Artificial intelligence (AI) holds tremendous promise for insurance underwriting, claims, and operations. Properly applied, AI opens teams’ capacity, boosts accuracy, speeds decision-making, and ultimately improves customer outcomes.
Realizing this potential requires more than adding to your tech stack. Building and maintaining effective AI solutions is a complex, multidisciplinary process, and the challenges are frequently underestimated.
But before encountering these obstacles, it’s important to recognize that your journey begins not with the technology itself but with selecting the proper use case.
Select Your Use Cases
Selecting AI use cases should be a strategic, ROI-focused decision. Here are some best practices to keep in mind:
- Align with business goals to support core objectives like improving customer experience, streamlining operations, enhancing risk management, or ensuring compliance.
- Target high-impact, attainable wins across repeatable processes where AI can drive immediate benefits—like claims indexing, submission intake, or Certificate of Insurance (COI) generation.
- Ensure that your chosen use case complies with data privacy laws, explainability requirements, and internal governance policies.
- Pilot first with a low-risk project and use real-world feedback to refine the solution before scaling.
- Establish clear KPIs – e.g., cycle time reductions, cost savings, etc. – and use the results to guide broader AI deployment.
A well-selected use case can lay the groundwork for long-term success.
Let’s dive into the five key challenges of building and maintaining AI solutions.
Challenge #1 – Data quality and privacy
AI is only as good as the data it’s trained on – and data is only as good as the data scientists that manage and apply it. Getting good quality data is expensive. You need to recruit, and hire data scientists and provide them the tools necessary to overcoming common data challenges, like:
- Data Quality – Inconsistent formatting, incomplete fields, or outdated information can seriously degrade AI performance. For example, an underwriting AI trained on unreliable loss run data may misclassify risk, leading to pricing errors or missed underwriting guidelines.
- Data Privacy – Insurers handle vast amounts of sensitive personal and financial data. Any AI system must be built with strong safeguards to ensure privacy and compliance with GDPR, HIPAA, CCPA, and other regulations. Failure here risks not just stiff penalties, but also loss of customer trust.
- Data interoperability – Many insurance systems are siloed, complicating access to clean, structured data. Getting your core systems to work together is often a project in itself.
Challenge #2 – Bias mitigation
AI systems don’t have intent, but they do have influence, which can reflect:
- Algorithmic bias – AI can repeat or reinforce patterns from past decisions. This can seriously affect underwriting decisions, claims triage, or fraud detection.
- Transparency and explainability – Advanced AI models can lack decisioning transparency (the so-called “black box effect”). Regulatory compliance and ethical standards require insurers to be able to explain and audit AI decisions.
- Ethical considerations – Fairness can affect customer satisfaction and brand integrity. Insurers can build trust by embedding bias monitoring to detect and correct unfair outcomes.
Avoiding model drift and bias are full-time concerns needing dedicated expertise in fairness-aware machine learning and explainable AI methodologies. Investments in specialized data scientists and/or comprehensive bias detection solutions can effectively address these challenges, however at significant cost.
Challenge #3 – Regulatory compliance
Insurance is consistently among the most closely regulated industries. Organizational agility and effective AI governance support contending with:
- Evolving regulations – What’s compliant today may not be tomorrow. AI systems must be adaptable and auditable to meet shifting requirements.
- Accountability and oversight – Insurers need robust governance frameworks in place to satisfy regulatory requirements around AI accountability concerning automated decisions (e.g., claim denials, pricing recommendations).
- Regulatory complexity – Rules vary across regions, states, and business lines. AI solutions must be configurable and context-aware, not hardcoded to a single standard.
These compliance challenges require specialized technology expertise, and tools, such as comprehensive model governance platforms, automated audit trail systems, and real-time compliance monitoring tools, which can add complexity and cost to building AI solutions in-house.
Challenge #4 – Continuous improvement and maintenance
Like any business-critical system, AI models need ongoing support and integration to remain effective over time. Common challenges include:
- Model drift – As conditions change, models trained on past data may lose accuracy. Data scientists performing ongoing monitoring and retraining are critical to keeping AI systems effective and relevant – and this adds significantly to the cost of building and maintaining AI systems.
- System integration – AI must connect to policy administration systems, claims platforms, CRMs, etc. Seamless integration often requires APIs, custom workflows, and resources for maintaining legacy infrastructure.
- User training and change management – Teams need training not just on how to use the tool but also on when to rely on it and when to override it. Without proper onboarding, even the best AI solution may be underused or misapplied.
- Monitoring and feedback – Human-in-the-loop (HITL) feedback systems are critical to capturing and applying data from workflow exceptions for continuous learning. These essential systems require developers to build HITL tools and subject matter experts to perform HITL checks and correct faulty AI decisions in real time to build accuracy, trust, and continual learning into the system.
Challenge #5 – Other considerations
Beyond data, bias, regulation, and maintenance, below are a few additional hurdles every insurance leader should be aware of:
- Implementation costs – Building and scaling AI systems for maximum ROI – especially for core processes – requires substantial outlays for human experts, tech infrastructure, data preparation, training, and integration depend on planning and budget.
- Customer acceptance – Consumers regularly express skepticism – or worse – about AI-based underwriting and claims decisions. Transparency, speed, and the ability to escalate to a human are critical components of winning people over to any AI-powered experience.
- Cybersecurity risks – AI systems expand your digital footprint – and potential attack surface. From data poisoning to adversarial attacks, insurers must deploy strong cybersecurity measures to protect models and the data they rely on.
- Overreliance on AI – A successful AI rollout can create its own risk: confirmation bias. AI is proven to deliver greater ROI by augmenting human expertise, especially in high-stakes decisions.
- Competitive pressures – The most sustainable AI strategies balance speed with thoughtful execution. Succumbing to “FOMO” can backfire by rushing adoption without proper governance, alignment, and training.
Build vs. Buy: Why Insurance-Specific AI Matters
Faced with these challenges, insurers must decide whether to build and maintain in-house solutions or buy from a specialized partner. While building offers potential long-term differentiation, it’s a path fraught with cost, risk, and complexity.
Off-the-shelf LLMs (e.g., from Meta, OpenAI, Anthropic, etc.) lack specific understanding of insurance nuances – particularly around regulatory compliance. Also, selecting a public model is far from a turnkey solution. These tools necessitate deep internal AI expertise for continued development and maintenance.
Partnering with an insurance-specific AI vendor can accelerate time-to-value at a lower risk. These solutions come pre-trained on industry data, and have embedded capabilities for regulatory compliance, continuous monitoring and performance improvement, to solve real-world underwriting, claims, and policy servicing use cases.
AI is not just another tool – it’s a transformation.
However, this transformation comes with real challenges. Insurance carriers, MGAs, and TPAs must take a deliberate, responsible approach to building and maintaining AI systems that are accurate, fair, compliant, and adaptable. The good news? You don’t have to solve it all at once. Start with a single, strategic use case. Build trust, measure success, and scale from there. With the right foundation, AI can help your organization improve outcomes for customers, underwriters, adjusters, and everyone in between.
Curious how insurers are putting Roots to work? Check out our case studies to see insurance-specific AI in action.