The AI Mistake That's Costing Companies Millions

Jan de Vries · 2026-09-16

AI is moving into boardrooms faster than many businesses can figure out what to do with it. For Jim Harris, the bigger risk isn't moving too slowly. It's bolting new technology onto old ways of working and expecting transformational results. Harris is a specialist in GenAI and disruptive innovation. He advises executive teams and has worked with organizations including American Express, Barclays, IBM, SAP, Walmart, and the UK Cabinet Office. In 2024, he was named Speaker of the Year by TEC Canada for his work on AI. In this exclusive interview, Harris explains why companies need to redesign processes from scratch, how leaders can reduce employees' fear of AI, and why keeping humans involved remains critical as adoption accelerates. ### Create Psychological Safety First When introducing AI, the first step is to create psychological safety. The majority of people in developed nations actually fear AI. They fear job loss. As adults, we tend to fear things or are uncomfortable with new ways of working. An organization has to work really hard to say three things: - Your jobs are safe. - We're going to provide you training. - This is going to enhance your skill set and your employability. Those are a few things organizations have to do. ### Don't Just Sprinkle AI Fairy Dust Applying AI to an existing business process can limit the gains companies achieve. Harris shared a powerful example from Ernst & Young (EY) in the US. EY has 400,000 employees and has spent a billion dollars on AI. They had a 46-hour risk management process they'd historically run for clients. They thought, "Let's apply some AI to it." With the people who designed the process, they got it down to 32 hours—a 20% gain, which is great. But then they said, "Let's try something different. Let's take people who've got no experience with the process, but who are technology-first, AI-first, chomping at the bit, new individuals in our organization, and pair them with risk managers who've never been involved in this process." They got the process down to 15 minutes. Then they said, "Well, we're EY, we have to have data compliance and we have to verify everything." So they added some steps back in and got it back to six hours. Look at this: 46 hours, 32 hours, 15 minutes, then back to six. It highlights a couple things. Number one, you can't just sprinkle some AI fairy dust on your existing processes and hope to get the greatest gain. You really need to start with a blank sheet of paper and say, if we were designing this process today from first principles, what would we do? In other words, we're not going to take a 2018 or before process and try and enhance it with AI. How do we redesign it? Secondly, the people who have been involved in designing the process, unconsciously or consciously, are wedded to it. We need to have fresh eyes to bring in. When you actually look at 46 hours down to six hours, that's a more than 80% reduction. That is staggering. And now you can redeploy people for more important activities. ### AI Beyond ChatGPT: Transformational Gains There are so many examples, but Harris has had the pleasure of being in interview sessions with Demis Hassabis, the founder of Google DeepMind, who won the Nobel Prize in 2024. Why? PhD student... > "You can't just sprinkle some AI fairy dust on your existing processes and hope to get the greatest gain." – Jim Harris ### The Bottom Line The lesson is clear: don't just layer AI onto old processes. Rethink from the ground up, involve fresh perspectives, and keep humans in the loop. That's how you unlock AI's true potential.