CVS Health

CEO David Joyner feels the core problem with AI transformation is fragmentation. Data sits across multiple businesses and records, and the patients are expected to hold it all together.

Enterprises usually solve fragmentation by pointing AI at every workflow and building governance over it. CVS did the opposite. Before scaling anything, Tilak Mandadi, the company’s Chief Experience and Technology Officer, drew a line the entire AI program is built on top of.

CVS has committed to three specific things AI will never do inside the company.

Every use case has to fit inside that boundary, or it doesn’t ship.

A promise like this only means something if it survives contact with 88 million pharmacy claims and 37 million insurance members.

It’s the same instinct we saw at Goldman Sachs. They didn’t transform the whole firm with AI, but picked where to invest.

Now that we know what CVS refuses to automate, you can see how the company’s AI applications stem from the “AI can’t decide” principle.

The most important metric in insurance claims is speed. The provider waits for it and doesn’t want any delays in an emergency.

So what does CVS do if it doesn’t want AI to approve or deny claims, but also doesn't want to compromise on speed?

In May 2026, Aetna launched the second generation of its Claims Assist Manager (CAM), an agentic AI platform for claims that need manual review. Inside CAM, AI agents pull together eligibility, coverage, member, and provider data, then recommend a next step. A human still closes the claim.

Because of CAM, the processing time for complex claims dropped more than 20%.

The CAQH Index, a benchmark that tracks administrative spending across U.S. healthcare, puts the annual cost of claims-related paperwork at roughly $80 billion, with more than $20 billion of that recoverable through better automation.

CAM is Aetna’s attempt to claim its share of that number.

If claims are the hardest test, pharmacy is where AI pays off first. It’s also one of the most literal-minded jobs in healthcare.

A prescriber writes dense, jargon-heavy instructions, and someone has to translate them into an exact dose, frequency, and quantity without a single error.

CVS trained open-source LLMs to do that translation.

The models parse conditional dosing logic and clinical terminology that older, rules-based systems couldn’t handle, then calculate the precise dosage supply from a prescription’s frequency, quantity, and duration fields.

Pharmacists still sign off on every output. AI removes the manual transcription.

CVS’s primary care arm, Oak Street Health, follows the same pattern. Ambient AI scribes now run at 90% of facilities. A physician talks with a patient.

The AI writes the clinical note. Automate the paperwork, keep the human on the decision, repeat across every business unit.

CVS is running the same playbook on its own workforce that it runs on members.

For members, Aetna’s conversational AI assistant, launched in fall 2025, now fully resolves roughly 75% of interactions without a human agent. In May 2026, CVS expanded its partnership with Salesforce to bring Agentforce Health to both Aetna and Caremark call centres, unifying member data that had previously resided in separate systems.

For employees, CVS built ARTEY, an internal AI assistant on the Moveworks platform, and rolled it out to roughly 380,000 colleagues. Live agent chats dropped 50% within 30 days of launch. ARTEY has since logged more than 2.5 million conversations, and 75% of employees who use it come back to use it again.

More tools and use cases within the company:

In March 2026, CVS partnered with Google Cloud to launch Health100, a new subsidiary built on Gemini models, Google’s Cloud Healthcare API, and BigQuery. The company wanted to build an AI-native health platform open to any consumer, regardless of which pharmacy, insurer, or doctor they use.

Google Cloud CEO Thomas Kurian described the goal as “agentic, AI-powered health care that enhances human touch and eliminates complexity.”

CVS could have kept its AI infrastructure as a private advantage, walled off for its own 185 million consumers. Instead, it is positioning Health100 as engagement-as-a-service, something CVS can eventually sell to employers, payers, and other health systems that aren’t its customers today.

It's the same move McKinsey made with its solution, QuantumBlack. Prove the tool works on yourself first, then package it for everyone else.

Health100 draws on CVS’s multiyear, multibillion-dollar technology commitment, announced in mid-2025. Initial launch is targeted for later in 2026.

Decide what AI won’t do before deciding what it will. In a regulated industry, a specific, enforced no-list builds more trust.

Operational wins can outpace consumer-facing ones. CVS’s claims, pharmacy, and workforce AI all shipped and proved out before Health100 existed. The consumer platform sits on plumbing that already works.

Human-in-the-loop is a speed strategy, not only a safety one. By keeping a person on every decision, CVS can move fast in claims and pharmacy at the same time, without triggering the regulatory scrutiny that slows down less disciplined competitors.

Sources