AI does not fail in hospitals because it is inaccurate. It fails because it does not fit.
Wings+ April 2026 Edition
From the WINGS Editorial Desk
Healthcare is usually explained through outcomes.
Diagnosis. Treatment. Recovery.
That is what people see.
What determines whether anything new is used sits elsewhere. Approvals, clinical workflow, data access, internal risk decisions. Different teams, each with their own constraints, rarely moving at the same speed.
AI has entered this environment with strong technical results. Models can detect disease, prioritise scans, support decisions in ways that were not possible a decade ago.
And yet, a large share of these systems never become part of daily practice.
They are tested. Sometimes validated. Occasionally approved.
Then they sit.
Not because they fail.
Because they do not fit into what already exists.
For early career professionals, this is the part that matters.
Performance is visible.
Adoption is negotiated.
Why This Matters Now
The conversation around AI has shifted.
It is no longer about what the technology can do. That question has largely been answered.
The pressure now is operational.
Can it be deployed?
Can it be trusted?
Can it sit inside existing systems without disrupting them?
Healthcare makes these constraints harder to ignore.
Workflows are rigid for a reason.
Responsibility is shared across multiple layers.
Regulation is continuous, not occasional.
That combination changes how decisions get made.
A system that performs well in isolation may still fail once introduced into a real clinical environment.
That failure is rarely technical.
It is usually organisational.
In many cases, the issues are already known. Data is fragmented. Ownership is unclear. Workflows are not designed for change. Fixing those problems requires time and coordination, so they are deferred.
Until they block progress.
Case Study: When Accuracy Is Not Enough
AI tools in radiology have been delivering strong results in controlled settings for years.
Adoption, however, has been uneven.
Not because clinicians doubt the technology. Because it interferes with how they work.
An extra step.
A separate interface.
A delay in review.
In a controlled environment, those are minor issues.
In practice, they are enough to be ignored.
Clinicians will default to what is reliable under pressure, even if it is less precise.
So the system works.
It just is not used.
This pattern repeats more often than most organisations admit.
The Hidden Constraint
From the outside, AI adoption looks like a technical problem.
Inside hospitals, it is not.
It is a question of alignment.
Does the system sit inside existing workflows, or does it sit alongside them?
Does it reduce effort, or quietly increase it?
Even small friction points matter. An additional login, a duplicate step, a delay in retrieving results. These are enough to change behaviour.
Most systems are built to perform.
Far fewer are designed to fit.
That is where adoption breaks down.
Exclusive Interview
Irina Wade, Business Development & Strategy Lead, Qure.ai
Irina works across clinical systems and global health programmes, focusing on how AI is deployed in environments where performance alone is not enough.
Why AI tools fail to embed in clinical workflows
WINGS: Many AI tools demonstrate strong validation results but never become embedded in daily clinical workflow. In your experience, what quietly prevents adoption inside hospitals, even when the technology itself works?
Irina:
AI adoption in healthcare depends less on model performance and more on mindset, alignment, and integration. For AI to be fully adopted, stakeholders across clinical, technical, and administrative departments must work together to ensure it fits seamlessly into existing systems.
The primary reason adoption fails is workflow misalignment. If an AI tool does not integrate smoothly with systems like PACS, RIS, or EMR, even minor disruptions can create resistance among clinicians. Hospitals are not looking for “better AI” but for invisible AI, tools that enhance workflows without adding friction.
Another key barrier is ambiguity in ownership. Because AI is still relatively new, responsibility is often unclear across IT, radiology, procurement, and clinical teams. When no single group owns outcomes, pilots can stall or fail without accountability.
Finally, incentives matter. Even if AI improves efficiency, it will not be prioritised unless it also increases revenue, reduces liability or improves measurable hospital metrics. Without clear value alignment, AI remains a “nice to have” rather than an operational necessity.
Regulation as constraint and advantage
WINGS: You operate in a highly regulated environment. Has regulatory clearance accelerated your strategy, or constrained it in ways founders often underestimate?
Irina:
The impact of regulation in healthcare AI depends on where you are in the journey. In the early stages, regulatory clearance is often underestimated. Requirements such as documentation, clinical validation, and post-market surveillance take significant time and effort. More importantly, they force companies to demonstrate that the product works reliably in real-world settings, not just in controlled environments.
As organisations move into the scale-up phase, regulation becomes an advantage. Clearance acts as a trust signal for hospitals, lowers perceived risk and creates a barrier to entry for competitors. It also provides a framework for standardisation, making it easier to deploy solutions consistently across sites.
Governance introduces what can be called productive friction. Regulators push for representative data, preventing overreliance on curated datasets. They also shift the focus toward outcomes-based validation rather than technical metrics. Requirements around traceability and auditability further strengthen reliability and accountability.
While regulation can slow development, it improves the long-term trajectory of a product by ensuring robustness and trust in complex clinical environments.
Where AI creates real impact
WINGS: In global TB and lung health programmes, where does AI genuinely move the needle, and where does it risk becoming a surface solution?
Irina:
For us, the most meaningful impact of AI is in screening at scale, particularly in low-resource settings. In these environments, the ability to detect cases faster and prioritise high-risk patients is essential and delivers immediate, measurable impact. AI can triage chest X-rays where radiologists are overburdened or scarce, ensuring critical cases are identified early.
Another contribution is the standardisation of interpretation. Variation between clinicians can lead to inconsistent diagnoses. AI reduces this variability, delivering more consistent outputs across different settings.
It also reduces missed early-stage diagnoses, which are often subtle and easy to overlook under time pressure. By flagging potential abnormalities, AI acts as a safety net.
Finally, AI accelerates workflows by prioritising abnormal scans, enabling faster decision-making and reducing delays. Its impact is strongest when embedded into end-to-end systems, not used as a standalone layer.
The real sustainability trade-offs
WINGS: Sustainability is often discussed but rarely defined. In practice, what is the hardest trade-off you face?
Irina:
The toughest trade-off is scale versus depth. Organisations must choose between deploying widely with lighter integration or going deep with full workflow integration and outcome tracking. Trying to do both often leads to diluted impact.
There is also tension between speed and rigour. Moving quickly captures opportunity but risks insufficient validation. Prioritising rigour slows deployment but strengthens reliability.
Another challenge is commercial viability versus accessibility. The regions with the greatest need often have the least ability to pay. Sustainable models must balance both.
Finally, there is the challenge of consistency versus localisation. Global models offer scale, but must adapt to local realities such as infrastructure and clinical protocols.
Sustainability is not fixed. It is a continuous negotiation.
AI as responsibility
WINGS: When reviewing early-career candidates, what distinguishes someone who understands AI as a technology from someone who understands it as a responsibility?
Irina:
This difference is visible in how candidates think. Those focused on technology discuss models, metrics and performance.
Those who understand responsibility ask different questions. Who is affected if the system fails? Has it been tested in real-world conditions? Are there biases in the data?
Their focus extends beyond performance to accountability and consequence.
As a result, their approach to optimisation changes. It is no longer about improving metrics. It is about ensuring reliable outcomes for people.
Career Clarity
Healthcare makes one thing very clear.
AI is not judged by how well it performs.
It is judged by how safely it operates inside a system that already exists.
This changes what organisations look for.
Technical ability matters. It is assumed.
What stands out is awareness. Understanding where data comes from. Who is responsible for it. What happens when something goes wrong.
People who see that layer early tend to be brought into more complex conversations sooner.
Not because they are more senior.
Because they reduce risk.
Cross Sector Relevance
This is not unique to healthcare.
In enterprise systems, projects stall at approval.
In hospitals, they stall at integration.
In infrastructure, they stall at permission.
Different environments. Same constraint.
Technology moves quickly.
Institutions do not.
Understanding where that friction sits changes how you move inside them.
Closing
AI is often framed as progress.
Inside organisations, it exposes something more familiar.
How decisions are made.
Where responsibility sits.
What is allowed to move forward.
The technical layer is visible.
The rest is not.
Those who understand both do not just build.
They recognise what will actually be used.
That is where leverage begins.