The context of our study is industrial reliability, where lifetime data are usually censored and in small number. Background information is available from experts. Our prior subjective knowledge is only about the lifetime of an industrial component and not about the parameters of a Weibull distribution which represents this lifetime. We propose to focus the discussion between the experts and the industrial analyst about the size of {\it virtual} data representing the variability of the expert opinion. Indeed, this size is one of the scarce indicators that both can understand. The prior calibration is made easy, and some methods and indicators including a default calibration method are proposed to help the Bayesian analyst (they can be extended to inferences on other distributions than Weibull). Besides, the posterior computation by importance sampling is simple and satisfying. Finally, through a real example, the flexibility of the elicitation is illustrated.