A persistent issue in the AI discourse is the quiet assumption that organisations must adopt AI. It’s rarely stated outright, but it sits underneath strategy decks, boardroom conversations, and vendor pitches. “AI” becomes an inevitability rather than a choice. The problem is that, in many cases, this assumption arrives before a clear understanding of what adoption actually entails.
That lack of clarity starts with the term itself.
The problem with calling everything “AI”
“AI” has become an umbrella term stretched to cover a wide and often incompatible range of technologies. In one context, it refers to advanced machine learning systems capable of generating content, making predictions, or learning from large datasets. In another, it’s used to describe deterministic automation, rule-based workflows, or even basic analytics tooling.
This elasticity is not just semantic – it has practical consequences.
When organisations say they are “adopting AI,” they may be referring to entirely different capabilities. One team might be experimenting with probabilistic models that introduce uncertainty and require new governance approaches. Another might be implementing process automation that behaves predictably and fits neatly into existing operational structures. Both get labelled as AI, but they carry different risks, costs, and organisational implications.
Without a shared definition, comparison becomes meaningless. Benchmarking progress across teams, sectors, or competitors becomes unreliable because the underlying thing being measured is inconsistent.
The result is a fragmented landscape where adoption is claimed, reported, and sometimes celebrated, but not necessarily understood.
Measurement without a baseline
This ambiguity carries through into how organisations attempt to measure impact.
There is no universally accepted framework for evaluating AI initiatives. In practice, organisations default to a mix of familiar but often insufficient metrics:
- Efficiency gains (time saved, throughput increased)
- Cost reduction (headcount, operational expenditure)
- Output volume (more content, more decisions, more interactions)
These are easy to quantify, which makes them attractive. But they are also partial.
What’s often missing are harder, more meaningful dimensions:
- Decision quality: are outcomes actually better, or just faster?
- Risk exposure: has uncertainty increased in ways that are not being tracked?
- Capability development: is the organisation becoming more adaptive, or just more automated?
- Innovation capacity: are new possibilities being unlocked, or are existing processes simply being accelerated?
In many cases, metrics are defined after the fact, retrofitted to justify investment rather than designed to test it. This creates a subtle but important distortion: success becomes something that is demonstrated, not something that is interrogated.
“You can’t measure AI if you haven’t defined it”
If the goal is to prove that AI works, almost any metric can be made to cooperate. If the goal is to understand its impact, the bar is much higher.
The leadership constraint
It’s tempting to frame AI adoption as a technical challenge: selecting the right tools, integrating systems, managing data pipelines. These are real concerns, but they are not the limiting factor in most organisations.
The more significant constraint is leadership.
Specifically, the ability to do three things well:
1. Define strategic intent
Why is AI being introduced in the first place? Is the goal to reduce cost, improve quality, increase speed, enable new products, or reposition the organisation competitively? Vague objectives like “stay competitive” or “drive innovation” are insufficient – they don’t translate into actionable decisions.
2. Establish evaluation criteria upfront
What does success look like before implementation begins? Not in abstract terms, but in measurable, observable outcomes. What would count as failure? What trade-offs are acceptable? Without this, initiatives drift, and post hoc justification becomes the default.
3. Align initiatives with organisational reality
AI systems do not operate in isolation. They interact with processes, people, incentives, and culture. If these are misaligned if, for example, performance metrics reward speed over accuracy, then the impact of AI will follow those incentives, regardless of stated intentions.
Where these leadership capabilities are weak or underdeveloped, AI adoption tends to become performative. Tools are deployed, pilots are launched, announcements are made, but the underlying organisation does not meaningfully change.
Performative vs. transformative adoption
This distinction matters.
Performative adoption is characterised by visible activity without structural impact. It often includes:
- Isolated pilots that never scale
- Metrics that highlight activity rather than outcomes
- Narratives that emphasise innovation without demonstrating it
- Tool-centric thinking (“we need AI”) rather than problem-centric thinking (“we need to solve X”)
Transformative adoption, by contrast, is quieter but more substantive. It involves:
- Clear linkage between AI capabilities and strategic goals
- Redesign of processes to incorporate new capabilities
- Willingness to challenge existing assumptions about how work is done
- Measurement frameworks that capture both benefits and trade-offs
The difference is not the technology itself. It’s the clarity of intent and the discipline of execution.
“With AI, definition must precedes evaluation.”
Starting with better questions
Before adopting AI, organisations would benefit from pausing, briefly, but deliberately, to answer two foundational questions:
1. How do we define AI in terms that are specific enough to be useful?
This does not require a universal definition. It requires a working definition that is precise within the organisation. What capabilities are included? What are not? Where are the boundaries between automation, analytics, and AI? Clarity here enables coherent decision-making later.
2. What tangible outcomes would we accept as evidence of genuine impact?
Not just activity, not just output, but outcomes that matter to the organisation’s objectives. These should be defined in advance, measured consistently, and revisited critically.
These questions are not complex, but they are often skipped. And in their absence, organisations default to imitation – following industry trends, vendor narratives, or competitor behaviour without a clear sense of why.
A more grounded path forward
AI is not a singular capability, and adoption is not a binary state. It is a spectrum of tools and approaches that can be applied in different ways, with different consequences.
Treating it as an inevitability obscures that nuance.
A more grounded approach starts with definition, moves through deliberate measurement, and is sustained by leadership that is willing to be specific – about goals, about trade-offs, and about what success actually looks like.
Without that, the discourse will continue to generate noise rather than insight, and organisations will continue to report progress without being able to meaningfully compare, evaluate, or learn from it.
The technology will keep advancing. The question is whether the way we think about it will keep up.
First dropped: | Last modified: April 24, 2026