The great metamorphosis:
From insurance provider to health partner

insight
Oct 21, 2026
10 min read

The conversation that makes you a health company

Every big health insurer now says the same thing - we're becoming a health company. UnitedHealth and CVS say it explicitly; Bupa already operates as one, running hospitals and clinics alongside insurance; Ping An is the only one that rebuilt itself around the principle.

And it’s not merely a branding exercise; it's the only growth story left. In the US, sector EBITDA fell from 11.2% to 8.9% of national health spend between 2019 and 2024 (McKinsey), and when most of every premium dollar goes straight back out as claims, you can't price or cut your way to growth. You must sell the member something they use all year, not a card they think about twice.

It all depends on one thing: being part of the conversation that matters.

Someone notices a symptom at 11 pm. Before AI, they would have typed it into Google, read three contradictory pages and either sat on it or waited for a Practitioner’s appointment. Now they open ChatGPT, along with around 230 million other people worldwide every week, to ask health-related queries (OpenAI, Jan 2026). Moreover, only 38% of insured US members open their health plan's app at all, and in Medicare Advantage, where the insurer carries full risk for the member's health, it's 20% and falling (J.D. Power, 2026).

That first exchange sits at the front of everything that follows: “Is this serious?” comes before “should I go?”, which comes before “who do I see?”. And the insurer appears at no point in that sequence. Last in line, Insurance providers are left with the task of processing the claim (which, as we know, doesn’t make for the best contact surface).

Insurers were never a part of the conversation to begin with — Think about the last time you contacted your own insurer: it was about a claim, a bill, whether something was covered and not "I found a lump, and I'm scared." Members have always reached insurers for the paperwork of health rather than the experience of it, and the health conversation was always happening somewhere else. What's changed is that the “somewhere else” is now an AI that's very good at it.

The health episode has three owners

The insurer holds only the final slice, the one stage the member never experiences as a relationship.

Figure 1

Figure 1: Who the member is talking to, at each stage of a health episode.


The diagram shows the whole episode and a gap running through it. Nothing holds the conversation at the front, carries what the plan covers into the middle, and routes into the care that follows. That joining role is the orchestration layer, and no incumbent owns it at scale.

That isn't an oversight; it's a trust problem. Why would a member describe a symptom to the company that prices their risk? No insurer has answered that convincingly, so none of them have built it, and everyone has walked away from the conversation instead.

 

2. Insurers can't grow inside the premium. The way out is a conversation they're not having.

Regulators have already decided how much of a premium an insurer keeps. In the US, the Affordable Care Act is explicit: at least 80% must go to care, 85% in large-group and Medicare Advantage, leaving 15 to 20 cents to cover staff, technology, marketing, and profit. What survives as profit is a single-digit net margin of a nickel for years, falling to just 0.4% in 2025, the industry's lowest point of the decade (NAIC).

Europe reaches the same ceiling through regulated competition rather than a mandate, with 90% to 94% of premium going on care. In the Gulf, the cap is the product itself: cover is compulsory, and the regulator writes the benefits, with Saudi Arabia binding every licensed insurer to a single policy, so what one insurer sells is close to indistinguishable from the next. 3 rulebooks, one outcome - no room to grow inside the premium.

You cannot grow inside the premium

The law caps where the dollar goes; what’s left as profit is a nickel that some years turns negative.

Figure 2

Figure 2: How much of every premium dollar the law allows an insurer to keep.


Which leaves one direction, and it sits outside the premium: care delivery, virtual visits, chronic-disease management, pharmacy, wellness. These are the services around the insurance rather than the insurance itself, and none of them count against the medical-spend cap because none of them are the insurance product. This is why becoming a health company stopped being a slogan and turned into a necessity. But nobody enrols in a diabetes programme because it exists; they enroll because someone put it in front of them at the moment it mattered.

Take the relationship away and ask what separates one insurer from another. Price and network, both of which a competitor can match, and even the old data advantage is going now that a member can hand their entire medical history to ChatGPT in a few taps. The relationship is the only asset left that cannot be copied or undercut. It is also the one thing almost none of the current AI spending is pointed at.


3. Where the spending goes & where it doesn’t

Insurers have already built the assistant. UnitedHealth had guided to $1.5 billion for 2026; by mid-2026 that had doubled to a $3 billion, two-year commitment; Elevance, formerly Anthem, has one reaching 22 million members; Humana runs a Gemini-built assistant behind more than 20,000 care advocates. Ask what any of them do and the answer is the same: explain what's covered, find someone on network, chase a claim. Good tools for the conversation insurers were already having. Only Oscar has pointed one at symptoms instead of coverage, and it now handles 86% of member questions (Becker's, 2026).


Only one player can own both sides

Everyone else holds half. The winnable corner is reachable from exactly one direction.

Figure 3

Figure 3: Who owns the conversation, who owns the risk, and who owns both.

The reason the rest haven't is visible in how the spending gets justified. UnitedHealth reports its AI returns entirely as savings - about a billion dollars of the cost taken out, with most projects paying back within a year or two (STAT, 2026). Money not spent, never money newly earned. "Cut contact-center cost 20% by Q3" fits in a board deck; "be there when a member first wonders if something is wrong" does not. And while insurers optimized the paperwork, a $14 billion global care-navigation market grew up around them - a meaningful share of it already selling guidance straight to the employers who buy US insurance (The Business Research Company, 2026).

Ping An didn't just build the assistant; trust was already there, and it's why members bought more from the company beyond the core policy - the same growth that only exists outside the premium (Ping An, 1H 2026). Treat it as directional, since the market and the moment differ, but it's the clearest proof anyone has that the join works, which is what the rest of this piece is about to define.

4. The conversation worth building for

So, what does owning it look like, and how does a conversation turn into growth?

Let's start with what it isn't. It isn't a slicker chatbot that's competing on the one thing that's becoming free. And it isn't buying hospitals the way Ping An had to; Ping An is the proof, not the template. Until now, being there when a health episode began meant owning the care that followed, because owning the care was the only way to have the relationship at all. That's the part AI removes: an insurer can orchestrate the care rather than own it, sitting between the member and a health system it partners with.

A chatbot answers a question. An orchestration layer acts on it, understands the symptom, weighs urgency alongside what the member's plan covers, routes to in-network care, books it, and remembers for next time. A general assistant can't do that, because it has no coverage authority. A single provider can't either, because it can't see across the system. Only the party already holding the claims, the data and the network can stand in that spot and it's the only party not trying to. Transcarent, which absorbed Accolade in a 2025 acquisition, is selling care navigation to employers today.

They are building the layer without the coverage, while insurers keep building the coverage without the layer.

Health Insurance IX - Health_v2-1

 

 

Which brings back the member's questions. How do I know you're not listening to reprice me? Reassurance is worth nothing here; the answer must be architecture. The layer sits walled off from underwriting and rating: what a member says at 11 pm never reaches pricing, and the separation is auditable rather than promised. In Europe, the case for that wall is only getting stronger. Software that interprets symptoms is already regulated as a medical device, and AI used to price health risk is classified as high-risk under the EU AI Act - a classification the industry now has until December 2027 to fully comply with, after a 2026 deferral, but one that isn't going away. Building the separation now is the difference between adapting on your own timeline and scrambling to one set by Brussels.

Get that right and the layer starts to earn, though never on its own. Owning the relationship is the precondition for everything sold around the insurance: virtual care, second opinions, chronic-care and wellness programs, pharmacy. None of it counts against the premium cap, because none of it is the insurance product. The relationship is what turns a worried member at 11 pm into a routed visit, a filled prescription, an enrolled program.

And the layer itself is sellable. Two-thirds of covered US workers are in self-funded plans, where the employer carries the risk, and the insurer is paid a flat fee to administer it - claims, network, member services. There's no premium to grow around because there's no premium, so the layer is the only thing on the table that isn't price and network. In the Gulf, the insurer still carries the risk, but the buyer is the same: an employer choosing between near-identical mandated plans.

Health Insurance IX - Control_v2-1

Retention runs on the same logic. When a member’s whole health life moves through your layer, there’s far less reason to leave. And every exchange deepens the health view that makes the next one sharper.

There's a further layer to how that conversation gets read. Voice carries what words alone don't. The tone, pace, the strain in someone's breath, and a layer built to read that can tell a routine question from a genuine emergency - not from what was said but from how it was said.

None of what this layer does i.e reading urgency, checking coverage, booking a slot, escalating to a clinician is one system doing everything. It's a set of specialist agents, each built for a single job, called on by whatever the conversation needs. The layer's task isn't to know everything itself; it's to know which agent to call, and when.

Everything this needs already sits inside an insurer: the claims, the network, the data, and now the models. What's missing isn't capability. It's the decision to point all of it at the other conversation not the one about bills, the one at 11pm.

Health Insurance IX - Empathy_v2-1

 

 

5. The question for the next planning cycle

The Intelligence Loop

Every completed episode makes the next one sharper — and a sharper layer is what growth is built on.

Figure 4 (1)

Figure 4: What the layer does, and where growth comes from once it's running.


The layer earns nothing on its own. Growth depends on the loop running not on any single episode, and not reachable from the claim alone.

Every avenue in that loop needs the loop running first. No relationship, no growth. The claim alone never gets you there.

 

How many of your members asked an assistant about their health last week instead of asking you?

Almost no one can answer that yet. Answer it, and the growth avenues aren't hypothetical anymore. They're just what comes next.

meet the authors

Professional Corporate Circle Headshot
Hemanth Bajaj
Associate Principal Consultant
Nagarro
bruno_round
Bruno Martins
Product Design Lead
Nagarro

Sources

Section 1
Section 2
  • McKinsey, What to expect in US healthcare — sector EBITDA declined from 11.2% to 8.9% of national health spend.
  • National Association of Insurance Commissioners, 2024 Annual Health Industry Commentary — health industry margin data
  • Affordable Care Act — Medical Loss Ratio requirements.
Section 3
  • Becker's Payer Issues, UnitedHealth is spending $1.5B on AI this year — UnitedHealth's original 2026 AI investment guidance
  • Modern Healthcare — UnitedHealth's $3B AI investment across 2026–27.
  • Wayne DeVeydt, CFO, UnitedHealth Group, Barclays Global Healthcare Conference, March 2026 — approximately $1B in cost savings and reported 2:1 returns from AI investments.
Section 4
  • Transcarent, Transcarent completes merger with Accolade — merger announcement
  • Council of the European Union, 29 June 2026 — EU Digital Omnibus on AI and revised compliance timelines.
  • European Commission, MDCG 2019-11: Guidance on Qualification and Classification of Software — MDR Rule 11 guidance