We all know AI tech is moving fast, and so is adoption broadly across industries. McKinsey’s 2025 global survey found that 78% of respondents were using AI in at least one business function.
If you drill down into the healthcare industry specifically, however, you’ll see that AI adoption is not happening equally across use cases. The Engagys 2025 State of Engagement Survey of healthcare payer organizations found that 60% of respondents had not yet started using AI to support member communications and engagement.
Unfortunately, the consensus is that health plans overwhelmingly want to leverage AI for content creation, but legal and compliance issues are hindering growth.
Watch the interview and read the full article below for practical guidance on speeding internal approvals and maintaining long-term compliance.
Establish an AI governance committee inclusive of compliance, clinical, legal, IT, and data personnel, and define policies and procedures that align with legal and compliance requirements for not only the short-term, but also for long-term sustainability.
- Organized communication and processes between teams required for approvals will mitigate the time spent playing telephone between teams within your organization.
- Continuous regulatory monitoring and updating of AI policies will ensure AI-enabled systems are always compliant
- Reviewing the health of AI implementation on a regular cadence can catch issues like bias or model drift (degradation of a model’s performance over time due to changes in real-world data distributions)
- Organization-wide agreed-upon situations and thresholds for escalation are essential for keeping your AI implementation in check
Architect your AI solutions with transparency, flexibility, and human oversight as core requirements instead of an afterthought. When something breaks, or an audit comes along, you don’t want to be retroactively trying to piece together evidence and non-existent data.
- Require all AI vendors to sign a BAA prohibiting the use of member data for training and sharing with third parties, and are compliant with all HIPAA security and privacy requirements.
- Scrub and de-identify your data before sending it to any AI tools, or create synthetic versions of your member records that represent the same data.
- Evaluate AI vendors for transparency. Many LLMs like ChatGPT are considered “black box” models, meaning that their reasoning and decisions are not fully transparent to users. There are methods to expose some of this transparency called eXplainable Artificial Intelligence (XAI) techniques that you can build around your model, but it’s not as effective as full API transparency.
- Logging and storing all AI inputs, outputs, and everything in between is crucial for performing high-quality internal and external audits of your systems, but also supports iterative improvement. You can send analytics on those logs to your AI governance committee to spot weaknesses and use those same logs for A/B testing.
Consider the advantages and feasibility of self-hosting, or on-prem AI. Given the lack of transparency and how token/credit pricing models are fluctuating, some executives at major corporations are finding that implementing AI is actually costing more than humans.
- On-prem models have a huge advantage in data privacy and security. While you may not be able to afford one of NVIDIA’s supercomputers like Mayo Clinic, you can still get some real advantages in cost by hosting an open-source model via cloud services as opposed to inking a deal with an AI vendor directly.
- If you are running your own open-source model, consider a hybrid of RAG for supplementing up-to-date knowledge and fine-tuning to adjust the weights of a model to fit your company and specific tasks.
All of this sounds great on paper, but of course, the devil is really in the details, and that’s where Engagys can help. Wherever you’re at in your AI journey, you’ve almost certainly come across legal and compliance challenges, or one of the following:
- Lack of AI strategy with clear ROI is leading to indecision.
- Your fragmented data, or gaps, are inhibiting you from implementing AI capabilities and/or receiving meaningful output.
- AI systems are opaque, and no AI governance team is in place, leading to auditing risks
- Even with a fully fleshed-out AI operating model and supporting infrastructure, the output you’re receiving is inaccurate and/or unusable.
- Analysis paralysis when it comes to choosing a vendor.
At Engagys, we have successfully guided clients through these AI obstacles, yielding outcomes like 50% savings in labor costs, 15%-20% increase in health understandability and accessibility, and 5% increase in setting and keeping appointments.
We do this by delivering innovations that are practical, measurable, and transformative.
