Predicting the unpredictable: Artificial Intelligence to understand CEO decisions

We live in unpredictable times, whether in terms of the global economy, climate, or geopolitics. It is important to anticipate risks in order to ensure humanity’s future resilience. This means making sense of increasingly complex information with the help of new tools. In this series, we look at the work of researchers striving to improve their forecasts.

Corporate governance is essential. It is the system by which companies are held accountable for their actions, balancing the interests of shareholders, management, customers, and communities. The challenge lies in the fact that human behavior can be unpredictable, as executives’ decisions do not always aim to maximize profits, shareholder value, or sustainability goals.

“Through our research, we try to understand what decision-makers focus on. What’s really exciting is that we can build an analytical framework that seeks to uncover the objectives CEOs appear to pursue through their decisions. Before this study, it was difficult to directly observe what they were trying to optimize,” explains Boris Nikolov, Professor of Finance at HEC Lausanne.

The researchers used a vast dataset on the investment and financial policies of publicly listed U.S. companies over several decades, including data on CEO compensation as well as ESG (environmental, social, and governance) ratings and other indicators.

They then designed a machine learning algorithm that they trained on the dataset. The technique used is called inverse reinforcement learning (inverse reinforcement learning). The goal is to infer which objectives corporate leaders seek to maximize when making investment and financial decisions. The researchers already knew the outcomes of these decisions, as recorded in the dataset. 

“We can now ask whether the decisions made by corporate leaders are optimal. Are they efficient? Do they match what we expect of them? Are they making good or bad decisions? Are they biased?” explains Professor Nikolov.

The professor and his team found that corporate leaders place great importance on profits. Notably, their objectives and those of shareholders are not perfectly aligned. ESG criteria also play a role, but complying with environmental standards represents a cost for executives. By contrast, the social and governance dimensions are perceived more positively, as they are seen as benefits rather than costs.

“We found that executives tend to overinvest in every aspect of the company, beyond what would be optimal for maximizing profits. This is explained by their desire to build true empires, even when doing so reduces shareholder value. It stems from the fact that CEO compensation is tied to the size of the assets they manage. They therefore have an incentive to accumulate them,” the professor explains.

While executives can hold press and analyst conferences as well as meetings with investors to talk about the future, it is their concrete actions that really matter, because these can be evaluated using this new tool.

“We propose an analytical framework that can now help board members hold CEOs accountable, not only for their statements but also for their actions. The aim is to eliminate biases in managerial decisions, provide better guidance, and put optimal policies in place,” says Boris Nikolov.  

He adds: “Ultimately, this machine learning tool can be used to improve decision-making and facilitate regulatory compliance, and to better analyze CEO behavior and performance.”

This research could also transform corporate governance, shifting it from a reactive, manual, checkbox-based exercise to a proactive, real-time, data-driven function. Further studies will examine how CEO actions influence other factors, notably ESG and carbon emissions, to determine whether these strategies can also be optimized. 

Reference:

1. AI in Corporate Governance: Can Machines Recover Corporate Purpose?Swiss Finance Institute Research Paper No. 25-23, European Corporate Governance Institute – Finance Working Paper No. 1048/2025, B. Nikolov, N. Schuerhoff, S. Wagner, 16 Mar 2025.