On 29 May 2020, Émile Servan-Schreiber shared his Les Echos Weekend interview with Pierre de Gasquet, appealing to free up intelligence as lockdown eased. His argument addressed a business problem: when familiar models lose their footing, how can an organisation gather observations from people facing changed conditions?
Forecasting when yesterday’s models lose their footing
Servan-Schreiber argued that the pandemic had disrupted assumptions behind familiar forecasting models. His proposed response was to collect observations from people on the ground. In his view, a distributed workforce could help an organisation understand changed conditions when historical data provided a poor guide.
His recommendations were concrete: seek perspectives beyond the top of the hierarchy, allow people to think independently before discussing ideas, and invite employees to contribute beyond their specialised roles. He also cautioned that participatory management depended on people’s willingness and autonomy. Participation was a design problem, rather than a guarantee.
Diversity needs a method
Different ways of seeing a problem were central to his argument. Digital tools, he proposed, could gather those perspectives across an organisation. Remote work could create room for individual reflection, while constant group brainstorming could interrupt it. The recommendation was to protect the work each person does before a group discussion, then make the separate contributions available for comparison. In his account, the organisation of participation mattered as much as the invitation to participate.
Related primary research helps delimit that argument. Lu Hong and Scott Page’s 2004 study modelled conditions under which a functionally diverse set of problem-solvers can outperform a group selected for high individual ability. Their result depends on the model’s conditions. It is not evidence that every diverse group will outperform every expert on any task.
For a forecasting team, this suggests a useful sequence: preserve separate observations, combine judgements and later compare them with outcomes. That is an editorial implication of the interview’s advice, rather than a protocol tested by it. The journal’s article on correlated forecasts explores the related risk of counting repeated information as many independent signals.
