Harrison.ai’s pivot from selling its radiology AI tools to hospitals in a transactional SaaS revenue model to contracting out clinicians using their tools to make their money reeks, just a little, of desperation.
The ABC ran a prominent story on one of our two defining Australian medical AI start ups this week – Harrison.ai.
The ABC pointed out that, despite accepting $32 million in its last funding round from the country’s National Reconstruction Fund Corporation – apparently on the grounds that it would keep its main business and investments headquartered in Australia – Harrison.ai made a few employees redundant earlier in the year and started a new business in the US in which it was investing heavily in recruiting local clinicians.
The story questioned the ethics of the group accepting $32 million to invest in AI locally, then presumably shifting some of that investment into the US pretty quickly.
It also revisited the fact that Harrison.ai had trained a lot of its tools on de-identified patient data from one of its major local investors – I-MED – without asking patients for permission, something that the Office of the Australian Information Commissioner reviewed and cleared because the data was sufficiently de-identified to meet privacy rules.
But the ABC may have missed the main point of what seems to be going on.
Harrison.ai was founded in Sydney in 2018 by brothers Aengus and Dimitry Tran with a pitch that was technically credible and commercially attractive – build AI tools that help radiologists read CT scans and chest X-rays faster and more accurately, sell them to hospitals, collect recurring fees, scale globally.
Why is it now pivoting into a highly established and competitive sector in the US on a completely different revenue model to what it started with?
The technology is real and apparently clinically rigorous.
Like Heidi, our other great medical AI start-up white hope, Harrison.ai was seeding the technology throughout hospitals around the world at little or no cost with the intention of getting clinicians to love it enough to presumably get the hospitals to start specifying it and paying for it.
Like Heidi, Harrison.ai has some serious backers – Blackbird Ventures (same as Heidi), Horizons Ventures, Sonic Healthcare, Ramsay Health Care, Aware Super, and of course, the National Reconstruction Fund Corporation – and by 2025, it had raised $271 million and was valued at close to $400 million.
But unlike Heidi – whose founder Thomas Kelly said in April that annualised revenues were as much as $70 million now, and whose resulting valuation is currently above the $600 million mark – the revenues of Harrison.ai have remained pretty much a mystery since they were last reported for the financial year ending 30 June 2024 at only $4.75 million.
But the comparison between the groups is now pretty stark.
Heidi isn’t quite six years old, has an ARR of around $70 million (reported by the CEO), up from roughly $1 million just two years earlier – a run rate that has famously now given the group the moniker of fastest growing local start-up ever (faster than Canva). Heidi has done all that on about half the capital Harrison.ai has, and it’s processing more than two million consultations per week across 190 countries – which suggests considerable further upside.
Heidi isn’t out of the woods here. It has significant strategic headwinds in trying to crack into the hospital enterprise market against global healthcare EMR gorillas Epic and Oracle, and it has reportedly struggled to find backers in the US for its next funding round, for which it is seeking about $100 million.
Neither company has fully cracked the monetisation question. Heidi’s revenue is not publicly filed, and the same structural question applies: its valuation is built on potential rather than current earnings.
But Heidi’s engagement data is undeniable, its geographic spread is real, and its growth trajectory is upward. It’s doing okay.
Harrison.ai has now been eight years operating, has raised $271 million and its only reliable reported revenue, albeit in June 2024, is $4.75 million. If it had more revenue to spruik publicly you’d think it would have, like Heidi has.
Related
Instead it is pivoting its revenue model in what could only be described as a fairly desperate way: it is starting a teleradiology contracting business in what is an established and highly competitive market, against major long-term players with far more capital backing, and it is gambling that, by having its new business use its own AI tools, it will create a business that is up to 30% more efficient than its established competitors.
What could possibly go wrong with this plan?
The problem is that it’s an admission that it’s not meaningfully converting any of its existing customers of its tools to paying for them. And that problem is apparently so stark it’s starting a business it has no experience in – contract teleradiology – with its only advantage being that its clinicians are using its AI toolkit.
Frontier Radiology is not a software product. It is a teleradiology company that hires radiologists, uses Harrison.ai’s AI to make them more productive, and sells the reporting service to US hospitals. Radiologists are offered a 25% individual bonus tied to AI-attributed productivity.
As of now, Frontier Radiology has no publicly disclosed hospital clients. Its website reads as a recruitment pitch to radiologists, not a client case study page.
The strategic logic isn’t that easy to follow.
If hospitals won’t pay enough for AI diagnostic tools to generate meaningful SaaS revenue, become the radiologist. Capture the economic value of the AI productivity gain directly.
Is this a rational response to the revenue problem? And if the product is so good why can’t Harrison.ai sell it to the companies it’s now proposing to compete against?
The US teleradiology market is worth approximately $1.4 billion and growing, driven by a genuine radiologist shortage.
But it is crowded and mature. vRad, the market leader, operates with more than 500 subspecialty radiologists and 11% global market share. Radiology Partners is the largest US radiology practice and building its own AI triage platform.
Dozens of established players are well ahead of Frontier Radiology on client relationships, regulatory approvals, and hospital integration.
And all of them are developing their own AI tools.
There is also a conflict that has received almost no attention so far.
I-MED Radiology – one of Harrison.ai’s original investors and the provider that supplied it with tens of millions of patient scans to train its AI – acquired StatRad, the US’s second-largest teleradiology provider, in July 2024.
So I-MED, one of Harrison.ai’s main initial investors, is now technically a direct competitor to Frontier Radiology in the US market.
Harrison.ai’s founding data partner is competing against the clinical services business Harrison.ai is building to generate the revenue its software business has not.
Heidi and Harrison share the same fundamental challenge: the technology works, but getting paid adequately for it in a health system designed around other incentives and with established enterprise players with their own long-term customer moats isn’t easy.
Heidi’s risk is that key EMR incumbents – Epic, Oracle, Best Practice, Genie for example – build ambient documentation natively into the platforms that already own the clinical workflow, displacing a standalone product however beloved it is early on.
Harrison.ai has a much bigger problem now. It appears to have run out of time for the software model to generate revenue at scale, so it has pivoted and is entering a mature US services market as a complete newcomer.
Of the two risks, Heidi’s is far more survivable. A product with genuine clinician loyalty and 86% utilisation rates has something to defend.
A software model generating $4.75 million, or whatever it actually is now, after eight years does not.
For the NRFC (and Australian taxpayers), which invested $32 million to keep Harrison.ai in Australia, and which the National Audit Office found had gaps in its investment due diligence, the question of whether the pivot succeeds is now the only question that matters.
Frontier Radiology has no clients yet.
The clock, and the $32 million investment, is running.