In his short story “The Evitable Conflict,” published 75 years ago, sci-fi writer Isaac Asimov described how machines might run entire industries. Today, there is mounting evidence that AI can outperform humans on many individual cognitive tasks. An experiment by a team from Cambridge University suggests that large language models (LLMs) can outperform humans in most tasks including product design, cost control, and market intelligence.

Impressive though such results are, the studies we are aware of have all been conducted under artificial laboratory conditions. Extrapolating these to real-world conditions is far from straightforward. Executives in the wild face unframed and shifting challenges, often with insufficient or inaccurate data.

We wanted to understand: What happens if we take AI out of the lab into a real company? Over the past year we joined a series of executive team meetings at Giesswein, an $85 million revenue company based in Austria, that sells organic, eco-friendly wool sneakers. Our idea was to experiment with different ways of integrating AI into their executive meetings to understand what works and how.

AI in the Boardroom

Our engagement with Giesswein started in October 2023. We designed three types of interventions and conducted at least two different variations of each to test for replicability. After each intervention, we conducted follow-up interviews with the two brothers who run the family firm to gather their perceptions of effectiveness.

The interventions happened during a period when the executive team decided to make several big strategic moves, allowing us to see AI at work when the future of the firm was being forged. The firm had decided to outsource production entirely, closing down their longstanding manufacturing facilities in Austria and transforming them into a logistics hub. They also decided to sell a sewing factory in Slovakia and to enter the U.S. market.

Our first type of intervention was to over the course of several meetings simply feed the agenda of the executive team into ChatGPT 4.0 asking for suggestions on which questions and issues to discuss. The output was shared with the team during the meeting as each of the agenda points came up. We also developed prompts to create more specific recommendations. For example, when the question of outsourcing was on the agenda we asked for pro and con arguments. We knew that the company was already outsourcing some production to China, so we asked ChatGPT to explore what the company needed to keep in mind if it wanted to outsource all production. The output was then shared during the executive meeting.