Exploration and the termination of inventions: the role of the structure of the firm’s knowledge base and its failure experience

ABSTRACT

Firms manage their R&D portfolios by continuously evaluating and selecting which inventive paths to maintain and which ones to terminate. Prior research found that the extent of exploration in an invention increases the invention’s likelihood of termination. We inquire about organisational contingencies that impact the evaluation and selection of exploratory inventions. We suggest that the structure of an organisation’s knowledge base and its failure experience are particularly relevant for better comprehending the conditions under which exploratory inventions are more or less likely to be terminated. We find that the positive effect of exploration on patent termination is weakened with increasing level of decomposability in organisation’s knowledge base and increasing failure experience but only up to the moderate levels of these moderating variables. Empirically, we examine patent maintenance decisions in the biopharmaceutical industry. We discuss the contributions of our study for the research on exploration, invention termination, knowledge networks and failures.

KEYWORDS:

• Invention termination
• exploration
• knowledge network structure
• decomposability
• failure experience
• patent renewal

JEL CLASSIFICATION:

• O32
• O34
• D22

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Notes

¹ We rely on the conventional distinction in the innovation and economics literature between invention and innovation. Innovation refers to the commercial application of an invention (Fleming 2001; Schumpeter 1939). Not all inventions give rise to an innovation. Thus, an invention is likely to be maintained for as long as the firm expects to extract value from it (or a related invention) through a commercial application.

² The empirical research on innovation often uses the technological categories listed on the patent or the prior art citations made in a patent in order to capture the exploration in an invention (e.g. Grigoriou and Rothaermel 2014; Rosenkopf and Nerkar 2001).

³ Given that highly exploratory inventions draw upon knowledge that is largely not local, these inventions share fewer interdependencies with other elements in the firm’s knowledge base (Greve 2007; Lowe and Veloso 2015). Yet, research has shown that when an invention has interdependencies with other inventions in the firm’s portfolio, decision makers will be hesitant about abandoning the invention to prevent possible negative impacts on the remaining inventions in the firm’s portfolio (Khanna, Guler, and Nerkar 2018; Liu et al. 2008).

⁴ This is different from Khanna, Guler, and Nerkar’s (2018) conceptualisation. They examined the knowledge interdependencies within and between research programs (as well as the interaction between them), but not across the firm’s entire knowledge base. This is also different from what Delerue and Sicotte (2020) investigated, as their unit of analysis is the R&D projects rather than the knowledge elements comprising the firm’s knowledge base.

⁵ In fact, discovery related costs represent only about 2% of the total drug development cost, whereas clinical trials amount to about two thirds of the cost (DiMasi, Grabowski, and Hansen 2016).

⁶ https://clarivate.com/products/cortellis-family/.

⁷ Many of these firms are diversified players with established position across multiple therapeutic areas and indications (and not limited to cancer therapy). By choosing to focus only on firms that have launched at least one product we include mainly large integrated companies that cover most of the stages of the pharmaceutical value chain. However, for our analysis we utilise the entire drug portfolio of firms, not only cancer drug projects.

⁸ Given that SDC Platinum does not provide complete coverage of M&A’s and divestitures, we had to complement and cross check this data by manual web searches.

⁹ https://bulkdata.uspto.gov/.

¹⁰ For a patent granted in January 1996, the first maintenance decision is made in 1999, and the third (last) maintenance decision is made in 2007. For a patent granted in January 2009, the first maintenance decision is made in 2012, and the third maintenance decision is made in 2020.

¹¹ Because we needed to have at least one patent that incorporates a coupling of technological classes for building this measure, in our analysis we included only firm-years for which the firm had at least one patent in the past three years that combined at least two technological classes.

¹² The time window of 1999–2022 was chosen in order to allow for three years’ time-lag for independent and control variables and to observe the first maintenance decision for the cohort of patents granted in 1996.

¹³ The analysis of marginal effects revealed that the marginal effect of exploration is negative but not significant at very low values of the variable and it is positive and significant starting from the value of 0.27. We also implemented the tests recommended by Haans, Pieters, and He (2016) and Lind and Mehlum (2010) to verify the existence of a non-linear relationship and did not find support for the significant negative effect of the variable at left-hand side of the variable’s value range.

¹⁴ In these models we observe each patent only once for the given maintenance period (unlike patent-maintenance level models used for main analysis including all the maintenance decisions for a patent).