Negotiating with AI: who sets the first number?
In Genève, preparing for a negotiation might involve private banking fees, a contract within an international organisation, supplier terms in luxury watchmaking or a logistics service in commodity trading. The managerial question is specific: should managers consult an AI estimate before or after setting their own target?
What the literature establishes
Amos Tversky and Daniel Kahneman (1974, Science) described anchoring: when an estimate starts from an initial value, subsequent adjustment may remain insufficient. Judgement then remains influenced by that starting point, even when it provides little relevant information. For managers, this suggests distinguishing two properties of a number proposed by AI: its informational value and its ability to steer the discussion.
The first number in a negotiation
Adam D. Galinsky and Thomas Mussweiler (2001, Journal of Personality and Social Psychology) showed in negotiation experiments that first offers could anchor final agreements. Their research also shows that this effect can be neutralised when recipients focus on information inconsistent with the anchor, particularly their own target or the other party’s alternatives. These findings concern offers made during a negotiation, not estimates generated by AI during preparation.
The premature conclusion
One might conclude that managers should always produce their own number before consulting AI. That would overlook the possibility that an internal estimate may itself be poorly grounded and become an anchor. The point is therefore not to favour the initial human judgement systematically, but to build a target from explicit criteria and examine what would justify revising it. (our executive and employee training programmes)
What the evidence cannot settle
These studies measure neither the effects of a generative assistant nor negotiation outcomes in the sectors mentioned in Genève. They do not establish that an AI-generated number has the same influence as an opposing party’s offer, or that it necessarily makes the resulting agreement worse. Simply comparing the initial target with the final agreement would also be insufficient: a shift may reflect anchoring, but it may also reflect the incorporation of useful information.
A practical check in Genève
As part of SHR’s « Manager à l’ère de l’intelligence artificielle » programme, an exercise could randomly assign participants to two ways of preparing for the same fictional negotiation, without confidential data. One group would first consult an AI estimate identical for everyone; the other would first record their target and criteria, then consult that same estimate. The measure would be the median absolute distance between each participant’s final target and the AI number, compared across groups, with written justifications retained. A closer alignment in the first group would indicate a possible effect of consultation order, without by itself demonstrating poorer preparation: that distinction should be the focus of the debrief. To go further: explore the Managing in the Age of Artificial Intelligence training in Geneva in the canton of Geneva, or browse our executive and employee training programmes in Switzerland.
In pictures: Managing in the Age of Artificial Intelligence in Geneva
- managing in the age of artificial intelligence
- Geneva
- research
- training geneva