How UnitedHealth Group Uses AI

UnitedHealth Group, known to most people as UnitedHealthcare, serves 49.8 million UnitedHealthcare members and more than 123 million people across its Optum businesses. In 2025, the company processed claims worth $447.6 billion in revenue, while Optum Rx managed $188 billion in pharmaceutical spending.

By the end of 2025, UnitedHealth Group (UHG) turned that history into more than 1,000 AI use cases running in production, double of what it had a year earlier. UHG has set aside close to $1.5 billion for AI in 2026.

So where does this advantage come from? And what does UHG do with it? Let’s break it down.

UHG runs two businesses:

Split across four segments, the numbers look like this:

In 2025, UHG pulled in $447.6 billion in revenue, up nearly 12% from the year before. UnitedHealthcare grew 16%, and Optum grew 7%. Both businesses have more than 390,000 employees, out of which (approximately)165,000 are in clinical roles.

While most companies only get to see one side of a transaction, UHG gets to be on several: Through UnitedHealthcare, it's the insurer. With Optum Health, it provides the care, and Optum Rx is the pharmacy.

Optum Insight manages the technology running underneath these businesses. So, when a claim gets stuck, UHG doesn’t need to wait for input from another company to resolve the issue. It owns each point in the chain, including the one where the holdup happens.

That's also why it can sell the fix. Optum Insight built Optum Real to speed up claims adjudication and coverage validation inside UHG's own business. Now it sells that same tool to outside hospitals, health plans, and providers who face similar challenges. 

How do you let 390,000 employees use AI without every team connecting sensitive health data to a random tool on the internet? 

When you are processing claims worth $447.6 billion in revenue, a stuck claim runs the risk of slowing down millions of transactions a day. And a prescription that takes hours to approve holds up every patient behind it.

So, how do you ensure everything runs smoothly?

UHG answered that with United AI Studio, an internal platform that gives employees access to large language models from major AI providers. The platform also allows them to use smaller models UHG built on its own data. This ensures that their employees work inside the same set of controls regardless of the model they pick.

The platform also gives UHG a way to bring its own data into those AI workflows. UHG handles multiple sides of the healthcare system, giving it access to years of records across the care journey. Every claim, prescription, patient visit and follow-up call has flowed through its systems. And because the company owns both sides of nearly every transaction in a patient's care, that data has taken the shape of history.

UHG then paired the platform with training. More than 10,000 employees signed up for an advanced AI course launched in 2025. So, while the platform gave employees access, the course taught them how to use it well.

Once the platform was in place, use cases spread across the business.

By early 2026, more than 80% of consumer contacts happened digitally, and digital visits hit 73 million in a single quarter, up 42% over two years.

Their Rules-based software already auto-approves about 90% of claims. Of the ones that need manual review, 98% get approved. 

UHG is now testing AI on the hardest cases, mostly to fill in missing or incorrect information before a claim ever reaches a human. 

Close to 95% of requests come in electronically. From these, about half get processed in real time, while more than 90% get approved within a single business day.

Optum Rx built PreCheck Prior Authorization, a tool that cut prescription approval time from over 8 hours to under 30 seconds. 

According to Conway, CEO of Optum, after implementing PreCheck, they saw: 

UHG’s Optum Insight also built Optum Real, a tool aimed at claims adjudication and coverage validation. 

Conway says it cuts manual contact costs by 76% for the payers and providers using it.

Every business unit reviews its own AI use cases every month. Chief information officers across the company meet every quarter to review adoption enterprise-wide. And a responsible AI board reviews hundreds of use cases a month before any of them reach production.

Because of this review process, UHG has been slow to hand things over to agents. Only about six use cases run on agentic AI today, and most of them handle repetitive admin work, not clinical decisions. 

The sequence UHG followed remains intrinsic regardless of the business you run. 

Identify your moat: 

Look at your internal structure. What do you already own that a competitor can't easily copy? It can be a dataset, a customer relationship, a process you run at a scale nobody else does. 

Build the access layer before you build the use cases:

Instead of building multiple pilots for different departments simultaneously, start with one platform that lets people use AI safely. Let the use cases come in gradually. 

Match your review speed to your platform speed:

A responsible AI board reviewing hundreds of use cases a month sounds slow. But it's also the reason UHG could scale past 1,000 use cases at all. With the review built in, it might slow down AI adoption, but catching problems early ensures sustainability.