The CEO Group's AI Leadership Forum, held at the Freshfields Law Firm in Lower Manhattan in July, looked beyond the hype and addressed the actual strategic governance and human capital implications of AI transformation. Through interactive panels and table discussions, participants considered how to identify valuable AI opportunities, build the necessary foundations, implement reliable systems, avoid innovation traps, navigate the regulatory environment, manage workforce transitions, and balance bold innovation with responsible oversight. This conversation revealed both the tremendous opportunities and significant challenges facing leaders who lead their organizations through this technology revolution. Some takeout.
Leaders and revolutions: “Think like a mid-level executive.”
Two executives leading large-scale AI transformations, Lara Abrash, chairman of Deloitte, and Sabastian Niles, president and chief legal officer of Salesforce, shared insights about leading hundreds of thousands of employees and customers through large-scale AI transformations. “Right now, in this environment, all of you in this room are grappling with the most complex set of problems I've seen in over 30 years of business,” Abrash said. How they do it:
Total trust, total honesty. For both Abrash and Niles, building employee trust during an AI transformation requires leaders to recognize what they don't know, rather than projecting a false sense of confidence. “It's more important to say what you don't know than what you know,” Abrash said. “Because if people are interested in this AI story and they say to their employees, 'Nobody's going to get fired,' that's not true. That's a lie. They're saying that everyone will upskill and it's going to be a great company, and that's probably not true either.”
“What good is that for me?” General messages about the benefits of AI fail to address individual employee concerns, especially among experienced employees. “Think about it in the shoes of a 50-year-old mid-level executive. They've been in this industry for 25, 30 years,” Abrash says. “They know that better than anyone. Basically, we want them to develop technology that will do one of two things in their heads: Either they put them out of work, or they put them through the last few years of working that they don't know if they can do it.”
Skeptics become champions. When properly engaged, the most knowledgeable employees are often the strongest advocates for AI. “I’ve actually been trying to find all the AI skeptics,” Niles said. “These people are the secret sauce because they know the work best, and they have become some of the most interesting advocates for the vision of human-AI collaboration.”
Beginner's mind. Successful AI implementation requires a balance between openness to new approaches and existing organizational knowledge. “When you have expertise, when you have deep lived experience, when you apply it as a company or in government, society, community, when you have established processes and workflows in place,” Niles said. “Beginner's mind is part of it, and one is to allow the people who know their job best to think safely, but can we do this differently?”
Focus more on the future. Modern leadership requires a constant shift in focus between immediate business needs and long-term strategic planning. “I found myself moving between horizons in a way I had never experienced before. For a very long time, my leadership was 90 percent focused on the here-and-now horizon and 10 percent focused on second and third horizons,” Abrash said. “The premium, the success factor, for boards and executives is what is effectively going to go from 1 to 2 to 3 very quickly right now.”
Leave the office to learn. “I meet with our technology partners and customers to understand what they want from us and where and how we can best help,” Abash said. “And what they want now is a business advisor. They don't want someone who's going to sell them some shiny technology. They want someone who's going to talk to 100 other CEOs.”
Take enough risks. Organizations need to find a middle ground between moving too fast and moving too slowly with AI adoption, but doing nothing is not an option. “We're talking about two things,” Niles said. “If you go to 100% AI, you're putting the whole company at risk. If you go to 0% AI, you're putting the whole company at risk, right?” Abash agreed, noting that risk is an essential part of innovation, adding that the company recently greenlit the largest single investment in its 180-year history pursuing generative AI. “If you expect management to be accountable for every penny, to not make mistakes, to be perfect, you're not going to get innovation,” she says.
Build trustworthy and valuable AI
As AI adoption accelerates, the gap between promise and performance is becoming a critical business challenge. With 88 percent of AI pilots failing to make progress and only 25 percent of CEOs realizing measurable ROI from their AI investments, moving forward requires more than technical prowess: trust.
Dominique Shelton Leipzig, CEO of Global Data Innovation, outlined a comprehensive framework for building trusted AI that enables companies to innovate quickly while maintaining risk control.
trust framework
Triage (T): Strategically align AI use cases with business objectives and legal obligations. This includes understanding jurisdictional requirements, such as the EU's 17 AI-prohibited areas, and working to ensure core business strategies are supported, rather than fringe projects where ROI is difficult to demonstrate.
Right data (R): Ensure data accuracy, proper governance, and legal rights for your training applications. “Are there intellectual property rights, privacy rights, business rights to train with that data? And are there audit logs in case someone challenges them?” Leipzig stressed.
Uninterrupted testing (U): AI models can drift at any time, so perform continuous monitoring and auditing. “I'm not saying AI can be hallucinating… It can be wrong every second, every minute of every day. So if your brand is at risk, you want to know and be able to fix it.”
Supervision (S): A human monitoring system that can quickly react if the AI performs outside of acceptable limits.
Technical documentation (T): Comprehensive logging and metadata allows technical teams to diagnose and fix model drift as it occurs.
“We are at a crossroads. We must harness these technologies to realize the potential of AI, which will soon add nearly $30 trillion to the global economy,” Leipzig said. “But you have to do it in a way that actually delivers ROI and long-term value.”
