Three Mechanisms of Innovation Search

Recombination as a mechanism for innovation can be traced back to at least Schumpeter (1934). More recently various organizational and strategy scholars have described the concept of recombination (Fleming, 2001; Kogut and Zander, 1992) with Fleming (2001) providing a seminal treatment of the recombinant process as it pertains to innovation search. As Fleming notes, inventions are fundamentally composed of combinations of prior existing components into new syntheses or the recombining of existing combinations (Fleming, 2001; Fleming and Sorenson, 2004; Fleming, Mingo, and Chen, 2007; Nasiriyar, Nesta, and Dibiaggio, 2010). Such recombinations may result in entirely new products and services or the application of existing products to new markets and uses (see Fleming and Sorenson, 2004, for famous examples). Pure recombination may however lead to the generation of far more combinations than can be meaningfully evaluated. To avoid a combinatorial explosion or ‘complexity catastrophe’ (Fleming and Sorenson, 2001) some decision rule needs to be invoked to reduce the set of combinations to a feasible number. Yayavaram and Ahuja (2008) suggest that ‘coupling’ may be one mechanism that organizations use to reduce the number of combinatorial choices to a meaningful number. Thus, recombination may work through the combining of ‘coupled’ groups of elements rather than individual elements.

Cognitive search processes work through the exercise of a causal reasoning logic (Fleming and Sorenson, 2004; Gavetti and Levinthal, 2000), the most common prototype of which is the scientific method. The inventor begins with an abstract model of the phenomenon of interest (Holland et al., 1986; Gavetti and Levinthal, 2000) and tries to understand the cause–effect mechanisms that underlie its behavior. To understand these cause–effect mechanisms the inventor may resort to one of two common processes, deduction or induction (Novelli, 2010). Deduction entails the identification of causes from systematic analysis using the tenets of established principles and general laws. These principles depict an architecture of relationships, a big picture of the terrain that the inventor can explore using causal reasoning and prior to or without observation of the phenomenon. In induction the researcher attempts to generalize backward from a set of observations to discover a common pattern that he or she uses as the basis of a general model of the phenomenon. Having understood the underlying causal effects the inventor can now design new products and processes that incorporate elements leading to desirable outcomes and eliminate elements that lead to undesirable effects.

When the process or product sought to be innovated upon is complex, embodying many simultaneous interactions, formulating a cognitive model of the process to engage in cognitive search may not be practical. In such situations experiential search processes may still provide a mechanism for innovation. Rather than rely upon a reasoned logic to explain the effects of a proposed change (impractical given the relative complexity of the change), in experiential search the inventor actually tries out the proposed change and then decides whether or not to accept the resultant product or process. Experiential search processes thus lead to innovation through a variation-selection-retention cycle. The decision maker varies an existing product or process and receives feedback from the environment on the performance of the entity after the mutation. If this feedback suggests that performance is improved following the mutation, the mutation is retained and the new version of the product or process is adopted. Subsequent variations continue the cycle of change leading potentially to further new products or processes.

Experiential search processes commonly differ from cognitive search processes on three key dimensions: the mode of evaluation, the range of alternatives considered, and the location of the search (Gavetti and Levinthal, 2000). Cognitive processes, since they can operate through the application of logic, may sometimes not require online evaluation. Logical analysis may suffice to provide an answer to whether some proposed change is a good idea or not. However, for experiential search processes putting the application into practice is necessary. Hence evaluation is, of necessity, on-line (Gavetti and Levinthal, 2000). Given the need to evaluate the effect of a proposed change experimentally, the number of alternatives that can be considered is also smaller than with cognitive search. Finally, experiential search is likely to be more incremental or local than cognitive search. Since the logic of experiential search involves assessing the effects of a change before accepting the change, it militates against simultaneous experimentation with multiple changes because the performance effects of the several changes would be confounded making it difficult to isolate what changes were beneficial and should be retained and what others were harmful (Sorenson, 2003).

Experiential search is also the basis of analogical reasoning (Gick and Holoyak, 1980; Gavetti, Levinthal, and Rivkin, 2005). Analogical reasoning deals with using experiential knowledge to cope with novel environments. It involves mapping from a source context of prior experience to a new target context (Gick and Holoyak, 1980). When the organization faces unfamiliar problems, analogizing managers choose a subset of the problem characteristics they believe distinguishes similar problems from different ones. Then, they transfer from the matching problem high-level policies or principles that guide search in the novel context (Gavetti, Levinthal, and Rivkin, 2005).