AI output being correct doesn’t automatically equate to it being trusted. Some pharma companies that have been drawing on AI powered adverse event processing for a while now are reporting a frustrating disconnect between the demonstrated accuracy scores and the way PV teams are responding to them. Too often, reviewers are found to be reopening cases already signed off by the system, when there is no logical reason to do so. Audits confirm that this is happening, even though there has been nothing formal from a quality control perspective to require it. So, what’s going on and what can be done about it – to ensure companies derive the expected benefits from the technology’s efficiency? Accuracy isn’t the whole story.