There’s an awful lot of talk about AI in healthcare but not always in ways that reflect what's actually happening in care settings.
What resonates with us at Blüm Health isn't the bold claims, it’s the ones grounded in honesty from real world experience. The reflections from people who are genuinely trying to make this work, who understand the gap between what technology can theoretically do and what people and a pressured healthcare system can realistically absorb on any given Tuesday.
As a digital health and AI manufacturer, with clinicians in our team, we're in care environments regularly and the picture we see is complicated, human, and often quite different from the one we hear being described in conference keynotes.
Healthcare is, and will remain, a human-first industry. No matter how sophisticated our models become, care is still delivered between people and trust, empathy, and context aren't edge cases, they are the system.
Care is relational. Technology isn't.
This might sound obvious, but it's worth saying plainly: healthcare has never been about information processing, it’s about what happens between people.
A clinician reading a patient isn't just analysing data points. They're picking up on hesitation, on fear, on something in the way a person describes their symptoms that doesn't quite add up. That interpretive layer, built from experience and shaped by the specific human in front of them, is what makes clinical judgement different from pattern recognition.
Patients, for their part, don't experience care as a transaction. They experience it as a relationship. Whether they feel heard. Whether they trust the person treating them. Whether they leave feeling safer than when they arrived.
Technology introduced without regard for this dynamic tends to create friction, not flow. We've seen it happen in practice far too frequently; a tool arrives that genuinely improves a process, and still it struggles, because it disrupts something relational that wasn't visible in the workflow diagram.
The role of AI, as we see it, isn't to replicate that human layer but to protect it. To take enough pressure off the system that clinicians actually have the time to be present with their patients.
Adoption is harder than it looks. Always.
Anyone who has worked in implementation knows this. The gap between a good tool and a used tool is enormous, and it's rarely a technical gap.
Behaviour in clinical settings is shaped by factors that are difficult to model in advance. The staffing levels on a particular ward. The culture of a team and how it handles change. The confidence of the person expected to use the tool. The timing of the rollout relative to everything else that team is absorbing. Whether the person who championed it is still in post six months later.
A well-designed feature can fail because it arrives at the wrong moment in a workflow. A capability that saves time in theory can feel like an imposition in practice. Across the environments we work in, unfortunately this tends to be more typical than exceptional.
The organisations we see making meaningful progress tend to share a few characteristics. They treat implementation as an ongoing process, not a project with a go-live date. They build in time to observe and listen before they optimise. They're genuinely curious about resistance rather than just trying to overcome it.
AI fits some parts of healthcare better than others. That's worth being honest about.
Administrative burden is one of the most consistent problems across health systems. It's also one of the areas where AI can make a clear and relatively straightforward contribution, structured tasks, repeatable processes, work that consumes clinical time without adding clinical value. There's a strong case for pursuing these opportunities, and we are and should continue to pursue them seriously.
The picture gets more complicated when workflows involve ambiguity, individual patient variability, or decisions where context matters as much as data. In those areas, efficiency is harder to define. The optimal path isn't always the fastest one. And standardised approaches can miss things that experienced clinicians would catch.
In our discovery work with clients, the most impactful applications of AI are rarely the ones attempting to replace clinical judgement. They're typically the ones that reduce friction around it. Lighter administrative load before a consultation. Better information flow between teams. Clearer visibility into what's happening so that clinicians can focus their attention where it's genuinely needed.
This is often a less dramatic version of AI than the one usually described at industry events but it's the one that holds up in practice.
So where does that leave us?
What healthcare needs from technology partners right now isn't more ambition, it’s more honesty. Honest about where AI performs well and where it doesn't. Honest about what's ready to deploy and what still needs work. Honest about the conditions that need to be in place for any of this to have the impact and outcomes it promises.
The most impactful solutions we're seeing aren't the ones trying to replace human decision-making instead it’s the ones quietly removing friction around it. That distinction matters more than it might sound. It's the difference between technology that competes with clinical judgement and technology that creates more room for it.
At Blüm Health, that's the kind of partner we're trying to be. We're not interested in AI that removes the human element from care, our focus is on AI that gives it more room.