Why AI Success Depends on Leadership, Not Just Technology

Most of the executive teams I work with have been investing in AI for a few years. The ones who are frustrated are not the skeptics. They are the believers whose programs have not connected to the P&L. They have the pilots, the internal momentum, the board slide showing everything in flight. What they do not have is a clear line between that activity and business performance, and at this point in the AI cycle, that gap is no longer acceptable.

I spent several years running AI at scale inside Kroger and its data science subsidiary 84.51°, where we processed millions of predictions per second across thousands of store locations. We measured work in margin, basket size, and customer retention rather than how many models were in production, whether the pilots were impressive, or if the work moved the business. That experience shaped how I think about what AI requires from leadership, and what most leadership teams are still getting wrong.

The executives I work with are not confused about whether AI matters. They are managing tighter margins, more expensive capital, and boards that want results rather than roadmaps. In my experience, closing that gap comes down to three things.

1. Value has to show up on the P&L

Most companies can tell you exactly how many AI models they have running. Very few can tell you what those models are worth to the business. AI can improve both sides of the income statement through better personalization and smarter pricing that support revenue. Automation and sharper forecasting cut costs and waste, but most companies are spreading investment across too many initiatives with too little connection to enterprise value. They are generating activity without changing their economics.

The question worth asking is not where the company is using AI. It is where AI is changing the unit economics of the business. Most organizations cannot answer the second one.

2. Velocity is an underrated strategic advantage

Almost every large organization knows more than it can act on. Data and insight exist, but the distance between signal and response is slow. Decision cycles drag, functions operate from different assumptions, and by the time internal alignment happens, the moment has often passed.

I watched this play out firsthand in financial services. A team built models to identify customers of competing firms most likely to switch in a specific line of business. The analysis was sound and the models performed. What followed was months of organizational hesitation and revisited governance questions long after the pilot had proven viable. By the time leaders made a decision, the market conditions had shifted, and they exited the business. Someone inside summed it up perfectly, “The surgery was successful, but the patient was dead.”

The technology worked. The moment was gone. AI can close that gap through faster reporting, better forecasting, and earlier anomaly detection. It is not about doing things cheaper. It is about being able to move when it matters, and that is as much