change rapidly, or move together, those assumptions stop holding, and more data alone does not repair them. What may help is a faster feedback loop: models re-fitted as data arrives, and underwriters close enough to them to notice when one has stopped working.

What Could the Future Underwriting Workflow Look Like?

Put the pieces together and a plausible workflow emerges:

• Digital submission: the risk arrives structured, or is converted on receipt, so nothing depends on a human retyping it
• Automated triage: the submission is checked against appetite, completeness, and capacity, then routed, held, or declined
• AI-assisted risk assessment: internal and external data are assembled, a score and an indicative price are produced, and the reasoning is recorded
• Human review: the underwriter accepts, adjusts, or overrides, with the referral threshold set by the insurer rather than the model
• Automated policy workflow: quote, bind, document generation, and downstream updates run without re-keying
• Continuous risk monitoring: exposure is tracked through the policy period, so material change is picked up before renewal rather than at it

The review step is where regulation is concentrating. The NAIC’s Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in December 2023, asks insurers to maintain a written AI systems programme covering governance, risk management, and internal controls. On the NAIC’s adoption record as of 31 August 2026, it had been adopted in 25 states and the District of Columbia. Documented human review is becoming what a regulator expects to see.

Key Takeaways: Preparing for the Future of P&C Underwriting

Technology and data are changing underwriting from the inside. More of the input is assembled automatically, more of the routine decision is handled by an AI model, and more of your underwriters’ attention goes to the cases that don’t fit.

Human expertise stays central and arguably becomes more concentrated. Judgement on unfamiliar risk, accountability, and the ability to explain a decision are things no model can supply.

The capabilities worth developing are less glamorous than the technology:

• clean and well-governed data
• core platform that can carry automated workflow from submission to renewal
• model governance that will survive a regulatory examination
• training path that helps your underwriters work alongside these tools.

Build those foundations first and get more from whatever you buy later.

Where Sapiens Fits

The underwriting model this article describes depends on two things: automation for straightforward risks and documented human oversight of every automated decision. Sapiens P&C insurance software supports the policy lifecycle across personal and commercial lines, from acquisition through to renewal. SapiensAIP, the Sapiens Autonomous Insurance Platform, adds insurance-specific AI agents on top of Sapiens Core to automate underwriting workflows, with underwriters reviewing and approving the recommendations. Audit trails, explainability, configurable guardrails, and human-in-the-loop oversight sit behind every agent, so you can show how a decision was reached.

Book a demo