**6. Discussion**

A natural comparison point for our survey is ACW, which also looked at innovation sourcing, although only for the most important projects and not for any secondary projects or any firm-level usage (at least as reported). They find that customers and suppliers are much more important sources of innovation projects (27% and 14% of most-important innovations<sup>3</sup> ), whereas eight years later we find 2% and 8%, respectively (**Table 3**). This drop over time aligns with the drop in the importance that firms report five years ago until today.

We observe a bigger difference with ACW in the share of most-important innovations that came from internal sources. ACW find that 51% came from an internal source, whereas we find that 77% do. This difference is even larger if we account for the rise of external sourcing that we observe in our data. There are many reasons why our results could differ from ACW. One important difference is the firm size distribution. Around 73–95% of their sample were firms with fewer than 500 employees, and the median firm size was 10–99 employees. In our survey, the median firm had 6000 employees, placing it at the high-end of ACW's "Large Firms" category. Even on its own, this might explain many of the reported differences, since such small firms

<sup>3</sup> 22% and 25% for their Large Firms category (500+ employees).

are less likely to be able to have substantial Central R&D departments or to found innovation labs.

Another potential difference is sample scope. The ACW data focuses on manufacturing, whereas our data covers seven industries. There is also a geographical difference, with ACW reporting US values, whereas we cover eight countries. However, subsetting our data to only US manufacturing firms reveals an even stronger contrast, with 86% of the most important innovation projects coming from internal sources. Thus, this difference in sample construction does not explain the differences in results, it heightens them.

Likely the biggest explanation for the difference between the two sets of results comes from the types of innovation projects included in each analysis. ACW takes an industrylevel approach, as recommended by [1], which has the benefit of providing a view of the frontier of knowledge in that industry. In contrast, we take a firm-level approach which has the benefit of being more managerially relevant since many of a firm's most important innovations do not originate outside the industry (we see this quantitatively in singlegeography testing that we did, where only 27% of the most important innovations to the firm were new to the industry). It is, therefore, not surprising that by including these other 73% of projects – which firms report as being their most important innovation but which are not new to the industry – we see a different pattern. In particular, since knownto-the-industry innovations are more likely to be areas of expertise for internal innovators, it is not surprising that we see a higher share of internal innovation than do ACW.

Thus, we conclude that our analysis provides a significantly different window on innovation than ACW. Whereas their analysis covers greater firm size variation, ours covers more geographic and industry variation. More importantly, our results cover a broader range of the innovations that are important to firms and drive their competitive advantage, and thus our findings are more directly relevant to innovation leaders for managing their entire innovation portfolio.

In addition to the strengths of the survey, highlighted above, our design also implies limitations. Like many surveys in this area, we only gather a single wave of responses. This limits our analysis to cross-sectional analysis, whereas repeating the survey over multiple years would allow panel analysis that typically allows for better covariate controls and thus better causal analysis. Hence, our analysis should typically be understood as implies that "the relationship with X and Y is consistent with explanation Z," rather than asserting the stronger "Z causes X and Y." Another limitation of the survey arises out of our sampling choice. By sampling only large firms, \$500 M+ revenue per year, our analysis excludes innovation trends occurring in smaller firms. This exclusion also has differential effects across countries, based on the share of firms above this threshold in each country.

These differences also highlight an important implication for policy. Policymakers must be vigilant to distinguish between industry- and firm-level views of innovation, lest their view of innovation be incorrectly biased. In particular, industry-level surveys are particularly useful to understand innovators drawing most successfully from outside their industry. By contrast, policymakers should use firm-level analyses when considering managerial behavior or firm productivity, where a majority of the most important innovations would be missed if industry-level analyses were used.

Our findings also have implications for future research. First, they suggest that more surveys should directly compare firm - and industry- level views of innovation. This will allow them to answer many important questions, such as whether firms that adopt first gain the most benefit, or whether those that wait until the idea is more mature do better.

Another area that should be explored in future research is how the deepening of firm capabilities in an area changes its innovation sourcing. Our research suggests that we would expect firms to move from external to internal innovation sourcing as they build industry-leading capabilities. That said, our data is entirely cross-sectional. This makes causal claims hard to make. Future research should look to panel data with plausible exogenous shocks to establish the causality of this and other findings in our paper.
