The divide you cannot see from the boardroom
In early 2025, executives around the world were still announcing AI initiatives with familiar optimism. Budgets were growing and pilots were multiplying. Inside many of those same organisations, a quieter divide was forming.
Some teams had woven AI into their daily work. They drafted, analysed and coded with it, and their output was visibly changing. A few desks away, other teams barely used it, or used it quietly with tools nobody had approved.
From the boardroom, both looked like “AI adoption”. On the ground, they were moving at different speeds.
AI is not lifting all boats. It is creating distance, between organisations and, just as importantly, within them.
The gap between organisations is well documented
The divide between companies is now measurable. Boston Consulting Group's 2025 research found that only 5% of companies are “future-built” and consistently generating substantial value from AI, while around 60% report minimal revenue and cost gains. Why that gap compounds over time, through learning that late movers cannot buy, is the subject of The Great AI Split.
Less discussed is the version of the same divide that runs through a single organisation. It matters because the organisation-level gap is largely the sum of the internal ones.
The divide inside the organisation
The internal divide runs along three lines.
Level. BCG's AI at Work 2025 survey of more than 10,600 workers in 11 countries found that more than three-quarters of leaders and managers used generative AI several times a week, while regular use among frontline employees had stalled at 51%. BCG called it a “silicon ceiling”. The people closest to customers and operations, where much of the value sits, were the least likely to be using AI routinely.
Support. The same survey found that only around a quarter of frontline employees felt they had strong leadership support for AI, and that people who received more than five hours of training were markedly more likely to be regular users. Adoption was not mainly a question of interest. It tracked investment in people.
Sanction. Where official tools lag, people bring their own. Microsoft and LinkedIn's 2024 Work Trend Index found that 78% of AI users brought their own AI tools to work, and only 39% of users had received AI training from their employer. That creates a divide between work done with approved, governed tools and work done with tools the organisation cannot see.
The result is an organisation in which some teams are compounding their capability while others stand still, and in which leaders may be reading high average adoption as broad progress.
Why the internal divide matters
The difference between organisations that pull ahead and those that stall is rarely the model. It is structural capability: the systems that let AI be absorbed, deployed and improved.
AI behaves less like a software upgrade and more like a learning system. Better data produces better models; better models improve workflows; improved workflows generate more data. That loop only runs where AI is used in real work, at volume. If adoption is concentrated in a few teams, so is the learning.
The divide also concentrates risk. Teams using unsanctioned tools move data outside governed systems. Teams that are not using AI at all may be the ones best placed to spot where it fails. And a workforce split into confident users and anxious non-users makes every later change harder.
The organisations pulling ahead treat data as infrastructure, embed AI outputs directly in workflows with clear decision rights, reuse what each deployment teaches them, and run AI as a shared operating layer rather than a set of isolated pilots. Each of those depends on adoption being broad, not patchy.
How to close the internal divide
Closing the gap inside an organisation is mostly a management task, not a technology one. Five steps make a measurable difference.
- Measure adoption by team and level, not on average. An overall usage figure hides the divide. Break it down by function, grade and location, and look for the teams that are not moving.
- Train on real work. Generic courses have limited effect. Short, hands-on sessions built around each team's own tasks, with time set aside to practise, change behaviour.
- Give frontline managers the mandate. Support from senior leaders matters, but the decisive support comes from the manager who protects time for experimenting and treats early mistakes as learning.
- Make the approved route the easiest one. If sanctioned tools are slower or weaker than consumer alternatives, people will route around them. Provide good tools quickly, with clear guidance on what data can be used where.
- Spread what works. Find the teams that are getting results, document how, and pair them with teams that are behind. Internal examples persuade more than vendor case studies.
What changes for people
For many organisations, the hardest adjustments are behavioural rather than technical. AI introduces probabilistic insight into systems designed for certainty, and it accelerates feedback cycles that traditional governance was never built to handle.
The role of people shifts with it. Instead of routine analysis, teams interpret AI outputs, manage exceptions and improve the systems themselves. A marketing team may spend less time producing reports and more time testing campaign ideas. A supply chain manager may move from manual forecasting to supervising automated planning.
This is where many organisations stall. Legacy structures, approval chains, risk processes and rigid job descriptions slow the feedback loops AI depends on, and they slow them most in the teams that are already behind.
Leaders who recognise this are rethinking organisational design: flattening hierarchies, embedding data specialists in business teams and building governance that allows experimentation without sacrificing accountability.
The result is not simply better technology. It is faster organisational learning, spread more evenly.
A gap that widens quietly
The internal divide will not look dramatic. There is no single moment that separates the teams that pull ahead from those that stall. The gap widens gradually.
One team launches products slightly faster. Another cuts its costs a little more each quarter. A third quietly builds data that improves every model it uses. Over time those advantages accumulate, and the organisation's overall position becomes the sum of them.
For boards and executive teams, the question is not only whether the organisation is investing in AI. It is whether that investment is reaching the people and teams where the work is done.
The real divide is not between those who use AI and those who do not. It is between those who learn faster with it and those who cannot, inside your own organisation as much as outside it.
