Olive AI’s Mirage of ROI

how the “AI co-worker” raced to a $4B valuation before hospitals realized ROI didn’t add up

Olive AI’s shutdown highlights how lofty promises in healthcare automation can collapse without sustainable ROI.

For a while Olive felt like the future. The company called its product an AI co-worker and spoke to hospital leaders in a language that was refreshingly human. Let this digital teammate log in, click through screens, move fields between systems, and carry the drudgery that burns out revenue cycle staff. The purple booths at conferences were packed. Investors agreed. In July 2021 Olive announced $400M of new capital led by Vista, lifting its valuation to four billion dollars. Since the start of the pandemic the total raised approached the nine hundred million mark. The pitch became a chorus. Olive was in hundreds of hospitals, in forty states, inside more than twenty of the top one hundred health systems. It sounded inevitable.

Inside hospitals the experience was far less glossy. 

What looks like one problem on a slide is actually many small problems that do not stay still in the wild. Every health system runs a slightly different stack with custom fields, idiosyncratic workflows, and payers who revise rules with little notice. Automating clicks across that terrain behaves less like a durable product and more like a script farm. It works today and fails tomorrow when a screen moves or a code set changes. In the short window of a pilot, minutes saved look promising. In production, every integration break reintroduces manual work and erodes trust. That gap between the stage and the shift floor set the trap.

Olive’s answer to the complexity was speed. More capital. More logos. More surface area. In December 2020 the company bought Verata Health to go after prior authorization, a workflow that truly hurts both providers and patients. The business case seemed obvious. Fix prior authorization and you improve access to care while protecting revenue. The acquisition also multiplied the places where things could go wrong. Prior authorization is not a single flow. It is a cluster of edge cases that vary by payer, plan, specialty, attachment type, medical policy, and CPT nuance. Each exception needs its own guardrail. Each guardrail increases the cost of maintaining reliability across sites. The company was now promising value across a broader front while the engineering problem grew harder with each go-live.

The narrative kept running ahead of the measurement. Olive described itself as a digital employee. Hospitals heard the word employee and expected the reliability they demand from people. If an EHR upgrade moved a field, the work still needed to get done that day. If a payer updated a rule, the process had to adapt without a week of reconfiguration. That is where the metaphor began to hurt more than it helped.

A script that breaks when a pixel shifts does not feel like a colleague. It feels like technical debt you have to babysit. The higher the promise, the lower the patience for breakage.

By mid 2022 the private whispers had turned public. Reporters in Columbus were hearing about over-promising and under-delivering. Then came the layoffs. 450 people were cut, roughly a third of the company. The CEO wrote to staff that leadership had made missteps, that the company had grown too fast and lost focus. The timing mattered. Capital markets were tightening and buyers were tightening with them. In a kind market, a loose proof story can survive. In a hard market, the proof has to clear a CFO’s bar. 

The gap between the deck and the ledger was now a risk no one wanted to carry.

Twelve months later the end arrived quickly. Olive announced that it would sell the heart of the business and wind down. The clearinghouse and patient access units went to Waystar, and the prior authorization line went to Humata Health. Everything else would be shut. The shutdown was striking not because companies never fail but because of how fast the arc flipped. In 2021 Olive stood for the promise that automation could free hospitals from administrative toil. In late 2023 the same company was a set of carve outs with the lights going dark.

So what actually went wrong, beneath the headlines and the brand voice. The first hypothesis is about proof. Olive treated provisional wins like durable ROI. A pilot that saves minutes is not yet a line that moves days in accounts receivable. True impact in revenue cycle shows up in fewer touches per claim, lower denial rates by category, faster cash for specific payer and service lines, and it survives staff churn and software updates. That kind of proof needs instrumentation that both the vendor and the CFO trust. Many buyers could not reconcile Olive’s aggregate claims about hundreds of millions in efficiencies with site level reports that would stand a finance review. 

In the absence of shared measurement, enthusiasm turns into skepticism. Skepticism stalls expansion. Expansion pressure pushes the company to chase new logos instead of hardening reliability at existing sites. The growth engine begins to feed on future promises because the present is too noisy to showcase.

The second hypothesis is about metaphors. Calling a product a co-worker is clever storytelling, and it helped the company stand out in a crowded automation market. It also sets a reliability bar that basic RPA cannot clear without a serious investment in change management and governance. If the promise is that work just gets done, then the product needs visible guardrails, documented failure modes, and a clear rollback path that does not require heroics from the hospital team. Marketing humanized the idea of automation. The operations did not consistently match that promise when screens changed or payers shifted policy. 

Over time, every breach of expectation subtracts trust faster than any new feature can add it back.

The third hypothesis is about timing. Olive tried to win breadth before it had depth. Buying into prior authorization made strategic sense on paper because the spend and the pain are real. The catch is that every new domain multiplies edge cases and raises the reliability bar. Scale is not the enemy if the core loop is boringly repeatable. Scale becomes the enemy when the core loop is still fragile. 

The company expanded where it was most tempting to expand, not where it was safest to prove repeatability beyond doubt.

There is also a governance story here that should matter to anyone building AI for complex enterprises. If you cannot show exactly what is automated, where errors will surface, how a human can take over, and how savings are reconciled to P&L at the end of the quarter, then your customer is not a blocker. Your customer is telling you how to unlock the deal. Hospitals wanted to believe. They also needed evidence that would survive a budget cycle and an audit. Without that evidence, even a friendly champion cannot keep the project safe when finance begins to ask hard questions.

If this story helped you see how a billion-dollar promise can collapse when proof of value is missing, share it with a friend, a colleague, or even that one person in your WhatsApp group, on X, or on Instagram who never stops dissecting business stories.

They might just look at “ROI” very differently after this.