Risk adjustment with AI: Key do’s and don’ts

Artificial intelligence (AI) has rapidly transformed risk adjustment, unlocking new levels of efficiency, coding accuracy, and clinical insight. But in a highly regulated environment like healthcare, success isn’t just about adopting AI, but about adopting it responsibly.

Separating hype from reality requires a clear understanding of what works, what doesn’t, and how to apply AI in a way that drives measurable outcomes while maintaining compliance, transparency, and trust. Here is a practical guide to the essential dos and don’ts for AI in risk adjustment, grounded in real-world application and industry expertise.

The do's: Building a responsible, high-impact AI strategy

Establish data quality procedures

High-performing AI models depend on high-quality data. This is particularly challenging in healthcare, where information is complex, highly variable, and often unstructured. AI helps bring greater consistency to this data, but model performance ultimately relies on how well the data is curated, contextualized, and aligned with clinically validated terminology, as well as how effectively variation in documentation is addressed. Bringing together structured and unstructured data, along with clinically meaningful relationships between concepts, helps reduce ambiguity and supports more consistent, trustworthy outputs.

Regularly refine models

AI models require ongoing evaluation to maintain performance and ensure compliance over time. Organizations should continuously monitor key metrics while regularly testing outputs for bias or inconsistencies. Models should be further updated as new data becomes available and as CMS and OIG regulations evolve. Ultimately, effective AI is not just about the capabilities of a particular model, but how it is governed, controlled, and integrated into workflows. This includes maintaining control over the models, protecting data privacy, and enabling the customization needed to meet regulatory requirements.

Utilize transparent and explainable models

Transparency is essential to building trust in AI outcomes. Using explainable AI enables clinicians, coders, and compliance teams to understand how outputs are generated and to validate them against source documentation. By providing clear evidence lineage and surfacing contextual support from across data sources, AI solutions can strengthen confidence, improve audit readiness, and support defensible programs.

Establish security and compliance in the foundation

AI in risk adjustment must be built on a secure and compliant infrastructure. Organizations should ensure that sensitive data remains protected within controlled environments. Strong governance processes, including cross-functional oversight from compliance and legal teams, are critical to ensuring alignment with regulatory requirements. Adhering to HIPAA safeguards and maintaining robust access controls helps protect patient privacy and reinforces trust.

Design AI to deliver actionable outcomes

AI creates the most value when it moves beyond automation to directly improve program performance. Effective integration into workflows, supported by high-quality, secure data, enables organizations to turn insights into measurable impact, including increased productivity, reduced manual effort, and improved data accuracy. By delivering more precise predictions of patient risk and surfacing potentially actionable insights for review, AI can help inform more timely, proactive interventions. This in turn may support  earlier identification of care or documentation gaps, stronger program results, and better member outcomes, ultimately helping plans control costs, protect revenue, and improve overall performance.

The don’ts: Avoiding common pitfalls

Don’t forget to mitigate bias

Failing to address bias in AI models can lead to inaccurate or inequitable risk scores. Where applicable, models should be validated across diverse populations and monitored continuously to ensure fair and consistent performance. Strong data quality, clinical grounding, and ongoing evaluation are essential to mitigating bias and fostering compliance.

Don’t undermine data security

Protecting sensitive health information is non-negotiable. Organizations should never input protected health information into public or unsecured AI tools. Instead, they should be HIPAA-compliant and rely on secure, compliant environments with strict access controls and appropriate data handling practices, such as de-identification where applicable. Maintaining strong data governance safeguards both stakeholders  trust and organizational integrity.

Don’t forget the experts

Expert oversight is critical to supporting the use of AI in risk adjustment in a manner designed to promote accuracy, compliance, clinically meaningful outcomes. This includes setting governance guardrails, embedding AI into real-world workflows, applying clinical review where appropriate to nuanced cases, adapting to evolving CMS requirements, and using insights to support provider education. This ensures AI drives trusted, sustainable results.

Don’t overpromise

While AI offers significant benefits, it is not a fully autonomous solution. Organizations should avoid overstating its capabilities and instead focus on realistic, measurable outcomes such as improved efficiency, increased accuracy, and enhanced visibility. Clear communication about both strengths and limitations helps set appropriate expectations and builds credibility.

Don’t bypass stakeholder alignment

Successful AI adoption requires alignment across clinical, compliance, coding, and operational teams. Without stakeholder engagement and seamless workflow integration, even the most advanced AI solutions may fail to deliver meaningful value. Organizations should ensure that AI tools are configurable and adaptable to their unique needs and processes.

Driving smarter, more accountable AI in risk adjustment

AI is a powerful tool to enhance expertise, improve accuracy, and drive measurable outcomes when applied thoughtfully. Organizations that take a disciplined approach, grounded in high-quality data, strong governance, transparent models, and human oversight, are best positioned to unlock its full value. By focusing on responsible adoption and real-world impact rather than hype, health plans can use AI to strengthen compliance, improve efficiency, and ultimately deliver better outcomes for the populations they serve.