4 min read
If you've ever sat by your phone waiting for a specialist's office to call you back, and waited, and waited, you're not alone. It turns out the reason has very little to do with your doctor not caring, and everything to do with a broken administrative system that is quietly drowning healthcare workers in paperwork. As someone who spends a lot of time thinking about where AI intersects with real human problems, the author finds this topic genuinely fascinating, and a little urgent.
A recent TechCrunch piece on Basata, an AI startup targeting the healthcare back office, puts a sharp lens on a problem most of us experience but rarely understand. The article captures something important: administrative staff in specialist clinics aren't slacking. They're overwhelmed. And that overwhelm is the silent reason your referral slips through the cracks.
To understand why this matters from an AI and systems perspective, it helps to map what's actually happening. When a GP refers you to a specialist, a surprisingly complex chain of administrative tasks kicks off. the author has looked at this kind of workflow bottleneck in other industries, logistics, legal, finance, and the pattern is almost always the same: highly skilled humans are being used as glorified data routers.
In healthcare, that typically means administrative staff are manually:
Each of these steps is individually manageable. Together, across dozens or hundreds of patients per day, they create a cognitive and operational avalanche. The TechCrunch article quotes Basata's founders noting that the administrative staff they work with aren't worried about AI taking their jobs, they're worried about drowning. That framing resonated with the author immediately, because it reflects something often lost in the AI displacement debate: the baseline experience for many workers right now is already unsustainable.
From a machine learning standpoint, the tasks Basata appears to be targeting are genuinely well-suited to current AI capabilities. Natural language processing has matured to the point where extracting structured data from unstructured clinical documents, referral letters, discharge summaries, GP notes, is increasingly reliable. Workflow automation, intelligent routing, and anomaly detection (flagging missing information before it causes delays) are all tractable problems for modern ML systems.
the author would argue that the more interesting engineering challenge here isn't the AI itself, it's the integration layer. Healthcare systems are notoriously fragmented. In the UK alone, NHS trusts often run on legacy infrastructure that predates modern APIs. In the US, the interoperability landscape is a patchwork of HL7 FHIR implementations of varying quality. Any AI system operating in this space has to be robust to messy, inconsistent, and sometimes missing data, which is where the real ML engineering work lives.
There are also non-trivial considerations around:
These aren't reasons to avoid AI in this space. They're reasons to build it carefully. And as the author sees it, the companies that will succeed long-term in healthcare AI are the ones treating these constraints as design requirements rather than obstacles.
The TechCrunch article raises the question that haunts every AI company automating human work: where is the line between augmenting workers and displacing them? Basata's founders are candid that this is a harder question they'll eventually have to face. Right now, the answer feels easy, staff are drowning, and AI is a lifeline. But the author thinks it's worth being honest that "augmentation now" and "displacement later" are not mutually exclusive outcomes.
The historical pattern in automation is fairly consistent. Technology first handles the most repetitive, high-volume tasks. Workers are redeployed to higher-complexity work. Then the technology improves, handles the higher-complexity work too, and the calculus shifts. This doesn't mean automation is bad, it means the transition requires deliberate policy, retraining investment, and institutional honesty about what's coming.
In healthcare administration specifically, the author suspects the near-term story is genuinely positive. There is so much unmet demand, patients waiting too long, referrals lost in the system, specialist capacity going unused because scheduling is broken, that AI-driven efficiency gains are likely to expand the effective capacity of the system rather than simply reduce headcount. But that assumption deserves scrutiny over time, not just reassurance.
What's clear is that the back-office problem in healthcare is real, it's consequential, and it's finally getting serious technical attention. The next time you're waiting for that call back, remember: somewhere, an admin worker is working through a queue that no human should have to manage alone. AI, built responsibly, might just be the thing that finally lets them breathe, and gets you your appointment.
If you found this analysis useful, the author writes regularly on AI, machine learning, and the systems shaping our world. Explore more on the research page, learn more about the author, or get in touch to discuss any of these topics further.