The AI Mirror: Can Generative Models Evaluate Strategy Better Than Humans?
For decades, the “holy grail” of business leadership has been the ability to accurately predict which strategic ideas will succeed and which will fail. While generating creative ideas is often seen as the spark of genius, the cold, hard reality of business is that evaluation, the rigorous assessment of an idea’s potential, is where value is truly created or destroyed. Strategic decisions are notoriously difficult to judge because they are uncertain, often irreversible, and lack immediate feedback loops.
In a groundbreaking study published in the Strategic Management Journal, “Generative artificial intelligence and evaluating strategic decisions,” researchers Anil R. Doshi, J. Jason Bell, Emil Mirzayev, and Bart S. Vanneste (2025) explore whether Large Language Models (LLMs) can bridge the “evaluation gap” that plagues human decision-makers. Their findings suggest that while AI is not a magic bullet, it offers a powerful new way to stress-test the logic of our most critical choices.
While individual AI outputs can be inconsistent, aggregated AI evaluations effectively filter out “noise” to match human expert judgment. Leaders can now use GenAI as a low-cost “Red Team” to pressure-test strategic assumptions and move from intuitive gut feelings to augmented, evidence-based decision architectures.
The Research: Benchmarking AI Against the Experts
To understand how GenAI handles strategic complexity, the researchers conducted two distinct studies involving 60 business models each. The first study utilized business models generated entirely by AI, while the second examined models submitted to a real-world competition.
The goal was to see if LLMs could rank these strategic alternatives in a way that aligned with the judgments of seasoned human experts. By testing multiple LLMs, diverse assumed roles (such as “Venture Capitalist” or “Strategy Consultant”), and various prompting techniques, the researchers uncovered a nuanced picture of AI’s current capabilities in the boardroom.
Key Findings: The Strengths and Scars of AI Evaluation
Practical Implications: What Strategic Leaders Should Do
Strategic Action Plan
- Implement AI Red Teaming: Firms can leverage GenAI as a low-cost, high-speed “Devil’s Advocate”. Before finalizing a major pivot or acquisition, managers should use LLMs to critique the plan’s underlying assumptions. By explicitly prompting the AI to find reasons why a strategy might fail, leaders can move beyond groupthink and uncover structural weaknesses before capital is committed.
- Moving to Augmented Evaluation: The role of the executive is shifting from the “Decision Maker” to the “Decision Architect”. Instead of relying solely on a gut feeling, leaders should design evaluation processes that combine human creativity with AI rigor. Use human teams to generate a broad range of creative options, and use aggregated AI evaluations to vet those options for logical consistency and market viability.
- The Importance of Prompt Discipline: The study demonstrates that the quality of AI feedback depends heavily on the role the LLM is asked to assume and the specific criteria it is given to evaluate. Managers must become proficient in “Decision Prompting”—structuring the AI’s task so that it focuses on the most critical strategic variables rather than surface-level details.
Conclusion: Prediction as a Service
Doshi, Bell, Mirzayev, and Vanneste (2025) conclude that GenAI’s greatest value to strategy lies in its ability to provide predictions at scale. While the “final call” remains a human responsibility, AI provides the data-driven map that makes that call more likely to be correct. By using AI to mirror our logic back to us and aggregating those reflections, we can strip away the biases and inconsistencies that lead to strategic failure. In the future, the most competitive firms won’t be those with the smartest humans, but those that best combine human intuition with the tireless, aggregated predictive power of generative AI.
Leave a Reply