IA on the job: does a good outcome prove a good decision?
A product launch succeeds, a procurement decision fails: an assistant can quickly produce a convincing review of either. The managerial question is more specific: how can AI be used to evaluate a decision without confusing the quality of the original reasoning with that of the eventual outcome?
What the literature establishes
Jonathan Baron and John C. Hershey (1988, Journal of Personality and Social Psychology) demonstrated experimentally that a decision’s outcome influences assessments of its quality, even when the information describing the decision remains identical. A favourable outcome can therefore make unchanged reasoning appear better. For managers, this suggests distinguishing between two objects of evaluation: the process followed and its consequences.
The past becomes more predictable in hindsight
Baruch Fischhoff (1975, Journal of Experimental Psychology: Human Perception and Performance) showed that knowing an outcome changes retrospective judgements of its likelihood. This hindsight bias differs from outcome bias: it changes what appears to have been predictable, rather than merely the assessment of the decision. A review prepared with AI therefore deserves scrutiny for what it portrays as obvious at the time.
The premature conclusion
One might conclude that outcomes should be excluded from every lessons-learned review; that would remove information essential to learning. The useful distinction is to assess the decision first against the information available at the time, and then consider what the outcome gives grounds to revise. Within the “IA on the job” programme, this separation can become a working instruction without being presented as a proven remedy. (our executive and employee training programmes)
What these studies do not demonstrate
These experiments concern human judgement, not reviews produced by generative models. They establish neither that AI systematically amplifies these biases nor that a particular instruction eliminates them. A fluent review therefore cannot, on its own, reveal whether a team has evaluated the decision better or merely told a better story about its outcome.
A practical check in Vevey
In Vevey, with its global food industry, group headquarters and Riviera SMEs, an “IA on the job” workshop could start with an anonymised procurement decision whose outcome is already known. Using a file limited to the information available when the choice was made, two comparable groups would prepare their evaluations with the same assistant and assessment grid, with only one group also receiving the final outcome. The measure would be the difference between their decision-quality scores, together with a record of the justifications that invoke that outcome. This difference would not isolate an effect specific to AI, but it would make the evaluation process’s sensitivity to retrospective information observable. To go further: explore the AI on the Job training in Vevey in the canton of Vaud, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Vevey
- ai on the job
- Vevey
- research
- training vevey