In my work with companies going through significant change, I see very different approaches to AI.

Some are still cautious. Others are experimenting extensively. Some are already committing significant resources to AI transformation, sometimes with the CEO personally driving it as a major priority.

It is difficult to know what the right approach is. AI is evolving so quickly that an AI transformation plan that makes sense today may need to be reconsidered six months from now. But waiting until the picture becomes clear is not really an option either.

There is no playbook for what comes next. Yet companies still have to move.

This is what makes AI transformation particularly challenging. How much should you commit and how much should you experiment? When should you stay focused and when should you adapt? How do you pursue AI transformation without losing sight of business fundamentals?

Looking at how companies are navigating these questions, I keep coming back to five tensions. Both sides of each tension make sense. Both also carry risks. The challenge is to find the right balance.

1. Focus vs. Adaptation

AI transformation requires focus. Companies need clear objectives, priorities and enough persistence to turn decisions into results.

But the environment keeps changing. New capabilities emerge, costs fall, and solutions that looked promising six months ago can quickly become outdated.

The risk on one side is rigidity: sticking to a plan that no longer makes sense. On the other, it is constant change: reacting to every new development, shifting priorities and exhausting teams without ever going far enough.

A useful distinction is between the objective and the path. Business objectives can remain relatively stable while the technology and the way to achieve them evolve.

Be persistent about the outcome, flexible about the path.

2. Experimentation vs. Transformation

Experimentation is essential. Companies need to test use cases, learn what works and understand where AI can create real value before making major commitments.

But experimentation can also create an illusion of progress. Pilots multiply, employees use new tools, training programmes are launched and AI initiatives appear across the organization. There can be a lot of activity without much actually changing.

It is a little like running five kilometres every week and assuming that this alone will prepare you for a marathon. It is useful. You are learning and building capabilities. But if you really want to run the marathon, at some point the depth and intensity of the effort need to change.

The opposite risk is moving too quickly into large-scale transformation, committing significant resources or redesigning the organization around assumptions that have not yet been sufficiently tested.

The challenge is knowing when you have learned enough to make a bigger bet.

Experiment enough to learn, commit enough to transform.

3. AI Transformation vs. Business Fundamentals

There is a real risk in doing too little. Companies have customers to serve, targets to deliver and many other strategic priorities. AI can easily remain a side topic until competitors have moved much further.

But I increasingly see the opposite risk too. Some leaders, including CEOs, become so passionate about AI that it starts consuming disproportionate attention, resources and organizational energy.

Every week brings a new tool, a new possibility or another initiative. The risk is that AI gradually becomes an agenda of its own.

AI should not become a parallel agenda. It should help the business perform its fundamentals better: serve customers, grow, innovate, improve productivity and quality, make better decisions and execute faster.

The risk is therefore on both sides: missing a major transformation by treating AI as secondary, or becoming so focused on AI that the company loses sight of the business it is supposed to improve.

Transformation is not the objective. Better business performance is.

4. Mobilization vs. Workforce Reduction

AI transformation will ultimately be executed by people.

Employees will have to learn how to use new tools, change the way they work, redesign processes, test what works in practice and adjust what does not. They will implement the change in day-to-day operations and, in many businesses, explain new ways of working to customers and help them accept those changes.

They are not simply contributors to the transformation. They are the ones who will make much of it happen.

This makes training critical. Companies cannot expect people to transform their jobs without giving them the skills, time and support to learn how to work differently.

But there is an obvious tension. Some of the productivity gains companies are seeking from AI may ultimately mean fewer people, different roles or very different skills.

Employees understand this too.

One risk is to focus primarily on technology, productivity and headcount reduction. The company may have the right tools and plans but struggle to execute because people are insufficiently trained, engaged or willing to make the change happen.

The opposite risk is to avoid the difficult workforce implications in an attempt to preserve engagement. If leaders communicate only opportunity and empowerment while employees anticipate potential job losses, trust can quickly disappear.

Resistance should not automatically be treated as a lack of adaptability. Some concerns are entirely rational.

Companies will need to train people, involve them in execution, communicate honestly and explore reskilling or redeployment where it makes sense. At the same time, they may eventually have to make difficult workforce decisions.

How do you fully mobilize people to execute a transformation that may ultimately require fewer people or very different roles?

5. Bottom-Up Innovation vs. Centralization

Many of the best AI use cases will probably emerge close to the work itself.

People inside functions understand their problems, processes and customers. Giving them room to experiment can create speed, learning and applications that a central team might never have identified.

But too much decentralization creates its own problems. Different teams adopt different tools. Similar use cases are developed several times. Costs increase, knowledge remains fragmented and security or data risks become harder to control. What works locally may also never scale across the organization.

Centralization addresses many of these issues. Companies can choose platforms, establish governance, share capabilities, allocate resources and turn successful use cases into common ways of working.

But centralize too much or too early and the company can slow down learning, create bureaucracy and reduce the initiative of the people closest to the work.

The balance may also change over time. Some things need to be decentralized to discover what works, then standardized to scale. Others, such as security or data governance, may require stronger central control from the beginning.

The challenge is to know what to centralize, what to leave bottom-up, and when to move from one to the other.

Keeping the Business at the Center

AI may profoundly change how companies operate, compete and create value. But AI transformation should never become an objective in itself.

Companies still have a business to win: serving customers better, building a stronger value proposition, outperforming competitors, growing, improving productivity, protecting margins and creating value.

The risk is to become so focused on AI that the transformation itself becomes the objective. The opposite risk is to focus so much on today’s business that you miss how AI could fundamentally improve your ability to compete tomorrow.

This is perhaps the most important balance to keep: stay focused on the business, the customers and the competitive battle, while constantly asking how AI can help you perform better.

The objective is not to become an AI company. It is to become a better business, and increasingly, AI may be part of what it takes to get there.