Two companies roll out the same AI tool. At one, the CEO talks about it only during the kick-off presentation, then returns to the topic at a meeting six months later to ask why the investment has not paid off. At the other, the same CEO uses the tool personally, talks about how it has helped or disappointed them, and asks their people what is and isn’t working for them. The technology is identical. The results a year from now will be completely different. And the difference will not lie in any of the parameters companies usually monitor — licence costs, model quality, or speed of deployment.
At one of our clients recently, we saw a very clear example of how strong a signal leadership can send through its own behaviour. One of the top managers programmed a simple personal AI assistant for his work. Not as an “AI initiative” for the company. Simply because it made sense to him and saved him time.
When he later showed the other directors what he had built and how he used it during an off-site, their reaction was more interesting than the tool itself. Roughly half of his colleagues came up to him and asked him to send them his Markdown file because they wanted to try it too. The other half essentially told him that they really did not want this.
And that, in our view, is what real AI adoption looks like. Not everyone will be enthusiastic. Nor do they have to be. What matters is that someone in leadership stops talking about AI as a transformation programme and starts using it as an ordinary work tool. This creates something no mandatory e-learning course can: proof that it has practical value, and legitimacy for others to start experimenting as well.
This is the fourth and final article in a series about what a company needs to address to make AI work in day-to-day operations. The previous three articles covered roles, incentives, and processes. All three, however, have one thing in common: without visible leadership and meaningful communication, even the best-designed role, the best incentive system, and the best-redesigned process will not last more than a few months.
What the data tells us about leadership
In its State of AI 2025 research, McKinsey found that companies achieving real business value from AI are three times more likely to have their senior leadership openly take ownership of and responsibility for AI initiatives¹. This is not just about formal support on paper. Specifically, it means that leaders actively use AI themselves and thereby become role models for others¹.
MIT research adds that a lack of leadership support is one of the main barriers preventing AI projects from getting off the ground, alongside reluctance to adopt a new tool and concerns about the quality of its outputs². When leadership merely approves the budget and then pays no further attention to the topic, the signal sent to the rest of the company is not neutral. It signals that this is a side project, not a priority.
Why visible leadership support is more than a gesture
In the first article in this series, we wrote that employees reject AI when they feel their own roles are under threat, and that one of the five mechanisms for preventing this is visible leadership support that legitimises the redesigned role³. This is not an additional, standalone recommendation. It is the point where all three previous topics in the series come together.
A redesigned role without leadership support feels like a temporary solution that someone higher up will cancel as soon as a way is found to automate the role entirely. Changes to incentives without an explanation of why they are changing feel like an ad hoc bonus, not a signal that the company has decided to do things differently. A redesigned process that leadership does not follow and enforce itself will be quietly bypassed during the first crisis, and everyone will return to the old way of working.
In other words, leadership is not a fourth, separate item on the list alongside roles, incentives, and processes. It is what keeps the other three alive for longer than a few weeks after launch. A company can design the perfect role, a clever incentive system, and a well-thought-out process, but if no one in leadership publicly and repeatedly stands behind them, all three will gradually slip back to their original state, because nothing prevents that from happening. In this sense, leadership is not a trigger for change at the beginning, but a sustaining mechanism that must be present all the way through, not just on launch day.
Leadership’s role is not to manage the details, but to set the direction
An interesting finding from MIT research is that more successful AI deployments are found in companies that combine visible leadership support with decentralised decision-making about exactly where AI should be deployed⁴. Success rates were higher when proposals for specific use cases came from the people on the ground — line managers who know their operations — rather than from the top, from a central AI team that lacks detailed knowledge of every department⁴.
This means that leadership’s role is not to manage every detail of deployment. It is to set the direction, demonstrate through its own example that it is committed, and then give people closer to individual areas of work the space to decide exactly where and how to use AI. Leadership that tries to plan every use case centrally usually ends up with use cases that look good in a presentation but do not reflect how people actually work — precisely the problem we discussed in the third article in this series, about processes.
In practice, this can look quite unremarkable. A head of strategy who uses AI to prepare materials for a quarterly review and openly explains where it saved time and where its output confused them does more for adoption than any internal email about the company’s “AI strategy”. People a few levels down do not need an order. They need to see that someone they respect is trying the tool, making mistakes with it, and talking openly about those mistakes.
AI communication should emphasise growth and innovation, not savings
McKinsey found another difference between companies that get value from AI and those that do not: more successful companies set growth or innovation as their goal, rather than cost savings alone¹. This has a direct impact on how AI should be communicated. If leadership presents AI deployment solely as a way to cut costs, it falls straight into the trap described in the first article in the series: employees interpret it as preparation for reducing headcount and respond accordingly, quietly sabotaging the tool or not using it to its full potential.
Successful companies communicate differently: AI is a way to do more, not a way to do the same with fewer people. BCG describes this as a combination of commitment to behavioural change, a people-oriented approach, and long-term investment in core capabilities⁵, alongside a focus on building trust while maintaining clear rules about when and how AI outputs can be trusted⁵. AI communication that forgets this balance and talks only about efficiency risks precisely the loss of trust we discussed in the second article in the series: more than half of employees distrust AI and fear for their jobs⁶, and a poorly chosen communications framework only reinforces that fear.
What happens when leadership is absent
Without visible leadership, an AI initiative usually does not fall apart immediately. It unravels slowly. In the first few months after deployment, enthusiasm is sustained by a handful of people who try the tool out of personal curiosity. But without leadership acknowledging it or commenting on it, that enthusiasm fades as soon as people encounter the first difficulty and find it easier to return to the old way of working. Six months later, leadership notices that the investment has not paid off and asks why AI has not “taken hold” in the company, even though, according to the technical specifications, it has worked exactly as intended from day one. The answer is almost never the tool itself. It is that after the initial presentation, no one in leadership spoke publicly and repeatedly about their own experience with AI, so people concluded that it was not a real company priority.
Three questions for leadership
Before declaring an AI initiative officially launched, leadership should take the time to answer three questions.
1. Do I use the tool enough myself to talk about it from personal experience, rather than just from a presentation prepared by someone else?
2. Have I left decisions about specific use cases to the people who do that work every day, or have I tried to plan them centrally on their behalf?
3. Do people in the company hear me say that the goal is to do more and do it better, or do they come away from my communication thinking that this is mainly about savings that could personally affect them?
The technology works; what is missing is the adaptation layer
Four articles, four different areas of research, but one shared conclusion: in most unsuccessful AI deployments, the technology works. What is missing is systematic work on everything around it: roles that give people meaningful work, incentives that encourage the right behaviour, processes that genuinely integrate AI into the way work is done, and leadership that leads by example and communicates AI as an opportunity rather than a threat.
It is no coincidence that these four topics are so interconnected. In the 1960s, organisational psychologist Harold Leavitt described a model that now bears his name: a company can be understood as four interconnected elements — task, people, structure, and technology — and a change in one will always affect the others⁸. If a company changes only the technology and leaves tasks, people, and structure unchanged, the result is exactly what this series has addressed: a tool that runs, but does not change how anyone in the company actually works.
MIT research sums it up with one figure worth repeating at the end of this series: this year, 95% of companies are getting no measurable return on their investments in generative AI⁷. Not because AI does not work, but because the four interconnected changes described above are missing around it.
This is the final article in a series about what a company needs to address to make AI truly work: roles, incentives, processes, and leadership. It draws on publicly available research and Kogi’s practical experience.
Sources
1. McKinsey QuantumBlack, “The State of AI in 2025: Agents, Innovation, and Transformation”. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (verified 20 August 2026)
2. MIT NANDA, “State of AI in Business 2025” (The GenAI Divide). https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf (verified 20 August 2026)
3. Harvard Business School Working Knowledge, “Why Employees Resist AI, and How Companies Can Win Them Over”. https://www.library.hbs.edu/working-knowledge/why-employees-resist-ai-and-how-companies-can-win-them-over (verified 20 August 2026)
4. See source 2, section on strategic partnerships and decentralised implementation.
5. BCG, “The Leader’s Guide to Transforming with AI”. https://www.bcg.com/featured-insights/the-leaders-guide-to-transforming-with-ai (verified 20 August 2026)
6. Knowledge at Wharton, “How Can Companies Incentivize AI Adoption?”. https://knowledge.wharton.upenn.edu/article/how-can-companies-incentivize-ai-adoption/ (verified 20 August 2026)
7. See source 2.
8. Umbrex, “Leavitt Diamond (Task, People, Structure, Technology)”, a summary of Harold Leavitt’s model. https://umbrex.com/resources/frameworks/organization-frameworks/leavitt-diamond-task-people-structure-technology/ (verified 20 August 2026)