In his keynote at the PCD London Conference 2026, Gregory Perdon argued that one of AI’s valuable applications could be helping leadership teams visualise, test and improve the quality of their own decision-making.
The uncomfortable truth is that many senior teams continue to make consequential decisions largely without much systematic feedback on their past decision-making processes thus creating an opportunity for improvement.
When we think of elite teams, we often default to Navy SEALs, commercial pilots, surgical teams, or perhaps the ExCo of a fast-growing technology company. They operate in high-stakes, uncertain conditions, where decisions can have serious consequences for the livelihoods of others. But there is an important difference between these groups.
The military uses after-action reviews as feedback, aviation teams use crew resource management and surgical teams use checklists to reduce avoidable error. In each case, feedback is embedded in the process. The Boards/ExCos of SMEs and investment committees do not always work that way.
That was the central challenge posed by Gregory Perdon, Investment Manager at Charles Stanley, in his keynote at the PCD London Conference 2026. If leadership teams want to improve the way they make decisions over time, they require more than intelligence, experience and confidence. They need feedback loops that make the quality of their reasoning visible so that it can, in turn, be more easily analysed objectively.
Gregory’s argument was not that AI is a solve-all. Nor was it that boards should outsource judgement to machines. The point was more practical: used carefully, in-house AI tools can help committees examine how their discussions unfold, where evidence is strong or weak, whether alternatives were properly considered, and whether uncertainty was communicated clearly. In other words, AI can act as a mirror to promote reflection, not a judge to make a final decision.
Leadership teams often grade themselves by the seriousness of the discussion. But seriousness is not the same as quality. A room can sound intelligent, energetic and decisive whilst still failing to test the story, search for alternatives, invite dissent or quantify uncertainty.
The keynote focused on four recurring traps in high-stakes group decision-making. The first was the power of narrative. Gregory used Simon Sinek’s well-known “Start with Why” argument as an example. The Apple story is memorable and persuasive, but it does not meet a serious causal standard. The issue is not whether purpose matters. It may. The issue is whether a persuasive story about one successful company proves that purpose caused the success.
Apple may have had a clear “why”, but so did many companies that failed. WeWork wanted to “elevate consciousness”. Theranos wanted to “revolutionise lab testing”. Mission alone is not evidence of causality.
The broader point is important for any boardroom or investment committee. A good story can make a weak claim feel stronger than it is. Presenters can move quickly between fact, opinion, forecast and anecdote, often without the audience noticing the transition. Confidence, meanwhile, tells us only that the speaker has a coherent story in mind. It does not tell us that the claims are necessarily true.
Gregory’s warning was blunt: one of the most dangerous phrases in business is not necessarily “this time is different”, but “when n=1”. A single observation can be useful, but it is not the same as representative evidence.
The second trap was first-idea fixation. Drawing on the work of Paul Nutt, Gregory noted that many poor organisational decisions begin with a rush to judgement. A proposal is placed in front of a committee and the discussion quickly becomes binary: should we proceed or not?
That can feel rigorous because the debate may be intense and the actors around the table credible. But often this feeling is misplaced. The first proposal becomes the anchor around which the discussion revolves. The hard part is not simply arguing for or against the first idea. The hard part is generating serious alternatives to compete with it.
This matters in investment committees as much as corporate boardrooms. A portfolio manager proposes a position and the room debates whether to buy. But perhaps the better question is: what are the genuinely distinct alternatives? Should we express the view differently? Should we wait? Should we buy something else that gives similar exposure with better asymmetry?
A decision is often only as good as the best alternative considered. Better debate helps, but more alternatives increase optionality.
The third trap was the suppression of dissent. Gregory used the Challenger disaster as the canonical example of what happens when challenge exists but is not properly heard. Engineers had raised concerns about the O-rings and cold weather, yet hierarchy, time pressure and institutional momentum muted the challenge when it mattered most.
Many committee members will recognise a milder version of this. There are moments when many people in the room privately disagree with an influential voice, but objection is softened, delayed, made performative or left unsaid. Dissent, done badly, becomes theatre. Done well, it can serve as an early warning system.
But disagreement alone is not enough. What matters is whether the disagreement is specific, timely and attached to a real decision consequence. The practical language matters: “What happens if we are wrong?” “Have we tested for this?” “What would an independent expert say?” “What is the counterfactual?” These are not aggressive questions. They are mechanisms for detecting errors in thinking.
The fourth trap was vague language around uncertainty. Gregory closed this section with the Bay of Pigs case study. President Kennedy was reportedly told the plan had a “fair chance” of success. The phrase concealed a much less favourable assessment: roughly a 25% chance that the plan would succeed.
The lesson is simple: language such as “likely”, “possible”, “confident” or “fair chance” can mean very different things to different people. A committee that says “highly likely” without defining what that means may think it has communicated clearly, when in fact it has only transferred ambiguity between parties.
Drawing upon the work of Tetlock & Gardner, Gregory argued that committees should get more comfortable using probabilities, ranges and scenarios. Absolutes may be psychologically comfortable, but they are epistemically dangerous. The decisive leader can look strong whilst quietly stripping out uncertainty.
The practical proposal was not complicated, but it does require a little courage. Where appropriate, record decision-making meetings. Produce word-for-word transcripts. Use an approved in-house LLM, with proper controls and human oversight, to analyse how the discussion actually unfolded. This should be done with agreed permissions, careful treatment of confidential material and a clear understanding that the model’s output is advisory, not determinative.
A useful first review might ask four questions. Which claims required evidence? Which alternatives were seriously considered? Where did challenge appear, disappear or become softened? Where was uncertainty expressed vaguely rather than numerically?
The purpose is not to let the machine decide. The purpose is to create a feedback loop. Gregory compared this to credit quality. A credit rating is not directly measurable. It is inferred from observable inputs: leverage, interest cover, revenue trends, cash flow, covenants and so on. Decision quality is similar. We cannot see it directly, but we can look at some of the underlying behaviours that contribute to it: the ratio of evidence to narrative, the presence of genuine alternatives, the quality of dissent, and the clarity with which uncertainty is expressed.
Good decision processes do not guarantee good outcomes. Markets, competitors and geopolitics will still surprise us. A good committee can be unlucky, just as a poor committee can occasionally be lucky. The aim is not certainty, but rather improved calibration.
AI can help make the language which expresses those behaviours, visible. It can identify check-worthy claims, distinguish evidence from assertion, flag areas where alternatives were not explored, detect vague uncertain language and generate counterarguments or requests for better evidence. Used well, it gives the team a baseline. Not a perfect measurement of judgement, but a more disciplined way of examining the ingredients that feed into judgement.
But of course there are risks. Some committee members may feel threatened by recording. Discussion may become more performative. Confidentiality, legal and governance considerations all matter. And LLMs can hallucinate, miss nuance, misread tone or disguise weak evidence. None of these concerns should be dismissed. But nor should they become excuses for inaction. Humans hallucinate too. LLMs are probabilistic models and since when do all forecasts come true?
Gregory noted that he had applied the same discipline to his own keynote. He ran the speech through an LLM, asked it to probe the claims line by line, and kept the audit trail. The objective was not to seek consensus nor modal thinking but rather to increase defensibility.
The final message was deliberately practical. Boards do not necessarily need military rituals nor aviation checklists but rather could benefit from systematic feedback on their decision processes. Because according to Gregory, “systematic feedback can be gold dust for edge-seekers or fool’s gold for theatregoers”.
