AI cannot replace the peer challenge

What happens when a leader has more information than ever, but fewer people willing to challenge the story behind it?

That is the question I keep returning to as CEOs and founders adopt artificial intelligence. AI can widen analysis and produce options at remarkable speed. It cannot carry accountability after the decision. It cannot exercise moral judgement when the commercially attractive option conflicts with what the organisation believes is right. It cannot bear the human responsibility of carrying consequences when a decision affects employees, customers, families and communities.

I am not arguing against AI. I am arguing against giving it a job it cannot do.

[DENNIS TO VERIFY: Replace the following plausible composite scene with a confirmed Boardroom story before publication.] A fintech CEO in Sandton brings an AI-supported market-entry recommendation to a confidential peer meeting. The analysis is clear and the opportunity looks attractive. A peer asks, "What would make you walk away?" The CEO realises that the analysis has shown how to proceed, but nobody has defined the point at which proceeding would no longer be responsible.

The room has not produced a more sophisticated model. It has made the decision honest.

What can AI contribute to an executive decision?

AI can contribute analysis, options and speed. It is useful when a leadership team needs to compare scenarios, find patterns in a large information set or pressure-test the assumptions inside a plan.

A logistics founder in Durban may ask AI to model a fleet expansion. A manufacturing CEO in Pretoria may use it to compare capital investment scenarios while energy costs and supply reliability move. A professional services leader in Cape Town may ask it to test a new pricing structure. An operations leader in Windhoek may use it to map supply risks.

A decision is not simply an answer extracted from data. It is a commitment made under uncertainty by people who must live with the result. Its quality depends on what they notice, question and consider.

Why does access to AI not produce equal judgement?

The first concept is the judgement filter . AI can offer advice, but the quality of the decision still depends on the person selecting, interpreting and applying it.

A Harvard Business School Institute for Business in Global Society article by Steven Melendez, published on 29 September 2025, summarised a field experiment involving 640 Kenyan small-business entrepreneurs. About half received access to a GPT-4 AI business assistant through WhatsApp. The other half received written business guides developed by the International Labour Organization.

Overall, the researchers found no statistically significant performance difference between the groups. The closer analysis was more revealing. Initial high performers appeared to improve performance by more than 15%, while initial low performers experienced a reduction of nearly 10%.

The primary working paper by Otis, Clarke, Delecourt, Holtz and Koning links the uneven result to which pieces of advice people selected and implemented. The groups asked broadly similar questions and received similar advice. Their implementation choices differed.

That finding cuts against the promise that access to a powerful tool automatically closes a capability gap. A capable user may use AI to expand an already sound process. A less capable user may choose generic advice that does not fit the business. The tool amplifies the selection process that already exists.

A peer can ask why you selected that recommendation, what evidence you ignored and whether the decision serves the organisation or merely protects your preferred identity.

What does a peer advisory group add that AI cannot?

The second concept is the accountability mirror . A peer advisory group does not exist to supply more information. It makes the leader's reasoning visible to people who have no reporting line to protect and no internal politics to manage.

Andrew Feghali's PhD dissertation, completed at the University of San Diego in 2022, examined executive peer advisory groups. The groups in his study typically used confidential forums of six to 16 members meeting regularly, often monthly. Participants described joining to address the isolation of the top role, grow as leaders and expand their networks. They identified trust, confidentiality, vulnerability and learning from peers as central benefits.

The research does not prove that peer groups cause superior performance. It does show why room quality matters.

A logistics owner in Durban may challenge a Sandton fintech CEO's assumptions about scale. A family-business leader in KZN may question whether a proposed successor has been tested or merely favoured. A leader from Windhoek may see a dependency that a South African team has normalised. Difference is useful only when the room has enough trust to use it.

The value is not that every peer has the answer. It is that the group creates enough trust for a leader to say, "I may be defending this because it is mine."

Why is challenge more valuable than another answer?

The third concept is the decision boundary . Every significant decision needs a clear line between what can be delegated, what can be tested and what must remain a human leadership call.

When AI generates a recommendation, the leadership team should ask: who owns the decision, what would cause us to override the system, what evidence is missing and who bears the consequence if the recommendation is wrong?

A prompt library, AI policy or dashboard cannot create the trust required for a CEO to admit uncertainty or the courage required for a peer to question a powerful leader. More information can become a socially acceptable way to delay a difficult call.

That is why the standard AI solution falls short. Giving leaders an assistant, adding more data and expecting decision quality to rise evenly across the organisation ignores the judgement filter. If the decision process is weak, an AI layer may make weak reasoning faster, more confident and harder to challenge.

The answer is not intuition over evidence. It is evidence inside a human decision process. That is the discipline behind why smart leaders make bad decisions and how to stop them.

What practical steps can leaders take today?

1. Ask AI for options, not permission. Request competing courses of action, missing evidence, downside scenarios and the strongest argument against your preferred plan.

1. Separate facts, assumptions and values. AI can help organise facts and test assumptions. The leadership team must decide which values govern the trade-off.

1. Define the decision boundary. Record what AI may analyse, what an executive may approve and what requires the CEO, board or family council to decide. Make clear what cannot be delegated.

1. Ask what would change your mind. If no evidence could alter the preferred answer, you are not evaluating a decision. You are seeking permission to proceed.

1. Take the decision outside the hierarchy. Use a confidential peer group or advisory forum before the commitment becomes difficult to reverse. Challenge is most useful before consequences arrive.

1. Carry the result in public. Keep a decision record with the reasoning, assumptions, risks and review date. Examine the process honestly rather than blaming the tool or mistaking luck for judgement.

Can AI and the boardroom make better decisions together?

Yes, but only when each is given the right job. AI should accelerate analysis, expose patterns and widen the range of options. Leaders and their peers must supply context, judgement, courage and accountability.

The boardroom is valuable because consequential decisions require people who can question the question, understand the cost of being wrong and remain responsible after the meeting ends.

At The Leadership Boardroom, CEOs and founders meet in Pretoria, Sandton, Cape Town, Durban and Windhoek to work through decisions that cannot be solved by more information alone. The work is connected to the loneliest role in the room.

If your most important decisions are being made in isolation, the next investment may not be another tool. It may be a room where the truth can arrive before the consequences do. Learn more at The Leadership Boardroom.

Dennis Kriel is the founder of The Leadership Boardroom, a peer advisory community for CEOs and founders meeting in Pretoria, Sandton, Cape Town, Durban and Windhoek. A serial entrepreneur, educator and international keynote speaker, he writes on leadership, decision-making and building businesses that outlast their founders. Connect at denniskriel.com.

Sources

• Steven Melendez, Harvard Business School Institute for Business in Global Society, "AI won't make the call: Why human judgment still drives innovation", 29 September 2025: https://www.hbs.edu/bigs/artificial-intelligence-human-jugment-drives-innovation
• Nicholas G. Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning, "The Uneven Impact of Generative AI on Entrepreneurial Performance: Evidence from a Field Experiment in Kenya", Harvard Business School Working Paper 24-042: https://www.hbs.edu/ris/download.aspx?name=24-042.pdf
• Andrew Feghali, University of San Diego, "Executive Peer Advisory Groups: Who They Are? What Are Their Benefits? Why Do Members Join and Stay?", PhD dissertation, 22 May 2022: https://digital.sandiego.edu/dissertations/930/

About the author: Dennis Kriel is the founder of The Leadership Boardroom, a peer advisory community for CEOs and founders meeting in Pretoria, Sandton, Cape Town, Durban and Windhoek. A serial entrepreneur, educator and international keynote speaker, he writes on leadership, decision-making and building businesses that outlast their founders. Connect at denniskriel.com.