Imagine two companies. 

Both are investing in artificial intelligence. Both are running pilots. Both are exploring the same technologies. 

But that’s where the similarity ends. 

One is translating those efforts into real revenue growth and using AI to reshape its business model. The other remains stuck in experimentation – busy, but not materially better off. 

This is the divide highlighted in PwC’s latest Global AI Performance Study. It finds that nearly three-quarters of the economic value created by AI is being captured by just 20 per cent of organisations. 

So, what is driving the disparity, and what can other businesses do to close the growing divide?  

The way David Lee sees it, the top‑performing companies are not simply deploying more AI tools. Instead, they are embedding AI into their business models as opposed to simply layering it on top.  

“It’s not just about doing things faster or more efficiently. It’s about doing things differently,” he told me on the latest episode of The Tech Agenda podcast. 

Lee has witnessed it first-hand. As Chief Technology Officer of PwC Ireland, in addition to supporting clients on their AI journeys, he is also responsible for ensuring his firm benefits from AI.  

He has seen the issue from both sides, and he is clear about the issues. 

“The reality is that the organisations that are actually seeing the return is that they are going beyond the back office,” he says, later adding: “It’s not about the tool, it’s about how you apply it.” 

*****

The PwC report is comprehensive. The firm collected data for a survey from 1,217 senior executives around the world, including from Ireland, at a director level or above, at companies across 25 sectors and multiple regions worldwide.  

It found that companies were not short on activity when it comes to AI. But it highlighted a disparity in how they were embedding it into their business, and on the returns it was generating. 

It found that companies getting the most value were those that were twice as likely to redesign workflows to incorporate AI rather than simply adding AI tools. Plus, those companies were three times more likely to have increased the number of decisions made without human intervention. 

“Importantly, it’s not just efficiency gains. It’s organisations who’ve also looked at it as a tool of reinvention,” said Lee in relation to why certain companies are benefiting while so many are not. 

“So, not only are they taking cost out of the organisation, but they’re also generating new revenue streams from the application of AI.” 

PwC’s analysis points to two broad areas of difference: how organisations are using AI, and the foundations they have built to support it. In practice, Lee said both come back to execution. 

Beyond the back office

Most companies have taken a cautious starting point. This is something Lee understands. After all, AI is changing so quickly that it is often hard for companies to keep up. In his experience, many have started the process with traditional functions. 

“If we look at most organisations, they tend to start in back office functions, whether that be finance, HR or IT,” he said. 

“People would be understandably reserved about trying a new technology on their customers. So, the logical starting point was to look at back office functions where effectively you were learning in a safer environment.” 

However, Lee said that the companies generating meaningful returns are not stopping there, as evidenced by the latest PwC survey. 

“The reality of the organisations that are actually seeing the return is that they are going beyond the back office. They’re applying it in front office, customer-facing settings, and also back into the supply chain,” he said. 

According to Lee, that shift – from internal efficiency to customer-facing and operational impact – is where the value begins to move. 

There is also a difference in the level of ambition. Many organisations are focusing on relatively simple use cases. These are easier to implement, easier to measure, and less likely to fail.  

But, according to Lee, they also tend to deliver limited returns. 

“There’s a fundamental difference between organisations who are using AI  in a true end-to-end manner within their organisations and those who are  not,” Lee said. 

The companies seeing results are tackling more complex problems, he said. 

“Unfortunately, like most things in life, the prize is in solving the complex, and the same is true of AI. So, those organisations who’ve actually stood back and tried to solve the more complex problems are those that are seeing a return,” he said. 

The problem with productivity

One of the challenges in assessing AI is that early benefits tend to show up at an individual level. 

Tools can improve personal productivity: summarising documents, generating content, or assisting with analysis. For employees, the gains are immediate. 

“It’s very difficult to argue at this point that the use of an AI tool does not help your individual personal productivity,” Lee said. 

However, as Lee has witnessed first-hand, it takes longer to see gains when AI is rolled out across the company.  

“It’s very hard to translate personal productivity to a level of enterprise return,” he said. 

That gap – between individual efficiency and organisational performance – is where many companies stall, he said, arguing that it requires changes to workflows, processes, and decision-making and not just access to tools. 

“The real challenge is for businesses to see a return on the AI investment from an organisational point of view, not just a personal productivity tool.  

Mindset over tools

Based on Lee’s experience, the key divide is how companies think about AI. 

“It’s not about the tool, it’s about how you apply it,” Lee said, adding that the tools being used today could well be totally different in just a matter of months. “It’s moving that fast.” 

Lee said companies also need to think about how they measure productivity, and what they  expect as a return on their investment.  

“The risk is we use traditional measures,” in reference to the fast-changing nature of AI.  

“We look at things like return on investment.  We are using traditional measures like return on investment to measure the return on AI investments.  There is an evolving school which says there’s a value in learning from the use of technology that will sustain beyond the advances in technology.  

So, you see people starting to talk about return on experimentation. By starting and learning, you bank the learning as the technology evolves.” 

According to Lee, waiting for the technology to mature may seem sensible, but it comes with a cost. 

“The risk is that people will wait until the technology gets better and not start,” he said.  

Despite widespread activity, relatively few companies have scaled their use of AI. 

“We see somewhere between 10 and 15 per cent of organisations having really scaled AI. In other words, something like 80 per cent to 85 per cent of organisations have not,” he said. 

The reasons are not particularly surprising. 

“One is understanding where to get the return… and legitimate concerns around the governance around its use,” he said. 

Trust, whether with customers, employees, or regulators, remains a key consideration, he said. 

“Organisations have worked hard to build relationships of trust with their employees, with their customers, and in certain industries with their regulators. They’re not going to risk those relationships by an inappropriate use of the technology,” he said. 

Building the foundations

A key theme of the report is that the companies pulling ahead are not necessarily those experimenting the most. They are the ones investing in the underlying infrastructure. 

“They’ve also invested time in the hard stuff, like building the foundations. AI without data, without trusted data, is of limited value,” he said. 

The organisations that are benefitting have invested “in the plumbing”, he said. 

“Get the data right so that you can actually rely on the insights that you’re getting from the AI that runs over that technology. They are investing in the upskilling of their people. So, they’re actually spending money. And they have an investment outlook which says, we do need to see a return here, but we’re going to take a broad view on the return. The return can be in terms of revenue we generate, costs we save, or learnings we accrue,” he said.  

A key difference highlighted in the study is how AI is positioned within organisations. 

“The organisations who are seeing a return are seeing this as a business tool and not as a technology tool,” Lee said. 

That shift changes how decisions are made and how initiatives are governed. 

“The governance protocols that they put in place to oversee the use of the technology, draw not just on the technology functions in the organisation, but have appropriate business representations who can then assess both the use of the AI and also the return on it,” he said. 

Rather than being layered onto existing processes, Lee said that AI needs to become part of how processes are designed. 

Reinvention, not optimisation

The concept of reinvention has been used widely in recent years, but in this context it has a specific meaning. 

“Let’s not just take the process that we have currently and see how we can apply AI to make it better or faster. It’s actually about saying can we do this in an entirely different way?” he said. 

He said the result is not just incremental improvement, but a different way of operating – whether in how companies engage with customers, manage operations, or structure their internal workflows. 

Another notable finding is the extent to which leading companies are automating decisions. 

“Less than 10 per cent of respondents were happy to use AI in an autonomous way, even for very basic decisions,” he said, adding that even in low-risk scenarios, companies are reluctant to remove human oversight. 

“There is just a resistance to using without the human in the loop.” 

But he said the companies seeing stronger returns are taking a different approach. 

“Those that are actually being a bit more ambitious in that space are seeing the return. It’s not that they’re reckless… they’ve actually put in place the right foundations to ensure that the AI behaves in a consistent manner,” he said. 

Trust and governance 

As AI becomes more embedded, governance becomes more important, something Lee is keen to point out. 

“Trust applies not just in terms of how you face off externally, but as importantly in terms of how you face off to your own employees,” he said. 

He added that internal trust is a significant factor. 

“It is hard to read media without getting a concern about, will AI replace my job?” he said. 

The companies seeing results are addressing that directly. 

“Their own employees are more likely to trust the technology… as a consequence of those organisations having invested in both the upskilling of their employees, but also giving them the comfort that the appropriate checks and balances are in place.” 

Closing the gap

For companies that have yet to move beyond experimentation, Lee’s message is clear. 

“Most definitely doing nothing is not an option.” 

The starting point is straightforward, he said: “Start with the right mindset. Think of it not just as a tool of efficiency. Think about how we can use it to do things differently, not just faster. Do invest in the ‘boring stuff’. Do think through the governance. Do think about what you need to do with data. Don’t trade your trust easily, but don’t hide behind it either.” 

For Lee, data is key.  

“Historically, organisations have always known that they should be doing more with the governance management of the data that they have,” he said. “They now actually have a need and an incentive for doing that. It can be overwhelming. Most organisations would stand back and say it’s very hard to undo the sins of 30 years in a year.  

“The organisations that have been successful in this space have focused on the data that’s important to the business need that they’re trying to address through AI. So, like the famous phrase about eating an elephant, bite it off in chunks.”  

The Tech Agenda with Ian Kehoe podcast series is sponsored by PwC.