The study
Who: Fabrizio Dell'Acqua and colleagues, with Boston Consulting Group.
What: Navigating the jagged technological frontier, published in a peer-reviewed management journal in March 2026.
Type: A pre-registered experiment. Groups with and without AI did the same tasks.
People: 758 BCG consultants. The experiment used GPT-4 in 2023. Today's AI tools are different.
What the researchers found
On tasks the AI handled well, consultants using GPT-4 completed 12.2% more tasks and worked 25.1% faster.
On one business case designed to be outside the AI's ability, 373 consultants took part. 84.5% of those without AI reached the correct recommendation. In the two groups using AI, 60% and 70.6% did.
Graders also rated recommendations made with AI as more persuasive, including wrong ones. A short overview of how to write prompts did not prevent the error.
What this may mean for managers
This is our interpretation, not the researchers' finding.
The same tool can help on one task and mislead on another. A convincing answer is not proof that it is right. Teams need to know which kinds of work their AI handles well, and to check more carefully when a task may be outside that range.
What this study did not test
- It used one model, GPT-4, in 2023. It does not describe the accuracy of today's AI in general.
- The task outside the AI's ability was one designed business case.
Read next
- Questions to ask before acting on an AI answer
- There are two ways to get AI wrong
- How AI answers become business results
Sources
- Dell'Acqua and others (2026). Navigating the jagged technological frontier. Organization Science 37(2):403–423. PDF. Source checked 4 October 2026.