To assess its impact on innovation, I focus on an important aspect of corporate governance that has been challenging to empirically identify — the CEO performance evaluation horizon. Because trial and error is a time-consuming process but is crucial to produce important innovations, evaluation horizon is an important determinant of innovation, an insight provided by Manso (2011). Taking advantage of the 2011 SEC rule change that requires all public US companies to vote on the “say on pay” frequency, I apply sharp regression discontinuity design (RDD) to show that firms with long term evaluation generate low value innovation at first but more valuable innovation in the long run. Furthermore, my evidence suggests that the underlying mechanism is that long evaluation horizons allow costly “explorations” in a broad spectrum of technological fields in the beginning. Later on, the firm benefits from the exploration by “exploiting” the field that was identified as the most promising during exploration. The patent “truncation problems” that traditionally limit investigations of recent (here 2011 onward) innovation are ameliorated by the collection and usage of patent application data. The differentiation of exploration and exploitation is enabled by two novel measures of exploration.