When AI enters the factory, what makes the investment pay?

A factory gains little from an accurate prediction if nobody can act on it. AI may help manufacturers detect defects, maintain equipment and coordinate deliveries, but the economic return depends on integration, reliable data and the costs of changing production. This article separates reported adoption from measured performance, using European manufacturing statistics, research on US factories and a survey of Japanese workers. The evidence includes initial productivity losses during adjustment, possible later gains and concerns about employment. A sector profile keeps the different populations and years visible, while an original chart distinguishes general AI use from specific production uses. The assessment follows what happens to quality, costs, suppliers and working conditions. It asks how gains can reach workers and customers, and what evidence would justify an investment beyond a successful demonstration.