IA on the job: should people estimate before consulting AI?
When a team estimates a deadline or workload, consulting AI first seems a reasonable way to begin. Should managers nevertheless ask for an independent initial estimate before displaying the tool’s suggestion?
What the literature establishes
Amos Tversky and Daniel Kahneman (1974, Science) describe anchoring: an initial value can influence a numerical estimate even when it does not provide relevant information. In the situations studied, adjustments away from that value are not always sufficient to escape its influence. This research provides a reason to examine the first number presented, but it does not concern AI assistants.
The mechanism behind the first number
Fritz Strack and Thomas Mussweiler (1997, Journal of Personality and Social Psychology) provide evidence supporting a selective accessibility mechanism. Considering a suggested value can make knowledge consistent with that value more accessible and influence the subsequent judgement. Applied to AI, this mechanism suggests a hypothesis to test: an estimate accompanied by a justification might structure the search for supporting arguments rather than simply inform the user.
The premature conclusion
One might conclude that people should always produce their own estimate before consulting AI. Yet these studies do not demonstrate that this sequence improves workplace decisions: an initial personal estimate can also be wrong and become an anchor. The task is therefore to compare work sequences, not to treat initial independence as a guarantee of accuracy. (our executive and employee training programmes)
What the evidence cannot settle
The experiments cited concern numerical judgements in controlled settings, not collective planning in a Fribourg company. They cannot establish how professional expertise, data quality or discussion among colleagues would alter the effect of an AI suggestion. Nor is observing convergence between a human estimate and the tool’s estimate sufficient: that convergence may represent an appropriate correction.
A practical check in Fribourg
Within SHR’s “IA on the job” programme, a check could involve estimating processing times for completed, anonymised cases from Fribourg’s agri-food sector or bilingual industry, potentially seeking methodological support from the local university community. Randomly assign participants to two sequences—personal estimate followed by an AI suggestion, or AI suggestion from the outset—using the same cases, a fixed AI suggestion for each case and concealed actual processing times. Then compare the mean absolute error of final estimates against recorded times, separately for French- and German-language cases. This test would provide a local indicator of the value of estimating first, without affecting a production decision or claiming to establish a general rule. To go further: explore the AI on the Job training in Fribourg, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Fribourg
- ai on the job
- Fribourg
- research
- training fribourg