Forecasting lead times with AI: does a precise date mean a sound forecast?
AI can turn scattered information into an apparently coherent schedule without removing production uncertainties. The management question is specific: how can a lead-time estimate be used without mistaking precision in its presentation for operational certainty?
What the literature establishes
Don Moore and Paul Healy (2008, Psychological Review) distinguish several forms of overconfidence: overestimating one's performance, placing oneself too favourably relative to others, and expressing excessive precision in one's beliefs. The last form is particularly relevant when a lead time is presented as narrowly predictable. A manager may remain cautious about the team's capabilities while underestimating uncertainty in the schedule.
An interval does not automatically make an estimate cautious
Jack Soll and Joshua Klayman (2004, Journal of Experimental Psychology: Learning, Memory, and Cognition) examine overconfidence in interval estimates and show that the way estimates are elicited influences their calibration. Asking for a range rather than a date therefore does not guarantee that uncertainty is represented accurately. For an AI-assisted forecast, this suggests examining how the bounds were constructed rather than treating their mere presence as a safeguard.
The hasty conclusion
One might conclude that announced lead times should always be widened. This would confuse a well-calibrated forecast with a safety margin: the former describes uncertainty, while the latter is a management choice. A very wide range may encompass the outcomes without providing useful support for coordination. (our executive and employee training programmes)
What the evidence does not establish
These studies concern human judgement, not forecasts generated by AI for a workshop in Neuchâtel. They establish neither that a particular tool produces overly narrow lead-time estimates nor that its intervals correspond to reliable probabilities. A range generated in natural language must be distinguished from an interval whose coverage has been evaluated against observed outcomes.
A practical check in Neuchâtel
Within SHR — Swiss Human Resources' programme « Manager à l’ère de l’intelligence artificielle », an exercise applicable in Neuchâtel would track the time taken by a recurring batch-validation operation in microtechnology, watchmaking or precision microelectronics. Before each operation, the manager would record the range adopted with AI assistance, its intended coverage level, its assumptions and the recording date, preserving earlier versions. The review would compare the proportion of actual durations falling within the bounds with the stated coverage level, while also reporting the average width of the ranges and the number of observations. This check would not prove AI's superiority; it would make the quality of time commitments developed with AI open to verification. To go further: explore the Managing in the Age of Artificial Intelligence training in Neuchâtel in the canton of Neuchâtel, or browse our executive and employee training programmes in Switzerland.
In pictures: Managing in the Age of Artificial Intelligence in Neuchâtel
- managing in the age of artificial intelligence
- Neuchâtel
- research
- training neuchâtel