Enterprise AI. Impacts that matter
AI has stopped being a future bet. It is already an operating reality — and advantage is accruing to the organisations that move on it decisively. The question is no longer whether AI matters. It’s whether your organisation is built to capture what it now makes possible.
It’s not ambition
It’s approach
The value is real. Yet investment stalls.
Most enterprises now agree AI matters. Far fewer are scaling it. The gap isn’t belief — it’s the operating discipline required to act on belief inside an enterprise.
Hesitation is a structural result of ambition outpacing operating discipline. Leaders are asked to commit without agreement on who owns value, risk, or the data underneath. AI investment gets framed as experimentation rather than enterprise change. In that environment, slowing down isn’t timidity — it’s a rational response to ambiguity that no-one has resolved.
Organisational factors — culture, management, talent practices — account for 2x the real AI impact of individual effort alone.
Source: Microsoft 2026 Work Trend Index.
Three ways it shows up.
The tension between value and compounding doesn’t look the same in every organisation. It tends to surface in three places.
Fragmented pilots.
Initiatives running across the business, each producing local value, none of it accruing to the enterprise. Different teams using different tools, learning different lessons, with no mechanism to capture or scale what works. The longer this runs, the harder it gets to unwind.
Zombie AI.
Programmes that look alive — budget allocated, governance in place, communications going out — but produce no compounding return. Activity gets confused with progress. The honest test: if you switched off your AI initiatives tomorrow, would the business notice?
Zombie AI (noun): An AI programme that has consumed investment, generated activity, and produced no compounding enterprise value — because the decisions that govern it were never made. Ambition without approach is how Zombie AI gets built — quietly, expensively, one fragmented pilot at a time.
The agent management gap.
Agents are not tools. They take action, hold context, and produce work. Microsoft’s telemetry shows active agents grew 15x in a single year, and 18x inside large enterprises. The management layer hasn’t grown at all. Most organisations are running a growing agent estate with no plan for how it’s governed, evaluated, or retired.
Why this is happening.
Microsoft just published the research that explains it. They surveyed 20,000 AI users across ten markets and mapped them across two dimensions: how ready they are as individuals, and how ready their organisations are to support that readiness.
The finding is uncomfortable. In 1 in 10 organisations, the most capable people have already built the skills, found the use cases, and changed how they operate. The system around them — the incentives, the decision rights, the manager behaviour, the way work is measured — hasn’t caught up.
That’s the gap. Not a skills gap. Not an ambition gap. The distance between what your people can now do and what your organisation lets them do.
It’s also measurable.
The same forces accelerating AI adoption at the individual level are suppressing it at the enterprise level. Your best people have already adapted. The organisation around them hasn’t.
It’s not a technology problem.
The instinct, when AI investment underperforms, is to buy more technology. More licences. More tools. More training programmes.
The data says the opposite. Culture, manager support, and talent practices drive twice the AI impact of individual mindset and behaviour. The factors most companies are investing in rank fourth.
This isn’t a tooling problem. It’s an operating model problem.
Don’t add AI to your roadmap. Rebuild the map around AI.
The most expensive mistake we see is treating AI as the next item on a list — finish the CRM, modernise the estate, then think about AI. It’s the wrong order. AI is already changing what apps need to be, and what data actually matters.
Apps are changing shape.
Modernising them first rebuilds the past, locking in assumptions AI will disrupt.
Data needs to change too.
AI doesn’t want more data. It wants the right data: structured, governed, semantically clean.
Foundations still matter.
AI now decides what ‘good’ looks like for cloud, security, and the data estate.
Roadmaps built before AI are roadmaps to the wrong place.