Organisations often talk about strategy as though they wanted to predict the future. But it is more effective if we first decide which futures deserve to be taken seriously.
For decades, foresight has used tools such as the futures cone to expand the space of possibilities. Possible, plausible, probable and preferable are not the same thing. A future can be possible without being probable. It can be preferable without being plausible. Clear enough, isn't it?
In every organisation, then, the struggle centres on plausibility. Ideas compete to be considered worthy of attention. Before they can even be approved, they must earn the credibility needed to be brought to the table as possibilities considered reasonable.
For centuries, this plausibility was a social and intellectual construction. It emerged from the interplay of experience, culture, intuition, authority, data, imagination and interpretative conflict. Managers, consultants, entrepreneurs, researchers and executives collectively helped define what appeared realistic and what seemed improbable.
The large-scale introduction of artificial intelligence into decision-making processes is changing this architecture. Because AI changes the criterion by which an organisation decides what is plausible.
The sophisticated median
By now, most people know that language models are tools for statistical synthesis. They work with enormous quantities of historical data. They identify regularities, recurrences, correlations and frequent configurations. They produce responses that maximise coherence, compatibility and continuity with the patterns in the training data.
For this reason, an LLM does not necessarily produce the most probable future, but it often produces the best available description of the past's centre of gravity.
Many organisations interpret AI output as a form of anticipation. In reality, they are often looking at a sophisticated representation of what the world has been up to that point.
It returns what appears most consistent with existing distributions. What emerges is not necessarily the future. It is the configuration that best represents the point of balance between configurations observed in the past and a possible future.
This function is extremely useful. It makes it possible to identify trends, consolidate knowledge, synthesise scattered information and reduce noise. But every median generates a centre that exerts a gravitational pull on the thinking of the human beings subject to it.
The cognitive baseline
The effect takes place in the room the model enters. When a team brings an AI-generated analysis to a meeting, that document quickly becomes a cognitive baseline. No one thinks ChatGPT or Claude is right one hundred per cent of the time. No one imagines the model is infallible. Yet almost everyone starts the discussion there, with what it said.
For decades, supporting a central position, an initial position, required reports, benchmarks, consultants, market analyses, data collection and systematic comparisons. Today, all it takes is a query to a model. Within seconds, a document appears that looks structured, reasonable, balanced and complete. AI thus produces a low-cost form of epistemic authority with a false sense of completeness.
From that moment, the burden of argument is redistributed. Hypotheses close to the model's output require little justification. Hypotheses that depart from it have to work harder; they have to be supported more forcefully. The effect is particularly significant in strategic planning.
The realm of what is defensibly plausible has effectively contracted. This is because the futures cone is being narrowed socially.
The normalisation of imagination
Many significant innovations arise from statistically marginal positions. The motor car was not the most plausible continuation of horse-drawn transport. The smartphone was not the linear continuation of the mobile phone. Netflix was not the natural evolution of video rental. Amazon was not the most obvious extension of an online bookshop.
The most transformative innovations tend initially to appear eccentric in relation to the dominant distribution. This does not mean that everything marginal is right. It does mean, however, that emergence and deviation are deeply connected. When plausibility is derived primarily from the frequency of past patterns, marginal positions receive less argumentative support.
A proposal can remain formally possible while at the same time becoming organisationally fragile. And this is the most interesting risk of the unreflective use of artificial intelligence in strategic processes.
The main problem is the normalisation of imagination. When plausibility is the result of statistical coherence, strategic exploration is reduced to nothing more than refining the present, even if it still seems to be much more. The organisation thinks it is innovating, but it innovates around the centre.






