New HEC Paris research suggests that shadow adoption of ChatGPT could benefit employees but not the firms

A study to be published and led by Professor David Restrepo Amariles of HEC Paris explores how consulting firms are integrating generative AI into their workflows and the challenges that arise when its adoption is left to individual discretion. The research highlights that while generative AI tools like ChatGPT can significantly improve content quality and perceived productivity, their unregulated use creates misalignment between managers and analysts, leading to undervalued effort and hidden AI adoption ("shadow adoption").

Higher Quality Output – But Effort is Undervalued

The study, conducted among 130 mid-level managers in a major consulting firm, found that:

• Content produced with the assistance of ChatGPT was evaluated more favourably by managers than the content produced without any AI generated content.
• However, when managers were aware that ChatGPT was used, they tended to undervalue the effort analysts put into the task.
• Analysts who used GPT without disclosing it were more likely to receive positive evaluations — suggesting that concealing AI use ("shadow adoption") might benefit them professionally.

Shadow Adoption – A Challenge for Managers

The research shows that managers struggle to identify when generative AI is used unless it is disclosed:

• When AI use was disclosed, 44% of managers still suspected that GPT had been used even when it hadn't — highlighting a trust gap.
• This creates an imbalance in accountability and evaluation, with analysts benefiting from undisclosed AI use while managers misjudge the effort involved.

Recommendations for Stronger AI Policies

The authors propose four key elements for developing comprehensive AI corporate policies:

1. Mandatory disclosure – Employees should inform managers when using AI to ensure quality and authenticity of AI-generated content.
2. Risk-sharing framework – To align the interests of managers and employees by clarifying responsibility and mitigating excessive risks.
3. Structured monitoring mechanism – To track AI usage, assess risks, and ensure compliance with quality and legal standards.
4. Incentive system – To recognise employee effort when working with AI, ensuring fairness and motivation without discouraging AI adoption.

"Our research demonstrates that AI adoption in consulting firms depends not only on technological capabilities, but also on managerial experience and structured policy frameworks,” said Professor Restrepo of HEC Paris. “Successful integration of AI tools like ChatGPT requires not only transparency, but also fair recognition of human effort and well-balanced incentives.”