P&C Automation: The Role of AI and Agentic AI in Insurance Automation Software

 P&C Automation 

Automation and AI in insurance are no longer emerging concepts, they are operational realities. In 2017, fewer than 50% of carriers reported using AI or machine learning in underwriting. By 2021, that figure had risen to 94%, according to benchmarking data from Luma Financial Technologies. The release of generative AI tools like ChatGPT accelerated this further, making advanced AI capabilities accessible across business functions almost overnight. 

The industry is now transitioning from non-agentic AI, bounded, task-specific tools, toward agentic AI, which executes entire end-to-end processes to deliver a business outcome. That shift carries significant implications for how MGAs are structured, how underwriting workflows are designed, and what operational readiness actually requires. 

MGAs are well-positioned to move faster than traditional carriers. The delegated authority market now represents roughly 40% to 60% of gross written premium in intermediated channels, depending on the source. In the US and other major markets, "greenfield" MGAs are launching with digital-first architectures and no legacy constraints, and the ability to build modular, API-driven ecosystems from day one. 

At the same time, consolidation continues across the MGA landscape. Mid-sized MGAs are being absorbed into larger platforms, while new niche MGAs are emerging with focused underwriting strategies and technology-enabled operating models. The common denominator is agility. Yet alongside optimism, concerns around data privacy, model hallucinations, non-determinism, and AI governance are rising. As automation and AI in insurance expand, governance frameworks must evolve in parallel. 

Underwriting Challenges: Why AI Is Being Considered 

The insurance industry is not yet in a golden age of underwriting, and P&C automation offers the clearest path toward one. According to a study by Boston Consulting Group, specialty underwriters spend approximately 40% of their working day on administrative work and data entry. Broker submissions routinely run to 230 or 300 pages, driven in part by brokers sending everything they have to manage their own E&O exposure. Underwriters are then expected to read, structure, and re-key that information across multiple systems before making a pricing or coverage decision. The burden is not a shortage of data, it is the cost of ingesting and organizing it. 

The commercial impact is measurable. Research cited by Ernst & Young suggests carriers are approximately 60% more likely to write a risk if they respond to a broker first. Faster response times correlate directly with revenue outcomes. This cultural dimension remains as important as technology itself. 

Where Automation and AI in Insurance Can Actually Help MGAs 

AI adoption within MGAs is strongest in bounded, practical use cases, particularly in submission intake, claims FNOL, and actuarial support. Fully automated specialty underwriting remains some distance away. 

1. Submission Intake 

Submission ingestion is the clearest early win for P&C automation. An incoming request for a quote can pass through a workflow, call an intelligent document processing API powered by an LLM, and return structured data into downstream systems, with a human reviewing flagged fields rather than the entire document. 

LLMs perform well on structured extraction tasks, achieving 85% to 95% accuracy on clean documents such as ACORD forms and standardized loss runs. However, insurance tolerance for error on pricing, limits, retentions, and compliance-sensitive fields is near zero. Two risks are worth noting: 

• Hallucination on sparse documents: when fields are missing, models can generate plausible-sounding values rather than flagging the absence 
• Degradation of messy documents: faxed PDFs, handwritten annotations, and mixed-language riders are common in real MGA submissions and significantly reduce model reliability 

For these reasons, carriers and MGAs consistently prefer a human-in-the-loop model at intake. Importantly, data accuracy captured at the point of entry, down to the asset level, has value even for risks ultimately declined. It informs portfolio management and maintains continuity of insight across the entire policy lifecycle. 

2. Claims

Claims automation has proven more structurally complex than submission workflows. At Insillion, a comprehensive motor claims implementation requiring 15 modules and 15 user profiles revealed a critical limitation: claims workflows are deeply embedded organizational habits, not standardized processes. The practical near-term value of AI in claims lies in ingestion and analysis:  

1. Ingesting policy forms and treaties 
2. Identifying endorsements and duplicates
3. Communication with customers and agents

 For most MGAs where TPAs manage claims handling, the priority is accurate first notification of loss (FNOL) data capture rather than full downstream automation. 

3. Actuarial and Renewal Insights

Actuaries are using AI to query large datasets, extract specific claims from thousands of records by keyword, and receive rapid coding support. The larger opportunity is at renewal: AI can surface changes in exposure, loss experience, or external risk signals almost instantly, shifting conversations between junior and senior underwriters away from data gathering and toward judgment and strategy. 

4. Why Human-in-the-Loop Is Non-Negotiable 

AI systems are non-deterministic. Outputs can vary, and hallucinations remain possible. In a regulated industry, every step of the decision process requires documentation and accountability. Best practice for high-stakes applications could involve having one model check another's output or running parallel models and using disagreement as a signal for human review. 

The aim is not to eliminate human judgment but to ensure humans focus their time on decisions that truly require it. The objective is augmented intelligence: AI handles structured extraction and pattern recognition, humans verify results, and apply commercial judgment. While AI has a clear role in standardized, high-volume lines, complex specialty underwriting still requires human judgment, business acumen, and relationship management.  

5. Future Use Cases: What the Data Signals 

Automation is more feasible in high-volume, repeatable segments. However, the industry is close to significantly improving human performance through AI and digital tools. The immediate opportunity involves augmenting underwriter capabilities and reducing administrative burdens.  

As AI continues to evolve, more complex and higher-impact use cases are likely to emerge across insurance operations. Over time, AI has the potential to reduce process uncertainty and strengthen confidence in pricing and underwriting decisions, possibly improving profitability and enabling carriers to consider risks that were previously avoided. 

Today, most carriers are applying generative AI internally to drive operational efficiency and productivity. Customer-facing use cases remain limited and cautious, typically focused on straightforward interactions such as helping a policyholder retrieve a policy number. 

As AI matures, more advanced use cases may become viable: 

1. Real-time premium calculations with automated data gap filling 
2. Deep research summaries condensing 100-page reports into actionable insights 
3. AI-assisted risk scoring embedded directly into underwriting workflows 

More deterministic underwriting decision support may develop over the next 4–8 quarters. Even then, human validation will remain central to ensure transparency and regulatory compliance. A recent executive study by the Geneva Association indicates AI deployment today is concentrated in: 

1. Data and analytics (57%) 
2. Claims (46%) 
3. Software engineering (44%) 
4. Customer support (41%) 
5. Underwriting (31%) 

This measured progression reflects the industry's need to balance innovation with governance. Over the next three years, underwriting and new business are flagged as high-priority areas for increased AI investment, precisely where MGAs stand to benefit most. With 48% of customer respondents already reporting challenges in claims navigation, the near-term opportunity for MGAs lies in deploying AI across the full lifecycle.   

Concerns Around AI: Data, Consent, and Governance 

As automation and AI in insurance scale, data governance becomes critical. Frank Senter in an Webinar organized by Insurtech Association states that Policyholders may not fully understand how their data flows through agency management systems, quoting platforms, SaaS vendors, and analytics engines. 

According to an insurance customer survey: 

1. 37% cite privacy and security concerns about how their personal information is used 
2. 35% question the correctness and accuracy of AI-generated responses 
3. 30% worry about over-reliance on automation, making it harder for humans to intervene 

These reflect a structural problem the industry has not yet resolved. Data entered into agency management systems and quoting platforms is frequently accessed by SaaS vendors and third parties for purposes unknown to the insured and, in some cases, monetized or used to train AI models without explicit permission. The concern is not automation itself but that the same data flows powering smaller-scale aggregation businesses can now operate at a significantly greater scale with AI, without adequate governance infrastructure. 

Regulators can audit licensed entities, but tracking data usage further up the supply chain remains complex. The standard being advocated is affirmative opt-in consent, not terms buried in a document nobody reads, for every use of policyholder data, whether by a quoting engine, a carrier, a data aggregator, or an AI training pipeline. This does not prevent agentic AI from functioning. It simply requires explicit customer agreement before automation begins. 

What MGAs Should Know About Agentic AI 

Agentic AI is still in the experimental and ideation stages across the industry. For MGAs, a clear understanding of AI governance is essential, including acceptable data use in their specific domain and how their AI usage aligns with developing regulations. To fully use Agentic AI, operational data access is critical. Generative AI offers opportunities to reimagine customer experience and drive industry differentiation, but much of the data needed for these solutions often resides in older systems designed for human access, not for agents. 

Before moving toward agentic implementations, MGAs should have two things in place. 

• First, clear AI governance: an understanding of acceptable data use in their specific domain and alignment with evolving regulations.  
• Second, programmatic data access: agentic AI requires machine-readable access to the same data humans currently access through a user interface.  

Legacy systems built for human navigation, such as COBOL and AS400, cannot simply have an agentic layer placed on top of them. Without API exposure and structured data architecture, agentic layers cannot function effectively. Major industry experts suggest starting with non-agentic, use-case-specific AI first. Understand the fundamentals of authentication, authorization, and data architecture. Then build toward process-level automation as both the technology and internal readiness mature. Agentic AI has the potential to increase productivity for agents and brokers, enabling them to manage more relationships by automating administrative chains.  

However, adoption should proceed incrementally, beginning with bounded use cases. 

Insillion CEO's Take on AI 

Insillion CEO Raja Raman, in an interview with TMPAA, is clear: AI adoption is not optional over the long term, but immediate full automation is premature. Today's AI remains non-deterministic, making human oversight essential. The focus should be on strengthening infrastructure now to adopt more mature AI capabilities over the next 6 to 12 months. 

The shift away from monolithic platforms toward modular, API-driven ecosystems is central to this. Modern APIs and low-code tools allow MGAs to connect underwriting workbenches with existing policy systems without costly overhauls. Insillion addresses this gap directly, offering carrier-grade PAS capabilities restructured for MGAs through a flexible, pay-as-you-grow model.  

In P&C automation, AI already supports structured data extraction and report summarization through integrations with LLM models. Workflows are increasingly configured as modular "flows" rather than rigid processes. 

Key takeaways for MGAs: 

P&C automation is not about replacing underwriters. It is about reallocating time from manual ingestion to commercial judgment. 

To future-proof, MGAs must  

1. Make technology central to strategy 
2. Build where you differentiate; buy where processes are standardized 
3. Prioritize modular, API-first systems over monolithic platforms 
4. Combine AI adoption with operational discipline and risk management 

As automation and AI in insurance mature, competitive advantage will belong to organizations that pair technological capability with strong change management and data governance. The near-term opportunity lies in augmentation, improving speed, accuracy, and consistency, while building the foundation for more advanced AI capabilities in the years ahead.