Mass producing innovation: why accelerators are not a paradox

As presented at The 1st Indonesia International Conference on Science Technology Park (IICSTP) on October 12-13th by Go1 CEO Andrew Barnes.

There are a growing number of accelerator programmes designed to start and support innovative startup businesses. Many accelerators are increasing the size of their intakes, with some programmes now launching over 200 new companies per year. On first inspection the large numbers and consistent approach taken to accelerating the participating companies appears to be in conflict with producing innovative and disruptive companies.

This paper uses Y Combinator as a single case study to investigate whether increasing the number of companies within a batch has resulted in longer or shorter timeframes for companies to achieve an exit (through acquisition or initial public offering). The paper finds that the timeframe for achieving an exit for Y Combinator companies is reducing, even while batch size has sharply increased. There is no statistically significant correlation between the cohort size and the initial money raised during the programme.

1. Introduction

Disruptive new ventures are a hallmark of capitalism, where entrepreneurship revolutionizes existing economic structures (Schumpeter 1942). Creating new industry paradigms can be highly profitable. In the last four decades high-growth, venture capital backed startups have accounted for less than half of new US public companies, but have captured over 63% of the total market capitalization (Gornall and Strebulaev 2015). More recently, over the last ten years startup accelerators have emerged as a new method for incubating and investing in disruptive businesses (Miller and Bound 2011). Accelerators provide resources (generally in the form of seed funding, networks and mentorship) to groups of participating companies.The accelerator model was first pioneered by Y Combinator (YC), a programme that originated in Boston and now operates from Mountain View in California (Kim and Wagman 2014; Miller and Bound 2011; Dempwolf, Auer, and D’Ippolito 2014).

The first cohort of YC companies in 2005 (termed batches) featured less than ten businesses. YC has grown over the last decade and now launches over 200 companies each year. Intuitively there would seem to be a conflict between launching disruptive companies at scale. Indeed, existing research suggests that as companies grow larger, their ability to innovate may decrease (Ackermann 2012). The team behind YC are not unaware of the difficulty inherent in scaling the pursuit of innovation, and despite this have continued to steadily expand. YC’s mass production of high growth companies offers an opportunity for study: does the increasing size of cohorts have a negative effect on startup outcomes? This paper investigates whether there is a relationship between cohort size (in a given year) on seed investment round size and whether cohort size negatively impacts the average length of time for an acquisition.

1.1 Can innovation scale?

There is reason to be both optimistic and pessimistic when considering whether innovation can scale. From a simplistic perspective there are certain types of advances that require sufficient size and scale to support the necessary research and development activities. The Large Hadron Collider offers unique insight into uncharted territory, and in turn may yield disruptive breakthrough findings. Without the support of countless scientists, many countries and many billions of dollars in investment it would not be possible. But this form of scale – concentrated on a single project – is not what YC and its peers are attempting. Instead YC’s programme supports a great diversity of startup businesses, all of which receive relatively few resources. As batch sizes increase, the time and attention from YC must be spread thinner still.

To support each company, YC must be able to systematize the elements required to support innovation. Isaac Asimov (2014, 2) suggests that “the process of creativity, whatever it is, is essentially the same in all its branches and varieties, so that the evolution of a new art form, a new gadget, a new scientific principle, all involve common factors.” If Asimov is correct, and if YC have been able to identify scalable means to deliver enough of the common factors, then there could be the opportunity to improve the outcomes for companies participating in YC, even while the size of cohorts increases.

In contrast to this, there are many examples where innovation has been negatively impacted as group size increases. In academia, research indicates there are decreasing returns as team size grow (Lee and Bozeman 2005). The productivity of corporate research and development teams decreases as staffing numbers increase, particularly for heterogeneous groups (Cummings et al. 2013). Venture capital firms seeking to increase the size of their portfolios also experience negative performance tradeoffs (Kim and Wagman 2014; Kanniainen and Keuschnigg 2004; Cumming 2006; Fulghieri and Sevilir 2009). Accelerators too may experience diseconomies of scale (Hallen, Bingham, and Cohen 2014).

Y Combinator differs to other accelerators in two important respects that may enable it to scale more successfully. First, by being renowned as the premier accelerator globally, YC is able to attract the top tier of potential startup applicants (Hochberg and Kamath 2012; Pauwels et al. 2016). Attracting the very top tier of applicants (ahead of other accelerators) is particularly important in an industry where outliers are essential (Kim and Wagman 2014). Second, YC strongly espouses independence and autonomy for the companies that participate in their programme. For example, unlike other accelerators which will often provide free or subsidized office space, Y Combinator does not (Radojevich-Kelley and Hoffman 2012). Paul Graham, one of the original co-founders of YC, cites the need for companies to develop their own culture (Stross 2012). Greater independence may certainly help YC companies innovate.

1.2 Pros and cons for YC’s increasing cohort size

There are potentially many negative impacts associated with increasing the size of an accelerator cohort. If the common factors for innovation described by Asimov are not systematized and designed to scale, then there will be less optimal outcomes on average for the participating companies. One obvious effect of larger cohort sizes is the increased competition for investment funding. If the available capital for seed investment is fixed, then increasing the average number of YC companies should result in a lower average seed round for each company. The increased competition for each investor dollar may be compounded by the perception of lower average quality as cohort size increases.

With increasing numbers of heterogeneous companies there will also be limits on the quality of advice and mentorship that can be provided. The homogeneity of advice given to an increasingly large cohort is more likely to be poorly suited to the unique needs of individual companies (Hallen, Bingham, and Cohen 2014). Unsuitable advice can negatively impact the chances of success.

1.3 Structure of paper

2. Research methods and data

2.1 Research Design

2.2 Data selection

2.3 Methods for Analysis

3. Results

3.1 H1: Does cohort size have an impact on the amount of seed funding raised by each company?

The 95% confidence interval for  has a lower bound of  and an upper bound of 2456.95.

3.2 H2: Is there an effect on the time it takes for a company to exit based on the cohort size?

The second hypothesis investigated is whether there is a relationship between the cohort size and the average time taken for a liquidity event (either an IPO or acquisition). As no Y Combinator company has been publically listed I only consider acquisitions. There were 127 acquisitions in the dataset and the results are summarized in Table 3 below. The 95% confidence interval for  in Table 3 has a lower bound of and an upper bound of 0.004.

There is a natural ceiling on the maximum number of days for an exit event based on the year that a company participated in YC: the average time taken for companies to be acquired from the 2015 YC cohort is likely much less than average time taken for participants of the 2010 cohorts. To control for this ceiling, I have included a variable for the total number of days that the company has been operating in Table 4. The 95% confidence interval for  in Table 4 has an upper value of  and a lower value of  -0.04.

The results in Table 3 and Table 4 do not provide support for any positive or negative impact with increasing cohort sizes. In both regressions there is no statistically significant relationship at  between cohort size and the average number of days for an acquisition. The effect size is also small and negative: the largest change recorded to cohort size (from around ten participants to 210) would only reduce the average number of days for an acquisition by four.

4. Discussion and conclusion

4.1 Seed round and time to acquisition

This paper began with the question of whether the success of Y Combinator’s portfolio companies can continue as YC scales the size of its programme. YC’s programme is now twenty-times larger than when it began. The investigation of 991 YC companies suggests that the increasing cohort sizes have not had a noticeable impact on the time taken for acquisitions or for the amount of seed funding raised. I hasten to note that a statistically insignificant finding does not itself suggest that the tested hypothesis is false (Ellenberg 2014). However, the narrow confidence interval for the effect size of cohort size does indicate that any potential positive or negative effect is minimal.

The minimal impact that increasing cohort sizes has had on the average amount of seed funding is unexpected. Intuitively one might expect that the greater the number of companies, the greater the competition and lower average investment each would receive. The results in Table 2 suggest that there is a statistically insignificant positive correlation between the size of cohort and the average seed round. Table 5 provides an overview per year of the average seed round and the number of YC companies in that year. The results suggest that YC’s brand as the world’s top accelerator may have meant its companies have been protected from the increased competition amongst peers. As Y Combinator’s brand has grown, so too has the number of applicants for its programmes. This has meant YC has been able to maintain an acceptance rate ‘below 3%’ (Yin and Luo 2015, 23). The negative impact of increasing cohort size is probably more acutely felt by accelerators that lower the entry bar.

The second finding was that YC’s increasing cohort size has had a negligible impact on the average time for an acquisition. There is already research showing that participation in an accelerator like YC has demonstrated reductions in the average time before acquisition (Smith and Hannigan 2015). What is unexpected is that cohort size has had no to minimal impact on the average time until acquisition.

4.2 Implications for other accelerators

The findings suggest certain implications for accelerators. It does appear that there may indeed be scalable “common factors” when it comes to producing innovative companies (Asimov 2014, 2). Y Combinator’s success in scaling the size of its cohorts suggests that it has systematized some of them. The literature review identified some of the unique strategies that YC uses to support innovation. The first is its focus on individual autonomy. Practices such as not offering office space are in contrast to ‘virtually all accelerator companies’ (Radojevich-Kelley and Hoffman 2012, 60). By strongly encouraging participating companies to develop their own cultures and environments, YC is attempting to encourage heterogeneity between their portfolio companies.

Second, YC has been able to keep acceptance rates low, despite the increase in cohort size. Increasing cohort size at the expense of quality would likely have a negative impact on both seed funding size and acquisition prospects. As Y Combinator has been able to maintain and even grow its brand, it has been able to attract an increasing number of applicant companies. With acceptance rates below 3% (which is better than industry averages), selection to YC continues to be a strong quality signal for participating companies (Yin and Luo 2015, 23).

Replicating the success of YC in is not trivial. However, a focus on the elements that will support innovation at scale appears to be an essential ingredient.

4.3 Limitations and Further Research

As with all studies, there are limitations which must be considered with this investigation. A primary limitation are the variables available for investigation. When considering whether cohort size has an impact on firm outcomes, the ideal outcome variable would be firm market capitalization. The amount of time taken for acquisition is an imperfect proxy for success. Unfortunately, data on valuation or market capitalization was not available. Further research in the area would benefit from access to such data.

With that said, the limited information on acquisition values seems to suggest that YC’s performance is continuing to strengthen. One of the most recent YC exits (and also its largest to date) is Cruise, an autonomous vehicle company, that reportedly sold to General Motors for $1bn (Crunchbase 2016). Cruise was part of YC’s 2014 cohort, which was the largest cohort year for YC to date.

Another potential limitation relates the incentives of accelerators at scale. For accelerators that are able to develop focused, larger portfolios there are strategic benefits from exiting underperforming companies quickly (Fulghieri and Sevilir 2009). This dynamic doesn’t seem to apply as much to YC as it does not have a focused portfolio (the diversity of participating companies appear to be increasing with time) and does not have control over exiting firms (either through acquisition or by shutting them down).

Finally, while not a limitation, it is worth noting that portfolio size is a function of an accelerator’s business model. The typical accelerator business model ‘is to invest in a set of ventures with a relatively small amount of money rather than continue to support the ventures in multiple rounds’ (Kim and Wagman 2014, 5). Accordingly, accelerators are able to invest in a large number of potential firms. More recently Y Combinator has opted to adopt a different model, guaranteeing follow on investment in all its portfolio companies up until a valuation of $300 million (Altman 2015). Given the unique nature of YC’s future strategy, there is an opportunity for further and deeper research.

Acknowledgements

I would like to acknowledge and thank Assoc. Professor Dr Damian Hine at the University of Queensland Business School for reading an early draft of this paper.

REFERENCES

Ackermann, Stephen J. 2012. Are Small Firms Important? Their Role and Impact. Springer Science & Business Media.

Altman, Sam. 2015. “YC Continuity.” Y Combinator Posthaven. October 16. https://blog.ycombinator.com/yc-continuity-fund.

Asimov, Isaac. 2014. “Published for the First Time: A 1959 Essay by Isaac Asimov on Creativity.” MIT Technology Review. October 20. http://www.technologyreview.com/view/531911/isaac-asimov-asks-how-do-people-get-new-ideas/.

Cohen, Jacob. 1994. “The Earth Is Round (P<. 05).” American Psychologist 49 (12): 997.

Cohen, Susan, and Yael V. Hochberg. 2014. “Accelerating Startups: The Seed Accelerator Phenomenon.” Available at SSRN 2418000. http://papers.ssrn.com/sol3/Papers.cfm?abstract_id=2418000.

Cottrell, A, and R Lucchetti. 2016. “Gretl User’s Guide.” http://ricardo.ecn.wfu.edu/pub//gretl/manual/en/gretl-guide.pdf.

Crunchbase. 2016. “Mar 11, 2016: General Motors / Cruise | Crunchbase - General Motors Acquires Cruise for $1B in Cash & Stock.” March 11. https://www.crunchbase.com/acquisition/d7ca7de8846a22a446e474e388617673.

Cumming, Douglas J. 2006. “The Determinants of Venture Capital Portfolio Size: Empirical Evidence.” Journal of Business 79: 1083–1126.

Cummings, Jonathon N., Sara Kiesler, Reza Bosagh Zadeh, and Aruna D. Balakrishnan. 2013. “Group Heterogeneity Increases the Risks of Large Group Size a Longitudinal Study of Productivity in Research Groups.” Psychological Science 24 (6): 880–890.

Dempwolf, C. Scott, Jennifer Auer, and Michelle D’Ippolito. 2014. “Innovation Accelerators: Defining Characteristics Among Startup Assistance Organizations.” SBAHQ-13-M-0197. US Small Business Administration. https://www.sba.gov/sites/default/files/rs425-Innovation-Accelerators-Report-FINAL.pdf.

Ellenberg, Jordan. 2014. How Not to Be Wrong: The Power of Mathematical Thinking. Penguin Publishing Group.

Fulghieri, Paolo, and Merih Sevilir. 2009. “Size and Focus of a Venture Capitalist’s Portfolio.” Review of Financial Studies 22 (11): 4643–4680.

Gornall, Will, and Ilya A. Strebulaev. 2015. “The Economic Impact of Venture Capital: Evidence from Public Companies.” Available at SSRN 2681841. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2681841.

Graham, Paul. 2013. “YC W13 Will Be Smaller.” http://old.ycombinator.com/w13smaller.html.

Hallen, Benjamin L., Christopher B. Bingham, and Susan L. Cohen. 2014. “Do Accelerators Accelerate? A Study of Venture Accelerators as a Path to Success.” Academy of Management Annual Meeting Proceedings, January, 747–52. doi:10.5465/AMBPP.2014.185.

Hochberg, Yael V. 2016. “Accelerating Entrepreneurs and Ecosystems: The Seed Accelerator Model.” Innovation Policy and the Economy 16 (1): 25–51.

Hochberg, Yael V., and Kristen Kamath. 2012. “US Seed Accelerator Rankings.” Kellogg School of Management, Northwestern University. http://tech.co/wp-content/uploads/2012/08/Accelerator-Companion-FINAL.pdf.

Kanniainen, Vesa, and Christian Keuschnigg. 2004. “Start-up Investment with Scarce Venture Capital Support.” Journal of Banking & Finance 28 (8): 1935–1959.

Kim, Jin-Hyuk, and Liad Wagman. 2014. “Portfolio Size and Information Disclosure: An Analysis of Startup Accelerators.” Journal of Corporate Finance 29: 520–534.

Lee, Sooho, and Barry Bozeman. 2005. “The Impact of Research Collaboration on Scientific Productivity.” Social Studies of Science 35 (5): 673–702.

Miller, Paul, and Kirsten Bound. 2011. “The Startup Factories.” In NESTA. http://businessincubation.com.au/wp-content/uploads/StartupFactories-Accelerators-Evaluation-NESTA-June-2011.pdf.

Morrill, Danielle. 2015. “Why Is the Number of Seed Rounds Raised in 2014 Down 30%? Exploring the Connection Between Startup Funding Activity and U.S. Interest Rates.” Mattermark. January 16. https://mattermark.com/exploring-the-connection-between-startup-funding-activity-and-u-s-interest-rates/.

Pauwels, Charlotte, Bart Clarysse, Mike Wright, and Jonas Van Hove. 2016. “Understanding a New Generation Incubation Model: The Accelerator.” Technovation 50: 13–24.

Radojevich-Kelley, Nina, and David Lynn Hoffman. 2012. “Analysis of Accelerator Companies: An Exploratory Case Study of Their Programs, Processes, and Early Results.” Small Business Institute® Journal 8 (2): 54–70.

Schumpeter, Joseph Alois. 1942. Capitalism, Socialism And Democracy. Kessinger Publishing, LLC.

Smith, Sheryl Winston, and Thomas J. Hannigan. 2015. “Swinging for the Fences: How Do Top Accelerators Impact the Trajectories of New Ventures?” In DRUID15. Rome. http://druid8.sit.aau.dk/druid/acc_papers/5ntuo6s1r5dvrpf032x24x5on5lq.pdf.

Starbuck, William H. 2006. The Production of Knowledge : The Challenges of Social Science Research. Oxford: Oxford University Press.

Stross, Randall. 2012. The Launch Pad: Inside Y Combinator, Silicon Valley’s Most Exclusive School for Startups. Penguin UK. https://books.google.com.au/books?hl=en&lr=&id=231A_MFcmzkC&oi=fnd&pg=PT6&dq=startup+accelerators+y+combinator&ots=ltuTHXLSFD&sig=l4Qvn5LsvZxZScBI2OZkpqDuYYY.

Yin, Bangqi, and Jianxi Luo. 2015. “Critical Factors in the Selection of Start-up Incubator Residents.” Available at SSRN 2735465. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2735465.

Yuxian Eugene Liang, and Soe-Tsyr Daphne Yuan. 2016. “Predicting Investor Funding Behavior Using Crunchbase Social Network Features.” Internet Research 26 (1): 74–100. doi:10.1108/IntR-09-2014-0231.