In Cyber-Physical Manufacturing Systems (CPMS), numerous distributed control architectures were suggested to make different production entities active with respect to decision-making and control processes, so that they can process information, interact, and make control decisions in an autonomous and adaptive way. Nevertheless, developing Product-Driven Control (PDC) mechanisms that enable Smart Products (SPs) to make control decisions to cope with disturbances is still a complex, open-ended, and challenging problem. This article suggests a PDC approach that enables SPs to learn how to make control decisions to react to disturbances and maintain continuity of operations. The control mechanism involves an Analytic Hierarchy Process (AHP) augmented with Expert rules to cope with the limitations of standard AHP in dealing with dynamic problems. The mechanism is applied to a dispatching problem in an industrial scale assembly process. A multi-agent discrete event simulation model is used to create a set of normal and disturbed production scenarios. SP agents use context indicators and performance assessment to activate Expert rules and update AHP preferences and scales before making dispatching control decisions to react to disturbances. Data analytics tools are developed to help manufacturing system Experts define and fine-tune rules, based on rule firing statistics and corresponding context indicators and performance assessment acquired from simulation. Experimentations and results show competitive performance and highlight interesting research directions.