Strategic Management Explorer
What AI Can’t Do: Two Thinking Skills Managers Need Now
By Rahul K. Mishra and Neha Gosain
The application of artificial intelligence (AI), machine learning, and robotics is automating processes and absorbing a growing share of routine analysis. Research published by Anil Doshi, J. Jason Bell, Emil Mirzayev, and Bart Vanneste in the Strategic Management Journal helps explain what human judgment is left to do.
Specifically, these SMJ authors found that a single AI evaluation of a strategic decision is often inconsistent and biased, but aggregating many AI evaluations, across different prompts, roles, or models produces judgments that closely resemble those of human experts. The lesson for managers is that good judgment comes from a deliberate process, not a single AI-generated answer.
The Practitioner “So What”
Never treat a single AI evaluation as signal. Aggregate across prompts, roles, and models before you act on the output — then bring the two skills the machine cannot supply: design thinking to frame what the customer actually needs, and critical thinking to test why the answer holds.
What Machines Now Do Well
Managers need to reimagine management; AI, robotics and digital twins can augment productivity, quality, delivery and customer value, but leadership now plays out inside a human-plus-machine organizational structure. We can reimagine a division of labor: machines handle synthesis and pattern-finding; humans own problem framing, innovation, and accountability for the choice.
What machines now do well is breadth at speed. In the Doshi et al. studies, large language models ranked 60 business models through pairwise comparisons, repeated across multiple models, ten assumed roles — journalist, strategy professor, and industry expert among them — and two prompting strategies. No human expert panel evaluates the same option set that many ways. The authors separate two sources of gain: a diversity effect from aggregating across different models, roles, and prompts, and a scaling effect from aggregating more evaluations of each kind.
Reimagining the division of labor starts there. Treat AI as a panel to be polled rather than an oracle to be consulted: ask the same question many ways and read the spread of answers as carefully as the answer itself. Machines supply volume, tireless consistency, and pattern-finding across unstructured evidence. What they do not supply is reliability in any single reading — in these studies, reversing the order in which two options were presented was often enough to change which one the model preferred.
What Stays Human, and Why
The human half comes from two time-tested tools: design thinking and critical thinking.
Design thinking is a human-centered approach to problem-solving, built on empathy, observation, and learning from users. There is little evidence that AI can empathize with the pain behind what users say, and its responses stay constrained by existing paradigms.
Research in Scientific Reports emphasizes that generative AI lacks the intrinsic curiosity and capacity to reveal novel mechanisms from scratch, often displaying overconfidence while failing to innovate beyond existing human paradigms.
Separately, research in the Strategic Management Journal on software products finds that generative AI mostly substitutes for or complements specific human tasks rather than replacing the judgment behind them.
Tim Brown, co-founder of IDEO, an international design company, affirms that a design thinker displays a remarkable talent for balancing technical, commercial and human considerations. Team members’ diverse backgrounds help in grasping and solving the customer’s problem as a truly human endeavor. Design thinking brings empathy for the end-user to center stage. No chatbot has yet demonstrated it can replicate this.
Roger Martin, professor emeritus at the Rotman School of Management, wrote the book “The Opposable Mind: How Successful Leaders Win Through Integrative Thinking.” In his book, he persuades leaders to take two opposing ideas, think through them and come up with a third and superior idea.
The habit working against this is instant gratification: reaching for an answer from the internet, or now from a chatbot, instead of stepping back to think the issue through.
Helen Lee Bouygues, the founder of the Reboot Foundation, argues that the opposite of critical thinking is selective thinking — quick reinforcement of what we already believe — and prescribes three habits: question assumptions, reason through logic, and diversify thought.
Japanese managers have a habit of understanding the crux of a problem by using a technique called the “five whys,” repeating “why” until the root cause, not its symptom, surfaces. The “five whys” is an analytical process created by Sakichi Toyoda, a Japanese industrialist, inventor and co-founder of Toyota Motor Company. Toyota uses this on the factory floor to find root causes fast.
The mechanism matters as much as the question. Any operator who sees a defect can stop the line, so the questioning happens at the machine within minutes rather than in a review meeting weeks later. Taiichi Ohno’s canonical chain runs: the machine stopped — why? A fuse blew from overload. Why? The bearing was not lubricated enough. Why? The lubrication pump was not circulating properly. Why? The pump shaft was worn. Why? There was no strainer, and metal filings got in. Stop at the first why and you replace a fuse; reach the fifth and you install a strainer. That is the same discipline the third check below asks of an AI recommendation.
What to Do on Monday
In an age when AI may take up every possible task, “thinking” remains a distinctly human domain. A single AI-generated recommendation should never be accepted at face value.
Practical Questions for Strategic Leaders
Before acting on one, a manager can run it through three quick checks:
- Does it reflect what the customer actually wants, not just what the data implies?
- What assumption would have to be false for it to fail?
- Why does it actually work? Ask yourself this several times over, until the answer stops changing.
Beyond any single recommendation, run an hour on a live decision rather than a case study: each participant names the assumption that would have to be wrong, and the group ranks which is cheapest to test first.
Finally, put AI inside the design sprint rather than at the head of it — let the team do its own observation, then use AI to cluster what it heard and widen the option set. One real decision and one short workshop beat any policy memo on AI use.
Human plus AI, questioned rather than accepted, is the way forward.
Rahul K. Mishra teaches Strategy at IILM. He has around 30 years of professional experience in management, including the last two decades in executive education, training senior executives on strategy and innovation. Neha Gosain is an assistant professor of finance and holds a PhD from IIT Delhi.
