AI assistants: who should give the first estimate in Basel?

In Basel, preparing a sample transfer, a Rhine logistics operation or an international R&D milestone often involves estimating a timeframe that remains uncertain. The precise managerial question is this: should the assistant’s estimate appear before the professional has formulated their own?

What the literature establishes

Amos Tversky and Daniel Kahneman (1974, Science) describe anchoring: an initial value can influence a numerical judgement even when it provides little relevant information. The starting point is therefore not necessarily just a reference that people freely adjust. For an AI assistant, this raises a design hypothesis rather than a directly established finding: displaying a timeframe could influence the estimate that follows.

Why adjustment may remain insufficient

Nicholas Epley and Thomas Gilovich (2006, Psychological Science) show that, for self-generated anchors, adjustment can stop as soon as a value seems plausible. Revising a starting point therefore does not necessarily mean moving far enough away from it. This mechanism does not automatically transfer to a number supplied by AI, but it challenges the assumption that simply allowing edits would neutralise any initial influence.

The premature conclusion

One might conclude that the human should always estimate first. Yet these studies demonstrate neither that human estimates would be better nor that an automated suggestion would systematically worsen the decision. A well-informed initial estimate can be useful; the challenge is to distinguish its informational contribution from its pull on subsequent judgement. (our executive and employee training programmes)

What the evidence cannot settle

These studies address neither generative assistants nor planning in the life sciences. They cannot establish the optimal display sequence for pharmaceutical teams working with international partners. Obtaining an independent estimate also has a cost: this step deserves testing where inaccurate time estimates have significant consequences, rather than being adopted as a universal rule.

A practical check in Bâle

As part of SHR’s programme « Créer ses assistants et agents IA sans coder », a Basel team could configure two workflows using anonymised retrospective cases involving sample transfers or Rhine logistics: display the assistant’s estimate immediately, or record a human estimate before displaying it. Participants would be randomly assigned to the workflows, with the same information available at the time of forecasting and no access to actual completion times. The primary measure would be the median absolute difference between the final estimated duration and the actual duration; time spent estimating would be recorded separately. This test would not, on its own, prove an anchoring mechanism, but it would establish whether the display sequence actually improves forecasting in this setting. To go further: explore the Building AI Assistants and Agents Without Coding training in Basel in the canton of Basel-Stadt, or browse our executive and employee training programmes in Switzerland.

In pictures: Building AI Assistants and Agents Without Coding in Basel

• building ai assistants and agents without coding
• Basel
• research
• training basel