Most Organisations Aren't Failing at AI. They're Failing at Adoption.

What I've learned running AI capability audits across dozens of businesses, and why the gap between using AI and getting value from it is almost entirely about how you start.

There is a number that should bother every leader more than it does.

Around 78% of organisations now use AI in at least one part of their business. Yet over 80% of them report no measurable impact on their bottom line from it. McKinsey's most recent State of AI research puts the share of genuine "AI high performer", companies seeing more than 5% EBIT impact; at just 5.5%.

Read that again. Near-universal adoption. Almost no value. That is not a technology problem. The models work. They are extraordinary, and they are getting better roughly twice as fast every few months. The problem sits somewhere else entirely, and after running structured AI capability and adoption audits across organisations of very different shapes and sizes; manufacturers, law firms, banks, charities, schools, marketing agencies - I've come to believe the gap is almost always the same gap. It is not about the technology a company buys. It is about the conditions it has, or hasn't, put in place before it buys anything.

This piece is an attempt to describe that gap honestly, and to give you a way of locating yourself on it. I'm not going to tell you AI is important. You know that. I'm going to tell you why knowing it changes so little, and what the organisations that break through actually do differently.

The expensive hobby

Here is the pattern I see again and again. A company "adopts AI." What this means in practice is that a few enthusiastic people start using ChatGPT or Copilot. Some of them are very good at it. They build little personal workflows, save themselves real time, and become quietly indispensable. Leadership hears the buzz, buys some licences, circulates a prompting guide, and considers the box ticked.

Then nothing scales.

The reason is structural, not motivational. Individual brilliance with AI is invisible and unrepeatable. The person who figured out how to cut their reporting time in half did it on their own laptop, in their own head, with no shared standard and no governance. When they go on holiday, the capability goes with them. When they leave, it's gone entirely. Sensitive data drifts into tools nobody approved. Excellent work happens and no one measures it, so no one can prove it happened, so no one can justify doing more of it.

This is the difference between activity and adoption. Activity is people using AI. Adoption is an organisation getting reliably better because of it. The first is easy and feels like progress. The second is hard and is the only thing that shows up in the numbers. Most companies are mistaking the first for the second, which is exactly why 78% have adopted and 80% see nothing.

Why guessing is the real risk

The instinct, once a leader notices this gap, is to act fast. Pick a use case, build something, prove value. I understand the impulse and it's usually wrong, because it skips the question that determines whether any of it works: are we actually ready?

The questions that should be answered before a single workflow is built are the ones that quietly raise the most anxiety, which is precisely why they get avoided. Where are we actually on our AI journey? Which teams are best placed to lead, and which would struggle? What can and cannot be put into these tools? Who owns this? How will we know if it worked? These aren't technology questions. They're readiness questions, and a company that can't answer them isn't ready to build, it's ready to guess.

Guessing has a cost, and it's not the obvious one. The obvious cost is wasted spend. The real cost is risk that grows quietly in the background: data ending up where it shouldn't, inconsistent practices hardening into habits, and a slow erosion of trust when the first AI experiment produces something embarrassing. The organisations that get hurt by AI are rarely the ones that moved too slowly. They're the ones that moved without knowing where they stood.

A simple way to locate yourself

When I assess an organisation's readiness, I look at four things. I'd encourage you to score yourself, honestly, out of ten on each right now.

• Strategy. Is there genuine clarity on what AI is meant to deliver, which use cases matter most, who owns them, and how success will be measured? Or is there interest without direction?
• Skills. Do your people have the confidence, the safe-use habits, and the practical ability to use these tools well, or is capability uneven, concentrated in a few self-taught enthusiasts?
• Systems. Are there approved tools, a usable policy, clear data handling, and real governance? Or is everyone improvising?
• Scale.

In my experience the most common profile, by a wide margin, looks like this: Skills are the strongest pillar, because people are curious and have been experimenting. Strategy is the weakest, because nobody has done the hard work of deciding what AI is for. Systems lag, because governance is unglamorous. And Scale sits in the middle, enough appetite to begin, nowhere near enough discipline to roll out.

If that sounds like you, you're not behind. You're normal. But notice what it means: your people are more ready to use AI than your organisation is to support them. That mismatch is the single biggest predictor of an expensive hobby. Training people without fixing strategy and systems just produces more skilled improvisation. The bottleneck isn't your team's ability. It's the absence of the conditions that let their ability compound.

What "good" actually looks like

I want to be concrete, because abstraction is where AI conversations go to die.

A mid-market law firm I'd describe as paralysed, partners split between treating AI as an existential threat and dismissing it as hype - didn't start by buying tools. They started by getting an honest, evidence-based picture of where they stood. That single exercise did something no strategy deck had managed: it gave a divided partnership a shared set of facts to argue from instead of competing opinions. Only then did training and targeted builds follow. Twelve months on, they had a firm-wide strategy, live tools, and partners who'd gone from sceptical to genuinely engaged. The verified first-year value ran into the hundreds of thousands. None of it would have happened if they'd led with technology instead of readiness.

A marketing agency losing pitches to faster competitors had three senior people quietly using AI in isolation. The unlock wasn't more individual brilliance, it was turning that scattered, invisible capability into shared infrastructure: a common toolkit, a governance framework, and training that embedded the skills into everyday process. Content production time dropped 40%. A monthly report that took eight hours took forty-five minutes. The difference between those two states wasn't talent. The talent was already there. It was structure.

A precision manufacturer trained four engineers and built three live automation pipelines, taking a weekly quality report from three days to twenty minutes. A healthcare provider gave clinical time back to patient-facing staff by upskilling existing admin teams rather than recruiting. The common thread across all of them is not the tools they chose. It is the order in which they did things: understand the ground first, build capability deliberately, then implement against measured baselines.

The sequence that works

If there is one idea I'd want a leader to take from years of doing this work, it's that the sequence is the strategy. Get the order wrong and even good components fail.

The order that works is unglamorous and reliable. First, audit, establish where you genuinely are, what's worth doing, and what's blocking you. Not a vendor demo, not a strategy slide. A clear, evidence-based read on your readiness and your highest-value opportunities, with the risky ones honestly flagged as later-phase. Then build and build narrow. Two or three lower-risk, measurable workflows with real owners and real baselines, so you can prove value before you bet on it. Then train, but train against the specific gaps and the specific workflows the audit surfaced, not as a generic course bolted on the side. Capability that's connected to live work sticks. Capability delivered in the abstract evaporates.

Audit, build, train. The reason this works isn't clever. It's that each stage de-risks the next. The audit stops you building the wrong thing. The narrow build stops you scaling something that doesn't work. The targeted training stops you teaching skills nobody then gets to apply. Most failed AI programmes are some version of doing these in the wrong order, or skipping one entirely; usually the audit, because it's the one that feels like a delay rather than a deliverable.

It is, in fact, the opposite of a delay. It's the thing that makes everything after it faster.

The honest part

I'll be candid about something the breathless AI commentary tends to skip. Most of what works is not spectacular. There's no single jaw-dropping use case that transforms a business overnight. The value comes from many ordinary improvements, a report written faster, a process made consistent, a bottleneck removed – happening continuously across every function, with measurement underneath so you can see them compound. It is volume and discipline that changes the game, not magic.

That's good news, because ordinary and repeatable is something any organisation can build. It just requires resisting the urge to start with the technology, and starting instead with the truth about where you are.

The companies pulling ahead right now aren't the ones with the best tools; everyone has access to the same models. They're the ones who replaced guessing with knowing before they spent a penny on implementation. They closed their readiness gaps, built capability in the right teams, proved value in a few measurable places, and only then expanded with evidence behind them.

The technology will keep getting better whether or not you're ready for it. That's exactly why readiness, not technology, is the thing worth your attention. The question was never should we adopt AI - you already have. The question is whether you know enough about your own organisation to turn that adoption into something that shows up where it counts.

Most don't. The ones that do aren't smarter. They just started in the right place.

If this resonated, I'd genuinely value your perspective: where do you think your organisation sits on those four pillars; strategy, skills, systems, scale? The honest answer is usually more revealing than any technology decision you'll make this year.

https://www.e-careers.com/business-solution/ai-consultation