Spanning Two Worlds? Corporate Accelerators and Corporate Venture Capital in Innovation Portfolios DRAFT Submitted to Wharton Technology & Innovation Conference 2020 January 2020 Sheryl Winston Smith BI Norwegian Business School [email protected]ABSTRACT This paper tackles a crucial problem in the theory and practice of global innovation strategy: In the face of technological and business model disruptions, often originating in innovative startups, how do companies respond with entrepreneurial ideas? Specifically, how do established companies use entrepreneurial approaches to facilitate access to startup innovation? This research examines the roles of corporate accelerators and corporate venture capital investing. Using a sample of companies with both corporate accelerators and corporate venture capital arms, this study provides preliminary descriptive evidence of the role of both types of external knowledge partnerships with startups. Specifically, the paper points to greater experimentation in CVC portfolios in terms of startup focus and industry distribution relative to corporate accelerator portfolios. At the same time corporate accelerators may provide access to greater geographic distribution in their portfolio companies. A novel algorithmic approach based on natural language processing and machine learning is used to characterize the distributions of portfolio companies. The paper raises questions for further study addressing the potential for complementarity and substitution amongst these two forms of corporate innovation in partnership with startups. Keywords: Corporate venture capital; accelerators; innovation; startups; natural language processing
32
Embed
Spanning Two Worlds? Corporate Accelerators and Corporate ......of “innovation theater”, and a recent report estimates that 60 percent of corporate accelerators that existed in
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Spanning Two Worlds?
Corporate Accelerators and Corporate Venture Capital in Innovation Portfolios
DRAFT
Submitted to Wharton Technology & Innovation Conference 2020
This paper tackles a crucial problem in the theory and practice of global innovation strategy: In the face of technological and business model disruptions, often originating in innovative startups, how do companies respond with entrepreneurial ideas? Specifically, how do established companies use entrepreneurial approaches to facilitate access to startup innovation? This research examines the roles of corporate accelerators and corporate venture capital investing. Using a sample of companies with both corporate accelerators and corporate venture capital arms, this study provides preliminary descriptive evidence of the role of both types of external knowledge partnerships with startups. Specifically, the paper points to greater experimentation in CVC portfolios in terms of startup focus and industry distribution relative to corporate accelerator portfolios. At the same time corporate accelerators may provide access to greater geographic distribution in their portfolio companies. A novel algorithmic approach based on natural language processing and machine learning is used to characterize the distributions of portfolio companies. The paper raises questions for further study addressing the potential for complementarity and substitution amongst these two forms of corporate innovation in partnership with startups. Keywords: Corporate venture capital; accelerators; innovation; startups; natural language
processing
2
INTRODUCTION
This paper tackles a crucial problem in global innovation: In the face of technological and
business model disruptions, often originating in innovative startups, how do companies respond
with innovative ideas? In particular, to what extend do established, global companies able
attempt to integrate internal corporate innovation efforts with external, internationally distributed
sources of entrepreneurial ideas? The relationship between established companies and startups
takes on growing importance as the locus of innovation increasingly comprises a global
ecosystem of opportunity seekers (Reuber et al., 2018) in which innovation efforts are dependent
upon a web of relationships with external parties (Papanastassiou et al., 2019).
Companies have long incorporated external partnerships to stimulate and extend internal
innovation efforts. Staying competitive in the face of disruptive technological and business
model innovation requires significant attention to novel ideas that originate in startups.
Traditionally, companies rely on corporate venture capital funds as a key component of their
strategic approach to accessing these external, early stage ideas (Dushnitsky et al., 2005; Maula
et al., 2012; Winston Smith et al., 2013). These funds allow companies to tap into ideas
emerging from startup companies through structured equity relationships. More recently,
however, established companies are using another tool to provide them with access to
entrepreneurial ideas and knowledge: partnerships with accelerators (Deloitte, 2015; Ream et al.,
2016). Corporate accelerators are increasingly deployed as an intersection between corporate
innovation and the startup ecosystem. The number of corporate accelerators has been growing
substantially every year since 2010, with 105 new corporate accelerators launched between 2013
and 2015 alone (Ream et al., 2016). A 2016 survey of the Forbes Global 500 found that 68% of
the top 100 companies engage with startups using a variety of mechanisms (Bonzom et al.,
3
2016). The survey identified a two-way flow of benefits of such collaboration, including faster
internal execution, changes in internal culture, enhanced brand image, and the ability to identify
new trends ahead of competitors (Bonzom et al., 2016). In practice, corporate accelerators face a
balance between the goals of venture launch and strategic fit with corporate innovation (Shankar
et al., 2018).
The literature provides deep insights into corporate responses to current and anticipated
disruptive changes that may impact their technological success or interfere with their business
model. The strategy literature includes a rich understanding of corporate partnerships that take
the form of strategic alliances (Gulati et al., 2009; Leiblein et al., 2002; Rothaermel et al., 2004)
and a variety of institutional arrangements companies use to be more entrepreneurial (Keil et al.,
2008b; Ott et al., 2017). The literature also points to the importance of a firm’s own experience
in shaping the ability to adapt to disruptive changes (Chen et al., 2012). Corporate venture
capital activity is perhaps most clearly related to the distinct relationship between the innovation
activities of large companies and access to startup knowledge (Wadhwa et al., 2006; Wadhwa et
al., 2016; Winston Smith et al., 2013) and of its limitations in that regard (Benson et al., 2010;
Dushnitsky et al., 2009). The literature further provides insight into challenges startups face in
the relationship with a corporate investor (Katila et al., 2008).
This paper brings together insights from the literature on corporate innovation
mechanisms with the growing literature on accelerators. Accelerators are characterized by
several distinct features that make them a novel organizational form, including formal
application and selection mechanisms for entry; pre-determined cohorts with a fixed length of
time (typically 3-4 months); and a formal ending point typically marked by a “Demo Day” event
in which startups in a given cohort pitch to investors (Clarysse et al., 2015; Cohen et al., 2014).
4
Most accelerators also take a small equity stake in the startup. The emerging literature on
accelerators suggests that accelerators facilitate experimentation by entrepreneurs, helping them
learn quickly which ideas have promise and which are likely to fail; this translates into quicker
exits both through acquisition and through quitting (Winston Smith et al., 2014). For
entrepreneurs, the impact of accelerators may be delivered through the intensive mentoring
experience (Cohen et al., 2018). Accelerators also foster intensive peer learning through the
cohorts (Winston Smith et al., 2015). The literature provides mixed evidence on whether
accelerators increase the speed of funding. In one study, entrepreneurs that participate in an
intensified learning portion of the government sponsored Startup Chile program are more likely
to raise subsequent funding (Gonzalez-Uribe et al., 2017). However, using a broader sample Yu
(2015) finds that accelerator-backed startups take longer on average to receive funding.
THEORETICAL FRAMEWORK
Companies regularly engage in external search for innovation, drawing on wide variety
of mechanisms to do so (Foss et al., 2013). Corporate venture capital (CVC) is one of the most
prominent and well-established approaches (Benson et al., 2009; Dushnitsky et al., 2005;
Wadhwa et al., 2006; Wadhwa et al., 2016). In contrast, corporate accelerators are a relatively
newer --and in some ways more controversial -addition to the corporate innovation toolbox
(CBInsights, 2019).
Several recent reports point to the growing use of corporate accelerators as an
increasingly practical intersection between corporate innovation and the startup ecosystem.
Similar to corporate venture capital, corporate accelerators are “outside-in” approaches to
innovation (Weiblen et al., 2015). In the general press, assessment of the value of corporate
accelerators is somewhat mixed. For example, one view is that corporate accelerators are a form
5
of “innovation theater”, and a recent report estimates that 60 percent of corporate accelerators
that existed in 2016 have since shut down (CBInsights, 2019). In practice, corporate accelerators
face a balance between the goals of venture launch and strategic fit with corporate innovation
(Shankar et al., 2018). Moreover, corporate accelerators also exist in relation to existing,
established forms of external knowledge seeking from startups, such as CVC.
As a first step in understanding the impact of corporate accelerators this research seeks to
first characterize the nature and extent of corporate accelerators in relation to existing CVC
programs. Similar to CVC, corporate accelerators provide the opportunity to assess the early
validity and relevance of nascent startups without a large commitment of resources.
Given the nascent status of this literature, this research takes a question-driven approach
to understanding the relationship between corporate accelerators and existing CVC investments
(Tidhar et al., 2019).
Experimentation and Corporate Innovation
There is a growing literature on the importance of experimentation and tolerance for
failure in corporate innovation (Keil et al., 2008a; Li et al.; Patel et al., 2015; Tian et al., 2011).
On the face of it, both corporate accelerators and CVC investing can be seen as approaches that
increase exposure to new, potentially different ideas, and thus serve as a funnel for greater
experimentation. Corporate accelerators and CVC investments might plausibly overlap or they
might plausibly diverge from one another, and in relation to larger to corporate innovation goals
and priorities.
Corporate accelerators and CVC are similar in that both represent portfolio approaches to
external innovation. In this regard, both approaches draw upon the logic of investing in a
relatively broad set of earlier stage ideas than would typically be developed using in-house
6
innovation (Fulghieri et al., 2009; Lin et al., 2011; Wadhwa et al., 2016). Traditionally, CVC
provides exposure to a larger set of ideas for a relatively small investment. As in independent
VC investing, CVC investors recognize that the outcomes of these investments will be skewed,
with many failures and some successes (Gompers et al., 2010). However, distinct from
traditional VC investing, which relies solely on financial return as a metric of success, CVC has
distinct performance goals of innovation outcomes as well as financial measures (Dushnitsky et
al., 2010). As in CVC portfolios, corporate accelerators are also tasked with generating
innovation returns while tracking financial performance metrics, such as VC investment, as well.
In keeping with the portfolio logic, both corporate accelerators and CVC units are
intended to increase exposure to a variety of ideas that are distinct from--yet complementary to—
in house innovation efforts. In this regard the goals of both types of programs are similar.
However, several fundamental questions pertain to the range of exposure to new ideas in
corporate accelerators relative to CVC investments. To the extent that corporate accelerators are
being established with goals of targeting innovation solutions, at the portfolio level, corporate
accelerators might plausibly be expected to provide relatively more targeted ideas compared to
CVC investments. The follow propositions apply at the portfolio level:
Proposition 1: Experimentation in range of ideas
Corporate accelerators provide a portfolio of more tightly aligned ideas
relative to the CVC investment portfolio.
Proposition 2:Industry diversity of portfolio
Corporate accelerators provide a portfolio of more tightly aligned industry
segments relative to the CVC investment portfolio.
On the other hand, the accelerators are able to draw on startups from a greater geographic
7
distribution relative to VC investing. Similarly, to the extent that corporate accelerators follow a
congruent, cohort-based and location-based approach as traditional accelerators, these corporate
accelerators might be able to bring in a portfolio of companies from a greater range of
geographic regions relative to CVC investments. Moreover, by having startups in house for a
concentrated period of time, corporate accelerators may be able to reduce the monitoring needs
that lead to geographic constraints in CVC investing. Proposition 3 follows at the portfolio level:
Proposition 3: Geographic diversity of portfolio
Corporate accelerators will invest in ventures from greater variety of
geographical locations relative to the CVC investment portfolio.
RESEARCH DESIGN, DATA, AND METHODOLOGY
This research draws on a sample of companies that have established both CVC
investment arms and corporate accelerators and analyzes these portfolios along multiple
dimensions. For the corporate investor, we seek to capture the value of experimentation through
exposure to the variety of new ideas from startups from two distinct sources, CVC investments
and corporate accelerators. We do this through several measures including deal size, industry,
and geographic distribution.
Further, we develop a novel, machine-learning natural language processing approach to
compare the relative similarity amongst the portfolio of startups in either the CVC portfolio or
the corporate accelerator.
Data and Sample
The data is derived at the deal level from corporate accelerator and corporate venture
capital deals in the CB Insights database, supplemented with information from other websites.
8
For the corporate partner, we seek to capture the value of experimentation through exposure to
the variety of new ideas from these startups. We do this through several measures including
industry, geographic, and business level diversity of investments into the startups that receive
investment.
The sample includes deals spanning the period 2009-2019 in order to include the time
period before and during which corporate accelerators become pervasive. The sample focuses on
several industries in which both CVC and corporate accelerators are prevalent, and where data
on both forms of investment are available. The final sample consists of corporate investors and
corporate accelerators in the following industries: financial services and internet
software/services.
The key criteria for inclusion are that the given corporate investor have both a CVC
investment program as well as an active corporate accelerator. Thus, while many companies
have CVC arms, they do not have corporate accelerators. Conversely, other companies have
taken the step to establish accelerators but do not make CVC investments. Finally, the sample
focuses exclusively on corporate accelerators and thus excludes stand-alone, traditional
accelerators. The final sample includes two companies within each of these two industries. The
final sample consists of Barclays and Wells Fargo in the financial services industry and
Microsoft and Amazon in internet software/services. These companies were selected for
preliminary analysis on the basis of having both CVC investment and corporate accelerators.
The final sample consists of 295 CVC investments and 678 CA investments by these four
companies.
Natural Language Processing (NLP) Analysis of Portfolio Companies
Natural language processing (NLP) is an algorithmic, machine-learning based approach
9
to computing similarity between text documents (Gentzkow et al., 2017; Hoberg et al., 2014;
Winston Smith, 2019; Winston Smith et al., 2013). CB Insights data provides uniform text
descriptions of each portfolio company. These text descriptions are then processed using Python
and the spaCy natural language processing module. Briefly, this approach involves
preprocessing steps of tokenizing and removing stop words, then mapping the text onto the large
English core corpus available in spaCy, producing a vector of words for each description. The
vectors corresponding to the texts of each of the portfolio companies are then compared
dyadically using a cosine similarity transformation trained using the GloVe w2v model
(https://nlp.stanford.edu/projects/glove/). This yields a cosine similarity score for each
combination, which is then the basis for analysis. Details are provided in the appendix.
PRELIMINARY FINDINGS
Deal Composition
The number of deals and total financing by source and company are summarized in
Table 1. In looking at the data several trends emerge. First, both the amount invested total and
amount per deal are substantially larger, in some cases by orders of magnitude, for CVC
investments relative to the corporate accelerators. This is true for all four firms and in both
industries. Second, true exits (through an IPO or an acquisition of the portfolio company) are
low overall, but particularly so for the corporate accelerator companies. Indeed, no exits are
Figure 1. Distribution of portfolio company descriptions using NLP: Barclays Accelerator
Figure 2. Distribution of portfolio company descriptions using NLP: Barclays CVC
20
Figure 3. Distribution of portfolio company descriptions using NLP: Wells Fargo
Accelerator
Figure 4. Distribution of portfolio company descriptions using NLP: Wells Fargo CVC
21
Figure 5. Distribution of portfolio company descriptions using NLP: Microsoft Accelerator
Figure 6. Distribution of portfolio company descriptions using NLP: Microsoft CVC
22
Figure 7. Distribution of portfolio company descriptions using NLP: Amazon Accelerator
Figure 8. Distribution of portfolio company descriptions using NLP: Amazon CVC
23
Figure 9. Distribution of portfolio company industries: Barclays Accelerator
Figure 10. Distribution of portfolio company industries: Barclays CVC
24
Figure 11. Distribution of portfolio company industries: Wells Fargo Accelerator
Figure 12. Distribution of portfolio company industries: Wells Fargo CVC
25
Figure 13. Distribution of portfolio company industries: Microsoft Accelerator
Figure 14. Distribution of portfolio company industries: Microsoft CVC
26
Figure 15. Distribution of portfolio company industries: Amazon Accelerator
Figure 16. Distribution of portfolio company industries: Amazon CVC
27
APPENDIX
The intuition and application of natural language processing techniques (NLP) and the
related field of information retrieval (IR) is developed in the computer science literature (Kwon
et al., 2003; Manning et al., 2008; Salton et al., 1986; Salton et al., 1975).2 Mathematically, IR
weights the relative similarity of knowledge from two sources of text to facilitate efficient
identification, based on the criteria of precision, i.e., the proportion of correct documents
retrieved, and recall, i.e., the proportion of relevant documents returned (Lease, 2007; Lewis et
al., 1996; Singhal, 2001).
The vector-space model underlies this process and enables direct quantification of document
similarity between a given pair of retrieved texts. This paper algorithmically executes a series of
steps to simplify text documents, assign relative weights to words, and map them into n-
dimensional vector space. A meaningful overlap between these vectors is then computed to
provide a metric for document comparison (Salton et al., 1986; Salton et al., 1975; Singh et al.,
2012). As such, this method is considered to be a ranked retrieval model (Manning et al., 2008).
These algorithms comprise the backbone of search engines (Google, 2013a, 2013b; Kwon et al.,
2003; Singhal, 2001)..1
Several papers in strategy research look to variants of NLP methodologies to quantify the
relative importance of words within a given text. Menon et al. (2018) use NLP to analyze
2 Briefly, IR as a field can be thought of as ‘… finding material (usually documents) of an unstructured nature
(usually text) that satisfies an information need from within large collections (usually stored on computers)
(Manning et al., 2008, p.1). IR covers wide ground: open-ended web search, domain-specific sea This paper focuses
on domain-specific search (Manning et al., 2008, p. 2). Search space can be conceptualized as a query seeking
answers relevant to a question at hand Technically, IR and NLP are distinct fields, where the former focuses on
probabilistic approaches and the latter on the use of “natural” (as opposed to “controlled”) queries (Lewis et al., 1996). However, although these fields were largely separate for decades, they are becoming increasingly
interchangeable as NLP adopts probabilistic approaches and IR broadens in scope (Lease, 2007; Lewis et al., 1996;
Manning et al., 1999; Singhal, 2001).
28
strategy change and differentiation from rivals. Lee and James (2007) use centering resonance
analysis (CRA), which utilizes network positioning of words in text to determine influence, in
this case to relate gender to media slant in the appointment of new chief officers. Ronda-Pupo
and Guerras-Martin (2012) use co-word analysis, in which they measure the likelihood of two-
words appearing together in a given document against a probabilistic backdrop, to study the
evolution of research concepts in the field of strategic management. In a similar fashion,
Kabanoff et al. (2008), apply machine learning techniques based on text classifying algorithms
to examine the evolution of strategic discourse by CEOs. In the context of innovation, and
closely related to the methodology in this paper, Kaplan et al. (2014) use topic modeling to trace
the evolution of breakthrough ideas in the emergent field of nanotechnology based on a form of
natural language processing known as latent Dirichlet allocation (LDA) analysis of patents. This
approach is based on probabilistic modeling of co-occurrence of words to infer meaning about
‘latent’ topics (Blei et al., 2003; Chang et al., 2009).
29
REFERENCES
Benson D, Ziedonis RH. 2009. Corporate Venture Capital as a Window on New Technologies:
Implications for the Performance of Corporate Investors When Acquiring Startups.
Organization Science 20(2): 329-351.
Benson D, Ziedonis RH. 2010. Corporate venture capital and the returns to acquiring portfolio
companies. Journal of Financial Economics 98(3): 478-499.
Blei DM, Ng AY, Jordan MI, Lafferty J. 2003. Latent Dirichlet Allocation. Journal of Machine Learning Research 3(4/5): 993-1022.
Bonzom A, Netessin S. 2016. 500 Corporations: How do the world’s biggest corporations deal
with the startup revolution?Startups (ed.), 500 Startups.
CBInsights. 2019. Why 60% Of Corporate Accelerators Fail After 2 Years.
Chang J, Boyd-Graber J, Wang C, Gerrish S, Blei DM. 2009. Reading Tea Leaves: How
Humans Interpret Topic Models. Neural Information Processing Systems.