Reflect
The time traveller
On organisational maturity, the debt beneath every AI initiative, and the future headline we are still in time to avoid.
I blinked hard at my screen, trying to force my vision back into focus.
A visual migraine had arrived without warning, scattering fractal colours across my sight like a broken kaleidoscope. My head pulsed in time with my heartbeat. I stood up from my desk, half intending to call it a day, when a headline on my screen caught my eye:
"Government Suing Industry for ROI Failure in AI."
I paused, one hand on the desk, the other reaching for ibuprofen.
The pain won. I stepped into the kitchen, swallowed the tablets with some water, and came back to my study. Curiosity got the better of me. I opened the article.
It was damning.
After years of investment, failed implementation, and underwhelming returns, the government was reportedly seeking damages from industry for overstating the maturity of AI and agentic technologies, and for overselling their ability to automate work across multiple sectors. Industry, in turn, was blaming poor digital foundations, weak process maturity, and a lack of organisational readiness. The article went on to describe how the situation had been made worse by the loss of experienced subject matter experts, leaving institutions with a growing capability gap and limited ability to recover.
I read the opening twice. Then I saw the date.
30 March 2030.
My head throbbed again. The room faded to black.
Yes, I know. It is cheesy. A bit melodramatic. Bad sci-fi at its finest.
But the point still stands.
Because while I have not actually travelled through time, I do think there is a version of the future where this kind of story writes itself. Not because AI is inherently doomed, and not because the technology has no value, but because the right tools in the wrong hands can create as many problems as they solve.
And right now, too much of the conversation around AI still assumes that buying capability is the same as being ready to use it.
It is not.
AI does not arrive in a vacuum
One of the most persistent mistakes in AI transformation is treating adoption as a tooling decision.
Buy the model. Procure the platform. Launch the pilot. Demonstrate the ROI.
That sounds neat in a slide deck. It sounds decisive in a board paper. It sounds modern, ambitious, and commercially sensible.
But it is not how adoption actually works.
AI does not land in an empty, neutral environment. It lands on top of everything that already exists inside an organisation: the quality of its data, the clarity of its workflows, the strength of its governance, the confidence of its workforce, the maturity of its leadership decisions, the consistency of its delivery model, and the trust people have in the systems around them.
If those foundations are weak, AI does not somehow rise above them. It operationalises them.
That is the part I think we still underestimate. AI is not just a new capability layer. In practice, it is often an acceleration layer. It makes underlying strengths more useful, and underlying weaknesses more visible. If your processes are coherent, your governance is solid, and your people know what good looks like, AI can support real augmentation. If not, it can amplify noise, inconsistency, and risk at speed.
A large language model connected to fragmented data, inconsistent workflows, and unclear accountability is not transformation. It is just a faster way to generate plausible-looking confusion.
Why organisational maturity matters
I work predominantly in healthcare and life sciences, and this question of maturity comes up again and again.
A while back, I was doing work with an organisation where I referenced the HIMSS Analytics Maturity Assessment Model (AMAM). What struck me about the model was not only the ambition of the upper levels, but the honesty of the lower ones. It starts from a very real place: fragmented point solutions, variable analytics capability, uneven governance, and a long journey toward anything that resembles predictive, personalised, or prescriptive intelligence.
That matters.
Because too many organisations want to talk about AI as though those earlier stages are optional. As though you can simply jump from fragmentation to intelligent automation without doing the slow, foundational work in between.
But maturity is not just about technical stack. It is about whether the organisation can absorb and use capability responsibly. It is about whether people understand the outputs, whether processes are stable enough to integrate the technology, whether data is trustworthy enough to support it, and whether governance can keep pace with the new risks it introduces.
This is especially true in healthcare, where the cost of ambiguity is not just financial. It can affect safety, quality, equity, and confidence in care.
Maturity is not the boring bit before AI. It is the enabling condition for AI.
The problem is not a lack of ambition
Most organisations are not short on ambition. If anything, the opposite is true.
The pressure to "do something with AI" is intense. Boards want a strategy. Executives want productivity gains. Clinical and operational leaders want relief from overwhelmed services. Suppliers want momentum. Governments want modernisation. Everyone is looking for proof points, quick wins, and signs that they are not being left behind.
That pressure creates a very understandable temptation: to move directly to visible AI use cases before addressing the invisible debt underneath them.
The trouble is that the debt does not disappear just because it is inconvenient.
It waits.
And when AI arrives, it tends to collect interest very quickly.
That is why I find the idea of organisational debt so useful. Often, what appears to be a technology problem is actually the compounded effect of longstanding debt in different parts of the organisation. The AI initiative simply happens to be where those debts finally become impossible to ignore.
Technical debt, connectivity debt, integration debt
In many healthcare settings, these are not edge cases. They are daily operating conditions.
Legacy systems, overlapping platforms, limited interoperability, brittle interfaces, locally optimised workarounds, and years of deferred architectural improvement all shape what is possible. Add capital constraints, competing priorities, and limited delivery capacity, and you have an environment where foundational modernisation is always important but rarely urgent enough to win attention against immediate service pressures.
This is not a criticism. It is reality.
The problem comes when AI programmes are designed as though this reality does not exist. If your architecture is fragmented, integration is difficult, and access to data relies on fragile joins between systems that were never designed to work together, then your AI solution inherits that instability. It may still demo well. It may still generate excitement. But scaling it into a reliable operational service becomes much harder.
That does not mean organisations must wait for perfection before doing anything with AI. Perfection is not coming. But it does mean they need to be honest about the condition of the estate they are building on. Responsible AI strategy is not just about aspiration; it is about designing within real constraints.
Literacy debt and capability debt
Another common failure point is the assumption that organisational understanding will somehow catch up later.
In many organisations, the knowledge sits with a small number of specialists, programme leaders, or enthusiastic early adopters, while the broader workforce is expected to absorb change without the same level of confidence, context, or support.
That creates fragility.
If only a handful of people can explain how an AI-enabled service works, interpret its outputs properly, describe its limitations, or spot when it is behaving inappropriately, then adoption remains shallow. Trust becomes uneven. Misunderstanding grows. Resistance hardens. And once the original champions move on, momentum fades with them.
Literacy does not mean turning everyone into a data scientist. It means giving leaders, managers, clinicians, operational teams, and support functions enough understanding to ask better questions, challenge weak assumptions, use tools safely, and recognise where human judgement still matters most.
Without that, organisations do not adopt AI. They borrow it.
Process debt: the part nobody wants to talk about
Many organisations rely on workarounds that have gradually hardened into "the way we do things here." Informal handoffs. Duplicate entry. Unwritten rules. Escalations based on who happens to be around. Variability hidden inside local custom and professional habit. Decisions that depend on tacit knowledge rather than clear operational design.
AI cannot sit cleanly on top of a process that is not actually understood.
If the workflow is inconsistent, the decision points are fuzzy, and the exceptions are doing most of the real work, then AI will not remove friction. It will add speed to confusion. It may even give the illusion of improvement while making the underlying process less transparent and harder to govern.
Before asking whether AI can automate something, organisations should first ask whether the task itself is sufficiently defined, governed, and standardised to support safe augmentation. That is not anti-innovation. It is what responsible innovation looks like.
Data debt: the truth beneath the demo
AI depends on data that is accessible, reliable, structured appropriately, understood in context, and governed well enough to be used with confidence. Yet many organisations are still dealing with fragmented records, inconsistent coding, missing metadata, variable data quality, ambiguous definitions, and limited interoperability between source systems.
Poor data quality does not just undermine reporting. It undermines retrieval. It undermines model inputs. It undermines user confidence. It undermines the ability to explain outcomes.
Generative AI does not magically bypass this. In some ways it makes the challenge more subtle. A powerful model can produce language so fluent and persuasive that weak grounding becomes harder to detect. If the source information is incomplete, inconsistent, or badly governed, then the resulting outputs may still sound credible while being operationally unsafe or contextually wrong.
If the data estate is immature, AI maturity will be constrained by it. Not eventually. Immediately.
Human-centred design is not optional
AI changes work, not just systems. It affects how people make decisions, where they place trust, how they interpret outputs, when they intervene, and what they feel responsible for. If these human factors are ignored, even technically capable solutions can fail.
An AI solution may be accurate in a lab, but unusable in practice. It may be efficient on paper, but disruptive in a real workflow. It may be impressive in a demonstration, but impossible to trust in the moment it matters.
That is why human-centred design must be more than a line in a strategy document. It has to shape discovery, design, implementation, assurance, and evaluation. The question is not only "what can the model do?" but "how does this fit into the way people actually work, decide, and care?"
Most failed AI programmes are not failures of intelligence
They are failures of organisational maturity.
When AI initiatives fail to deliver value, the post-mortem often focuses on the visible layer: the vendor, the model, the interface, the business case. Those things matter. But beneath them, there is usually a more fundamental issue.
The organisation was trying to deploy intelligence into an environment that had not yet made itself ready to receive it.
In that context, AI becomes a mirror as much as a tool. And not every organisation likes what it sees reflected back.
Protecting the future
So what do we do with that? We start by being more honest.
Honest that AI is not a shortcut around transformation. Honest that organisational debt does not disappear because we have found a new technology to talk about. Honest that pilots are easier than adoption. Honest that maturity is uneven. Honest that value depends on readiness. Honest that some of the most important work is still deeply unfashionable: governance, standardisation, integration, capability building, service redesign, and disciplined change.
If we want AI to support productivity, quality, resilience, and better outcomes, then we have to invest in the substrate beneath it. People. Process. Data. Governance. Architecture. Leadership. Trust.
That is not separate from AI strategy. That is AI strategy.
Because without organisational maturity, AI does not become a force multiplier for improvement. It becomes sand in the gears.
And if we are not careful, the future headline writes itself.