Aetna has rebuilt its cost-estimation system to cover more than 20,000 medical services and wrapped it in plain-language AI, and it has put an AI assistant in front of more than 15,000 care-management nurses that summarizes a case before the call and structures the record after it. Nathan Frank, Aetna's chief digital officer, puts the saving from the second one at 90 minutes per nurse per day. Both run on an enterprise data platform CVS Health built during the pandemic, which is the part that explains why this is arriving now rather than three years ago.
Scale matters for reading the rest. Aetna is CVS Health's health insurance division. CVS Health booked $402.1 billion of revenue in 2025; the Health Care Benefits segment that contains Aetna accounted for $143.4 billion and served 26.6 million medical members at year-end. Frank runs roughly 6,000 technology professionals supporting about 2,000 platforms, with responsibility for member, clinician and employee services, platform modernization and AI.
His framing of the problem is that health care makes the customer do the homework. Consumers agree to a service without knowing its price, carry their own information between organizations, and coordinate their own treatment — conditions that would be unacceptable in most other industries. The system was built around insurers, physicians and health organizations rather than the patient, and Frank's stated goal is to use technology to join coverage, pharmacy services and care into one coherent path.
When he took over Aetna's technology organization about four years ago, executives at other insurers were mostly spending on utilization management, cost control and the move to cloud. Frank bet on member experience instead. The cost-estimation rebuild is the clearest product of that choice: a system that prices both simple and complex procedures in advance, in ordinary language and with AI support, across more than 20,000 services.
The behavioral shift is the headline claim. Frank says fewer than 10% of people used the old version; among those who estimate cost before a procedure, the figure has risen to nearly 70%.
Aetna does not judge these products on standard customer-experience metrics alone. Teams run comparative tests, then match digital actions and proactive messages against cost of care, prior-authorization automation and medical outcomes — checking whether a feature helped someone get preventive care, avoid an emergency department visit, or head off a more complex procedure. Prior-authorization automation showing up in that list is telling: it is the same lever the peer insurers were pulling four years ago, reached from the consumer side rather than the cost side.
None of it predates the plumbing. During the pandemic CVS Health built an enterprise data platform that pulls information from across the company and makes it available along the patient journey, and invested in a clinical data repository to consolidate clinical information and share it with physicians and others in the system. Frank says today's AI, machine learning, prediction and proactive messaging would not have been possible without data organized properly first, and that the foundation now supports scenarios meant to improve human interaction rather than remove staff from the process.
The nurse workflow is the most concrete application. Aetna's care-management team runs to more than 15,000 nurses. Before speaking with someone newly diagnosed with cancer or another serious illness, a nurse previously had to open several systems and read hundreds of pages of medical records. Now AI summarizes the case notes in advance, ambient speech recognition runs during the conversation, and the information is structured afterwards so the nurse can move to action. Frank puts the return at 90 minutes per nurse per day and says the intent is that nurses spend more time with patients, not that there are fewer of them.
On the outbound side, next-best-action features can prompt a member to book a vaccination or a preventive check-up. Frank says Aetna was the first insurer to deploy RCS for the purpose — the standard that replaces ordinary SMS with interactive messages, letting a member read up or find a doctor inside the messaging app they already use. Engagement on RCS messages has run above 80%, and opt-outs are less than half the SMS rate. Aetna then tracks whether those contacts lead to preventive care and, eventually, to fewer complex interventions. The approach combines Aetna's data, its relationship with members and a familiar digital channel to make messages more useful rather than simply more numerous.
The constraint Frank names on all of it is trust. CVS Health employs more than 3,000 data specialists and has worked in machine learning, analytics and responsible AI for years. Every AI use case passes through a governance framework weighted toward privacy, security and transparency, and Frank stresses that AI should reinforce Aetna's relationship with members and be used in communication with them correctly and comprehensibly.
Two things here deserve separating from the pitch. The first is the arithmetic on that 70%. Fewer than 10% is a share of people; nearly 70% is a share of people who estimate cost before a procedure — a group defined by having already done the thing being counted. The two figures do not rest on the same base, and nothing in the account says what share of 26.6 million members touch the estimator at all. The improvement may well be real. The comparison as presented cannot demonstrate it.
The second is what the nurse product actually is. A nurse opening several systems and reading hundreds of pages before a phone call is not a language problem; it is a symptom of 2,000 platforms. On the most direct reading, Aetna's summarizer is an expensive adapter layered over integration debt that the same organization is separately trying to pay down through platform modernization. That does not make it a bad investment — 90 minutes a day across more than 15,000 nurses is genuine capacity, whatever its cause. But it is worth being clear which problem is being solved, because the summarizer only keeps returning that time as long as someone keeps paying for it, while a consolidated record would stop generating the cost in the first place.
And for an organization that says it measures features against emergency department visits and avoided procedures, the outcome numbers are exactly what is missing. There is a usage figure, an engagement figure and a time-saved figure. There is no figure for preventive visits generated, complex procedures avoided, or cost of care moved — the tests Frank himself describes as the real ones.
The operating model is meant to hold the work to that standard. Engineers, technology product leaders and business product leaders build solutions together, with nurses, care coordinators and service staff involved in development of the features they will use, and feedback embedded in the products so teams can see how services are used and tune models hour by hour and day by day.
Frank expects AI and real-time data exchange to reshape member experience, clinical care and provider operations while lowering administrative cost, and says the most significant shift will be to a system that predicts and prevents chronic and acute conditions rather than responding after they appear. The company's job, in his words, is to help people live better and spend less time inside the health care system.
That is also the part of the program carrying no number. Aetna can show what it has taken off nurses' calendars and what it has put onto members' screens; prevention is the claim it has not yet had to prove — and it is the one that would justify an insurer spending four years on price transparency in the first place.