Health plans sit on three things AI was made for: massive volumes of repetitive work, decades of proprietary data, and workflows where small improvements compound into real margin.
But health plans also sit on three things that make AI hard: regulated decisions, fragile member trust, and a matrix of stakeholders, clinical, compliance, IT, network, sales, service, who each have a legitimate claim on how AI gets used.
This page is a practical view of where AI creates the most value in a health plan, where it creates the most risk, and how leaders should think about the difference.
Managed care is unusually well-suited to AI.
Most industries have to manufacture the conditions for AI to work. Health plans already have them.
- Volume. Millions of claims, calls, authorizations, and member touches every year.
- Repetition. The same kinds of decisions show up over and over.
- Proprietary data. Years of claims, eligibility, network, and member behavior data that no competitor can replicate.
- Verifiable outcomes. Most decisions have a clear right answer.
- Regulatory guardrails. The rules are written down. They can be taught to a model.
The same conditions that make managed care operationally heavy are what make it AI-ready. The question is not whether AI applies. The question is where it applies first.
Where AI should help first in a health plan.
Not every function is the same kind of opportunity. Three categories, in the order most plans should pursue them.
High-volume, rule-bound work.
These are the workflows where AI delivers the cleanest, fastest, most defensible ROI. Low risk, high repetition, clear right answers.
- Provider directory verification and accuracy
- First-tier member services inquiries
- Claims triage and routing
- Prior authorization intake and document classification
- Appeals and grievances intake
- Broker support and quoting assistance
- Routine eligibility and benefits questions
The pattern: AI handles the volume; humans handle the exceptions. The financial case usually clears the bar on its own.
Judgment and personalization.
These are workflows where AI doesn't replace the human, it gives the human a better starting point or gives the member a clearer experience.
- Plan selection and enrollment guidance, especially for Medicare and ACA
- Care management risk stratification and outreach prioritization
- Stars and quality program targeting
- Member onboarding and year-round engagement
- Personalized care navigation
- Sales and broker copilots
The pattern: AI compresses a complex decision into something a member or a frontline employee can actually use. This is where long-term competitive advantage starts to show up.
Analytics at the speed of decisions.
These are the workflows where AI changes how leaders see the business, not how they execute it.
- Market and competitive intelligence synthesis
- Network adequacy and provider performance analysis
- Risk adjustment opportunity identification, with clinical validation
- Medical cost trend explanation and root cause analysis
- Product design analytics across markets
The pattern: AI doesn't make the decision. It cuts the time between question and informed answer from weeks to hours.
Where AI should not lead.
Some decisions in a health plan should stay with humans, even when AI could technically inform them. The leaders who get this right name those lines clearly, and early.
- Coverage and medical necessity decisions. AI can prepare the case. A clinician decides.
- Appeals at the member level. AI can route, summarize, and surface precedent. A human owns the outcome.
- Sensitive member conversations. End-of-life, serious diagnoses, financial hardship. These are not workflows. They are relationships.
- Anything where the wrong answer could harm a member.
This is not a limitation of AI. It is the design of the operating model. The plans that draw these lines clearly are also the plans that move fastest on everything else, because the rest of the organization understands what AI is for.
The Six Lenses applied to a health plan.
Each lens shows up differently in managed care. Here's how to think about each one.
Start with two cases at once, a near-term efficiency case the CFO can fund this year, and a longer-term member-experience case that builds competitive position. One pays for the other.
Map work at the task level, claims processor, member services rep, care manager, network analyst, broker support. Most roles contain tasks in all three categories: automate, augment, preserve. The role doesn't disappear. It gets redesigned.
Tier your models. Productivity tools, internal analytics, member-facing recommendations, and coverage-influencing decisions are not the same risk class, and should not have the same approval process. Centralize the risk framework, model inventory, and approval gates. Decentralize use case selection inside each function.
The crown jewel is not the model. It's the claims history, the member behavior data, the provider network data, and the years of operational data that sit underneath. Most of that data is what the industry calls "dark", it exists but isn't AI-ready. The work of preparing it is the highest-leverage work a health plan can do.
The roles most plans are underbuilt on are not data scientists. They are AI business translators, AI data stewards, responsible AI leads, and change champions inside each function. Hire one of each before hiring a tenth model engineer.
The biggest barrier is rarely the technology. It is medical directors who don't trust the model, clinicians worried about liability, member services reps worried about their job, and brokers worried about being replaced. The plans that handle adoption well bring the frontline into the design of the AI tools they will eventually use.
How most plans should approach the work.
The names will differ inside each organization, but the sequence holds.
- Build the inventory. What AI is already running, what data it uses, who owns it, and what decisions it influences. Most plans are surprised by what they find.
- Pick three workflows. One efficiency play, one member-facing play, one analytics play. Set a real baseline.
- Stand up the four roles. Translator, data steward, responsible AI lead, change champion network.
- Tier the models. Decide which approval path applies to which kind of use case. Publish it.
- Pick the moat. Identify the proprietary data and workflow design where a long-term advantage can be built.
- Measure end-to-end. Time from member inquiry to resolution. Not logins. Not seats. Not pilot counts.
This is twelve months of work, not three. The plans that try to compress it usually end up running pilots that never scale.
Most AI initiatives don't fail at the model.
They fail in the months after deployment, when the workflow hasn't been redesigned, the frontline hasn't been brought along, the measurement is still task-level, and the C-suite has moved on to the next priority.
There is a productivity dip before there is a productivity gain. The leaders who name that dip, and protect the team through it, are the ones who actually capture the value.
AI doesn't fix a broken process. It scales it. The work of fixing the process is still the work.
What separates the plans that will lead.
They are not the ones with the biggest models. They are the ones that:
- Choose where AI helps first based on value, not enthusiasm.
- Govern centrally what protects the enterprise and decentralize what creates value.
- Treat data as the moat and design accordingly.
- Build the right four roles before they build the tenth tool.
- Take human adoption as seriously as model accuracy.
- Measure what actually matters, productivity, not activity.