Anyone who thinks that the Chief Technology Officer or Chief Information Officer should own AI adoption has it wrong: The direct, personal involvement of a company’s CEO is essential for any significant transformation to succeed. Yet, many CEOs have remained on the sidelines, unclear of what exactly they must do, or happy to leave things in the hands of specialists.

Organisation-wide AI initiatives are difficult to orchestrate, and the effort put into them often goes no further than what the highest-ranking individual demands. Based on our experience with leading and advising AI and data efforts, companies with bold CEOs who have the courage to question everything and make fundamental changes are best placed to succeed. 

Below, we propose five non-delegatable responsibilities that CEOs must take the lead on.

1. Don’t limit your ambitions

Given the rapid pace at which the technology is evolving, we understand why many CEOs may prefer to make smaller bets on easier AI projects. While this suffices as a first step, CEOs shouldn’t limit their ambitions.

Using AI to help automate processes and improve productivity – be it by using generative AI assistants or chatbots such as ChatGPT, Perplexity and Claude – is just the beginning. AI is already enabling organisations across diverse industries to rethink their entire business models, from warfare and national planning to marketing and education.

CEOs who don’t think big risk being left behind. Every company and government agency should be asking itself how it can use AI to solve long-standing business problems, gain a competitive advantage and future-proof its strategies and operations. Of course, CEOs must decide which AI projects to prioritise and fund in the face of board pressure, internal company dynamics and employee concerns. Our advice? Don’t let politics hamper your ambitions.

2. Learn to live with uncertainty

The reality is that everything about AI is uncertain. Estimated benefits and feasibility vary greatly, and many initiatives have either failed entirely or failed to progress beyond the pilot stage. Trust in AI is often low, and national governments are unsure of how to regulate it.

This is not a comfortable environment for leaders who’ve spent their entire careers trying to reduce uncertainty. Making matters worse, we don’t see any relief in sight. Our best historical analogy involves the invention of the printing press, where it took two full generations for some semblance of order to emerge.

Senior leaders, especially CEOs, should embrace this uncertainty and learn to experiment in everything from technologies and funding mechanisms to people and organisational structures. Some experiments, such as those that involve moving from model to adoption, will require considerable effort. And plenty will fail. But part of experimentation involves reaping the benefits of successful experiments and learning from failed ones.

One final point: “Dabbling” in AI, perhaps motivated by FOMO (fear of missing out), is not a smart experiment to undertake. Such halfway measures aren’t comprehensive enough to succeed but may well convince you that “AI is not for us” or “the technology isn’t ready”.

3. Make data a first-class citizen

As James Betker of OpenAI noted, “The ‘it’ in AI models is the dataset.” Data availability and quality are generally regarded as the number one limiting factor for AI – as highlighted by the recent failed launch of French AI chatbot Lucie – and will play a big part in separating the winners from the losers. 

Of course, very few companies will actually develop LLMs (large language models) of their own, and they can’t do much about data scraped from the internet being used to train these models. Instead, companies must focus on the data they use to augment LLMs and to train and operate predictive AI models.

The issues run broad and deep: Much data is simply wrong, poorly defined or even hard to find; issues such as relevancy and bias bedevil individual projects; and models can provide bad answers (e.g. inferences) and hallucinate. Exacerbating this, most data is “unstructured” (i.e. emails, contracts, forms, recordings of meetings and so forth) and subject to less scrutiny than data held within corporate systems. 

Even when leaders recognise that their company’s data isn’t ready, some have misdiagnosed this as a technical problem – and not the management problem it really is. CEOs should therefore launch and fully support aggressive data quality programmes that focus on the cross-department processes that supply AI applications.

4. Build AI into your organisation rather than bolt it on

Arthur W. Jones, an organisational design expert, once observed that organisations are perfectly designed to achieve the results they achieve. AI presents some tough organisational challenges for CEOs, such as getting people to understand their responsibilities and implementing new ways of working within and across departments. In some respects, this shouldn’t be surprising, considering that most of today’s organisations were designed for industrialisation, not data and AI.

Here is where the CEO’s role is especially crucial. To meet these challenges, CEOs must take ownership of reshaping and readying their organisations for a world in which data and AI are central. This means driving a shift that promotes the right qualities: clear accountability, cross-functional collaboration, adaptability and a commitment to continuous learning. They must foster a culture where learning from both success and failure is valued, where teams are empowered to test and improve and where the entire company becomes more agile, resilient and ready to evolve.

CEOs shouldn’t do this by replicating a particular structure or best practice, but by building the capacities and virtues required to embed data and AI into the heart of the company’s operational and strategic work. This will require a lot of effort, considerable experimentation and some painful choices. But there is no getting around it.

5. Act boldly and quickly, but thoughtfully

Acting on AI requires urgency and boldness – but not haste. Many organisations today fail to address the fundamentals, leading to stunning inefficiencies. For example, employees spend around 30 percent of their time dealing with data issues, a figure that will likely be exacerbated by AI.

At the same time, both customers and employees are already experimenting with generative tools, often contrary to their organisations’ policies. If leadership waits too long or acts without direction, others will set the pace, sometimes in ways that compromise trust, quality or competitiveness.

CEOs must lead with intent to shape the conditions for responsible adoption. Success doesn’t come from rushing in, but from moving decisively with a clear understanding of where value lies, what the organisation can absorb and how to evolve both culture and capabilities to deliver lasting impact. Indeed, many companies “let a thousand flowers bloom” yet fail to get a single application off the ground. Experimentation is key, but only implementation counts as a win.

CEOs must set the pace

Many senior leaders and CEOs have remained on the sidelines when it comes to data, analytics and AI. And we get it – the technology is daunting, and the pace is dizzying. The responsibilities we prescribe may seem new, unexpected, unfamiliar and demanding. But the risks associated with adopting a “wait-and-see” approach or not driving this change are simply too great.

We understand that CEOs may have reservations about AI. After all, the success rates are low. But some companies are breaking through and, as best we can tell, the direct involvement of CEOs is decisive.

This is a based on an article published in Chief Data Officer Magazine.