A general definition of an experiment is:" A deliberate comparison of outcomes from a varied but repeatable set of conditions, with an effort to explain different outcomes by differences in conditions." This definition includes controlled experiments (in which possible causal conditions are deliberately manipulated), and natural experiments (in which they are measured but not deliberately altered). The definition excludes purely descriptive investigations such as surveys or satellite photographs in which the results are solely tabulated or displayed. For example, a control chart by itself is not an experiment, although control charts can provide data for natural experiments.

Experiments are a key mechanism for industrial learning, and are therefore an important managerial lever to accelerate learning curves (Adler and Clark 1991; Dutton and Thomas 1994; Lapré et al. 2000). The learning curve phenomenon has been observed frequently for measures of organizational performance such as quality and productivity. The rate of improvement is called the" learning rate.” Learning rates show tremendous variation across industries, organizations, and organizational units (Dutton and Thomas 1984; Lapré and Van Wassenhove 2001). Bohn (1994) and Lapré (2011) discuss the model inside the learning curve. Both experience and deliberate activities can be sources for learning; learning can yield better organizational knowledge, which in turn can lead to changed behavior, and subsequently to improved performance. None of these steps are automatic. Dutton and Thomas (1984) call learning from experience “autonomous learning" and learning from deliberate activities" induced learning." The typical examples of deliberate activities are quality improvement projects and productivity improvement projects. Such projects often rely on a series of experiments. Other induced learning methods at the level of the firm are deliberate knowledge transfers from outside the organization, and the training of workers-but these are available only for knowledge that already exists and is accessible. Even when firms transfer knowledge from the outside, some adaptation to local circumstances and thus experimentation-is almost always required (Leonard-Barton 1988). Hence, sound management of experimentation is important in order to attain the effective management of the learning rate. In an extreme example, experimentation can be the sole driver of a learning curve. Lapré and Van Wassenhove (2001) studied productivity learning curves at four production lines at Bekaert, the world's largest independent producer of steel wire. One production line was run as a learning laboratory in a factory. The productivity learning curve was explained by the cumulative number of productivity improvement projects, which consisted of a series of natural and controlled experiments. The other three production lines were set up to replicate the induced learning. Interestingly, the other three lines struggled with learning from experimentation and relied on autonomous learning instead. Even within the same organization, it can be difficult to manage the required scientific understanding and human creativity (we will later refer to this as the" value of the underlying ideas"). Experiments are used in a variety of settings. Generally, the experiments themselves are embedded into broader processes for deliberate learning, such as line start-ups, quality programs, product development, or market research. Examples of situations dealt with by experimentation include: