AI Underwriting: How to Use AI for Underwriting (2026)

Artificial intelligence is reshaping the underwriting process from a manual, time-intensive art into a data-driven science. For insurance professionals in underwriting, claims, and risk analytics, understanding how to use these tools is no longer optional. It’s the core of the modern role. The goal isn’t to replace human judgment but to augment it, absorbing repetitive administrative tasks and allowing underwriters to focus on complex risks and strategic portfolio management.

This guide provides a practical, step-by-step walkthrough of how to use AI in a real-world underwriting workflow as of September 2026. We cover the distinct stages of the process, review the specialist platforms built for each job, and detail the compliance landscape that every underwriter must navigate. ZEKAI reviews all tools independently; our recommendations are based on publicly available data and our analysis of each platform’s specific function within the underwriting value chain.

What AI Underwriting *Actually* Is in 2026

AI underwriting is the use of technologies like machine learning and natural language processing to automate and improve risk assessment. Unlike older rules-based systems that simply follow a static “if-then” script, AI models learn from historical data to identify patterns, predict outcomes, and process huge volumes of unstructured information (like broker submission emails or property inspection reports) without manual data entry.

A McKinsey analysis found that 30% to 40% of an underwriter’s time is spent on manual administrative tasks like re-keying data. AI-powered underwriting absorbs this work, freeing up human experts for higher-value analysis.

The practical application of AI in underwriting breaks down into five key stages:

1. Submission Intake & Triage: AI automatically extracts data from submission documents, digitizes it, and triages cases based on appetite and completeness.
2. Data Enrichment: The system pulls in third-party data—such as property records, telematics data, or litigation history—to build a comprehensive risk profile.
3. Risk Assessment & Scoring: Machine learning models analyze thousands of variables to score the risk, identify potential hazards, and flag anomalies for human review. Traditional underwriters might assess 15-20 variables; an AI system can analyze over 1,500.
4. Pricing & Quoting: AI can recommend pricing and policy terms based on the risk score and historical performance data, improving consistency.
5. Decision & Documentation: The system generates a complete, auditable record of the data and logic used to arrive at a decision, which is critical for compliance. A human underwriter reviews and makes the final bind or decline decision.

This workflow doesn’t remove the underwriter; it equips them with better tools to make faster, more consistent, and more data-backed decisions.

The Best AI Underwriting Platforms in 2026: Ranked & Reviewed

Choosing the right AI underwriting tool depends entirely on the problem you need to solve. The market is not a monolith; it’s a stack of specialized platforms for different jobs. Some tools focus on data intake, others on telematics, and a third group on end-to-end risk decisioning.

We evaluated these tools based on five core criteria relevant to a working professional as of September 2026:

• Underwriting-Specific Functionality: How well does it perform core underwriting tasks (risk assessment, data enrichment, etc.)?
• Integration & Data Sources: How easily does it connect with existing systems (LOS, policy admin) and what third-party data can it access?
• Compliance & Explainability: Does it provide clear audit trails and help meet regulatory requirements like the NAIC AI Model Bulletin?
• Pricing Transparency: Is pricing clear and based on value (e.g., per-submission) or is it opaque enterprise-level contracting?
• Role Suitability: Is it built for a specific line of business (P&C, Life, Cyber) or a specific user (carrier, MGA, broker)?

Swipe the table sideways →

Our Tool Verdicts

How to Use AI in the Underwriting Workflow: A Step-by-Step Guide

Integrating AI is not about flipping a single switch. It involves applying specific tools at specific stages of the underwriting process.

Step 1: Automate Submission Intake and Triage

The process begins when a submission arrives, often as a collection of PDFs and emails. Instead of an underwriter manually re-keying data, an AI platform with Natural Language Processing (NLP) and document intelligence capabilities reads the submission.

Analyze the attached broker submission for a commercial property policy. Extract the following fields: Applicant Name, Address, Requested Coverage Limits (Building, Contents, Business Interruption), and Prior Carrier Loss Runs for the past 3 years. Flag any missing information. Based on our underwriting appetite guidelines (document: UW_Appetite_v3.pdf), classify this submission as: (1) In-Appetite, (2) Out-of-Appetite, or (3) Refer to Senior Underwriter. Provide a confidence score for the classification.

Tools like BriteCore’s submission intake copilot claim to reduce manual data entry by 80-90%. This step transforms chaotic, unstructured data into a clean, digitized format ready for analysis.

Step 2: Enrich Data for a 360-Degree Risk View

Once the initial data is captured, the AI system enriches the profile by pulling data from external sources. For a homeowners policy, this could be Hippo’s use of aerial imagery to assess roof condition or public records for past permits. For a commercial auto policy, it could involve telematics data from a platform like Root. This creates a far more comprehensive risk profile than what is available from the application alone.

Step 3: Generate an AI-Powered Risk Score

With a complete and enriched dataset, machine learning models analyze the information to produce a risk score. These models can identify complex, non-intuitive relationships between variables that a human might miss. For example, an AI might find that a certain combination of business type, geographic location, and litigation history is highly predictive of future liability claims.

This is not a “black box.” Modern AI underwriting systems are designed for explainability, showing the key factors that contributed to the score. This is essential for both internal governance and regulatory compliance.

Step 4: Make the Decision with Human Oversight

The AI presents its findings to the human underwriter: the risk score, the key contributing factors, any flagged anomalies, and a recommended decision (approve, decline, or refer) with proposed pricing. The underwriter’s job is to review this package, apply their experience and judgment—especially on complex or edge cases—and make the final call.

No credible system in 2026 fully automates the final underwriting decision for all but the most simple, straight-through policies. The human is always the final authority.

The Compliance Landscape: AI Rules You Must Follow

Using AI in underwriting is not a regulatory free-for-all. A strict legal framework governs how these tools can be used to ensure fairness and transparency.

The NAIC AI Model Bulletin

The National Association of Insurance Commissioners (NAIC) has been proactive in setting expectations. Their Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in late 2023, has become a key piece of regulation. As of early 2026, over half of U.S. states have adopted the bulletin or similar guidance.

The bulletin doesn’t ban AI. Instead, it clarifies that existing insurance laws—like those against unfair trade practices—apply to decisions made or assisted by AI. It requires insurers to maintain a written AI governance program, document their risk management controls, and be accountable for any third-party AI vendors they use. Regulators can and will ask for these documents during market conduct examinations.

Colorado SB 21-169: The Bias-Testing Mandate

Colorado’s Senate Bill 21-169 is a landmark piece of legislation that prohibits insurers from using external data, algorithms, or predictive models in a way that results in unfair discrimination based on a protected class (race, color, sex, etc.). The law requires insurers to establish a governance and risk management framework and demonstrate to regulators that their AI systems are being tested for and are free of discriminatory bias. While it initially focused on life insurance, the principles are expected to expand to other lines.

As of March 2026, approximately 24 states had adopted the NAIC’s AI Model Bulletin, making AI governance a mainstream regulatory expectation.

FCRA and Adverse Action Notices

When an AI model contributes to an “adverse action”—such as declining an application, offering less favorable terms, or charging a higher premium—the federal Fair Credit Reporting Act (FCRA) and Equal Credit Opportunity Act (ECOA) have strict notification requirements.

An insurer must be able to provide the applicant with the specific, principal reasons for the adverse action. Simply stating “the algorithm decided” is illegal. The CFPB has been clear that using a “black box” model that cannot be explained is not a valid excuse for failing to provide specific reasons. This makes model explainability not just a good practice, but a legal necessity. Your AI system must be able to surface the key factors that drove its recommendation.

Common Mistakes and How to Avoid Them

Implementing AI in underwriting can create massive efficiencies, but pitfalls are common.

1. Treating AI as a “Black Box”: The biggest mistake is deploying a model you can’t explain. If you can’t tell a customer or a regulator *why* they were declined, you are exposed to significant legal and reputational risk. Solution: Prioritize platforms with built-in explainability and audit trail features.
2. Ignoring Data Privacy: Feeding sensitive policyholder information (PII) into unsecured, generic AI models (like the public version of ChatGPT) is a massive data breach waiting to happen. Solution: Use enterprise-grade, secure AI platforms designed for insurance that have clear data handling and privacy protocols.
3. Automating Judgment Instead of Tasks: The goal of AI is to automate administrative tasks, not to replace the nuanced judgment of an experienced underwriter. Models are good at spotting patterns in data; they are bad at understanding novel risks or complex context. Solution: Structure your workflow as AI-augmentation, where the system handles data prep and scoring, but a human makes the final, authoritative decision.
4. Forgetting Third-Party Vendor Risk: If you use an AI tool from a vendor, you are still responsible for its outputs. The NAIC Model Bulletin explicitly states that insurers are accountable for the third-party systems they use. Solution: Your vendor contracts must include audit rights and guarantees of cooperation with regulatory inquiries.

Ultimately, successful AI adoption in underwriting is a strategic change, not just a technical one. It requires redesigning workflows and upskilling your team to transition from manual data processors to data-driven risk managers. For professionals in the AI, insurance claims, and risk analytics field, mastering these tools and principles is the key to leading the industry forward.