Do Accelerators Actually Work? What the Research Says

Ask whether accelerators work and you will get an answer shaped by whoever is answering. Program directors cite graduate funding totals. Skeptics point out that the programs pick promising companies and then take credit for their promise. Both are describing the same data.

There is more rigorous research on this than the debate suggests, some of it going back more than a decade, and it converges on an answer that is narrower and more useful than either side's version. The effect is real in places, absent in others, and part of what gets measured as acceleration is selection working as intended rather than selection contaminating the result.

The Biggest Dataset, and What It Compares

The Global Accelerator Learning Initiative, run by Emory University with the Aspen Network of Development Entrepreneurs, has collected data from more than 23,000 entrepreneurs who applied to more than 360 acceleration programs across over 150 countries between 2013 and 2020.¹ Entrepreneurs complete a survey at application and then annually afterwards, whether or not they were accepted, which is what makes the dataset unusual.

In GALI's own published summary of its findings, over the year following application, revenue grew 50 percent for accelerated ventures against 30 percent for rejected ones, employee numbers grew 47 percent against 30 percent, and debt and equity financing grew 38 percent against 22 percent.² Those percentages come from GALI's earlier analyses rather than from the full 2013 to 2020 dataset. The 2021 five-year report presents the same three comparisons as average levels rather than growth rates, and on those averages the gap in employee and financing growth is a good deal narrower than the gap in revenue growth.¹ The revenue difference is the one that holds up across both presentations.

The comparison group is the part to be careful about, because it is routinely described wrongly. GALI compares ventures accepted into a program against ventures that applied to the same program and were rejected. That is not a randomized control group, not a matched sample drawn from startups generally, and not a regression discontinuity around the admission threshold. GALI says so itself, describing the rejected applicants as ventures that, "while not a 'control group' in the traditional sense, allow for a comparison of outcomes for ventures that were accelerated with ventures that have a relatively similar profile but did not receive acceleration services."¹

That distinction is the whole argument. Accelerators select on the same attributes that predict growth, so accepted and rejected applicants differ before the program starts. An emerging-markets investor made this point directly against GALI's 2018 report, which compared 526 participants against 1,733 rejected applicants: the study "doesn't control for the selection criteria that the programs use to choose their cohorts," and comparing the average performance of two differently selected groups cannot isolate what the program did.³

The Number That Should Change the Question

Averages across hundreds of programs hide the thing operators most need to know. In Cohen and Hochberg's survey of the accelerator model, drawing on the 2012 Seed Accelerator Rankings, the share of graduates receiving follow-on financing of 350,000 dollars or more within a year of graduating averaged 41 percent across the programs surveyed, with a range from 5 percent to 78 percent. The share achieving an exit by sale or IPO of a million dollars or more, measured as of the end of 2011, averaged 4 percent, ranging from 0 to 13 percent. Top programs accepted as few as 1 percent of applicants.⁴ The levels are a decade and a half old. The spread is the part that has held up.

A 5-to-78 spread is not noise around a mean. It says the average accelerator effect is close to meaningless as a decision input, and that a founder choosing a program, or a network deciding whether to partner with one, is making a program-level choice rather than a category-level one.

GALI's own data says the same thing from the other direction. Looking across the 52 programs with enough follow-up data for program-level analysis, the report notes that while average performance was above zero, there were programs performing badly enough that rejected ventures outperformed accelerated ones.¹

What Survives Causal Scrutiny

Two studies get closest to isolating an effect. Hallen, Cohen and Bingham, in Organization Science in 2020, used proprietary data on ventures accepted and "almost accepted" to a set of top accelerators, supplemented by a quantitative dataset built from public data and by qualitative fieldwork. Their finding is careful and worth quoting as written: they find evidence that "some, but not all, of the early accelerators we study substantially aid and accelerate venture development," alongside "some evidence of sorting dynamics."⁵ Two caveats travel with that. The sample is top accelerators, so it says nothing about the median program, and even inside that sample the effect held for some programs and not others.

Gonzalez-Uribe and Leatherbee, in the Review of Financial Studies in 2018, used a regression discontinuity design at Start-Up Chile, which ran two treatment conditions. One was basic services: equity-free funding plus co-working space. The other added structured entrepreneurship schooling on top. Schooling bundled with basic services significantly increased venture performance. Basic services on their own produced no evidence of any performance effect at all.⁶

It is worth being precise about what that rules out, because this result is frequently repeated in a form the paper does not support. The finding is not that money does not matter in general, and not that mentorship does not matter. It is that the cash-and-desk bundle showed no detectable effect on its own, and that adding a teaching component to that bundle is what produced the gain. The paper does not test schooling in isolation, so schooling alone is untested. The authors also limit their own external validity, saying the results are most relevant to ecosystem accelerators attracting young, early-stage businesses.⁶

The Finding Nobody Quotes

The uncomfortable result comes from GALI's own academic side. Lall, Chen and Roberts, in World Development in 2020, analysed a matched sample of 1,647 entrepreneurs who applied to 77 impact-oriented accelerators. Participants attracted significantly more outside equity than rejected applicants in the first follow-up year, and the authors argue this was not explained by cherry-picking obviously promising ventures during selection. Then the sentence that gets left out of program marketing: "the equity investment effect does not extend to ventures working in emerging markets, or to those with women on their founding teams. Thus, the benefits of accelerators for entrepreneurship-led development are not yet reaching the places and people that have the hardest time attracting capital on their own."⁷

Peter Roberts is GALI's academic lead at Emory. This is the same research program publishing a null result against its own headline finding, in a peer-reviewed journal, which is more than most program evaluations manage. GALI's 2021 report says the same thing in its own words: essentially all the net investment benefit went to teams in high-income countries, and participation in local accelerators does not significantly relieve the early-stage capital constraint in developing regions.¹

Faster Verdicts, Not Only Better Ones

Sandy Yu, in Management Science in 2020, compared roughly 900 accelerator companies across 13 accelerators against 900 matched non-accelerator companies, with a separate sample of rejected applicants as a check. Accelerator companies closed down earlier and more often, and raised less money conditional on closing. Her reading is that accelerators help resolve uncertainty about company quality sooner, letting founders make funding and exit decisions accordingly.⁸

Summarising that as "accelerators make startups fail more" is technically supported and substantially misleading. The more accurate version is that acceleration appears to speed up the verdict in both directions. For an investor, a fast no is close to as valuable as a fast yes. For a founder, an early verdict is a different proposition, and whether it reads as a service depends on whether the program's judgment is any good.

How Much of the Outcome Is the Program at All

One study puts a number on the ceiling. Chan, Patel and Phan, in the Strategic Entrepreneurship Journal in 2020, decomposed variance across 1,442 ventures from 117 accelerator programs in 22 countries and attributed between 11.13 and 14.18 percent of the variance in venture performance to accelerator membership, with 16.65 percent of revenue change, 5.15 percent of employee costs and 3.00 percent of employee growth attributable to it.⁹

Read that generously and it is a real effect. Read it plainly and roughly 85 to 89 percent of the variation in how these ventures performed had nothing to do with which accelerator they joined. A 2025 meta-analysis of 21 primary studies and 68 effect sizes found a statistically significant positive overall effect of accelerator participation, while also identifying publication bias in the literature and selection bias within programs, which tend to favour startups that already have funding.¹⁰ Both things are true at once: the effect exists, and the literature reporting it is tilted toward finding it.

The European Problem

Almost none of this evidence is European. The peer-reviewed causal work is American, Chilean or emerging-market. The one European exception we could find is not a journal article. A 2019 report for the UK Department for Business, Energy and Industrial Strategy by Jonathan Bone, Juanita Gonzalez-Uribe, Christopher Haley and Henry Lahr applied a regression discontinuity design to a single UK corporate accelerator, comparing applicants who scored just above and just below the interview threshold, and found roughly a 50 percent increase in survival, an increase of about one employee size band, and a 77.6 percent increase in funds raised.¹² The authors are explicit that this is a local effect for applicants at the margin of one program. That is the European evidence base: one accelerator, in one country, in one piece of grey literature. European operators are still largely reasoning about their own programs from studies run somewhere else.

The strongest European quasi-experimental result in the adjacent space points a different way. Santoleri, Mina, Di Minin and Martelli, in the Review of Economics and Statistics in 2024, used a sharp regression discontinuity on the EU SME Instrument, the Horizon 2020 predecessor of today's EIC Accelerator, and found large effects on employment growth, revenue growth, patenting, private equity receipt and survival. Their mechanism finding is the interesting part: funding effects were much more important than certification effects, and firms awarded the Seal of Excellence, which is certification without money, showed no significant performance improvement.¹¹

That is close to the mirror image of the Chilean result, where money and space alone did nothing. The two are not directly contradictory, since the treatments, populations and outcomes all differ, and an R&D grant competition is not a cohort program with mentors and a demo day. But it is a reminder that the mechanism is not settled, and that anyone claiming to know which ingredient does the work is ahead of the evidence.

What an Operator Should Take From This

Four things follow for anyone running a network, an incubator or a selection process.

Selection is a service, not an embarrassment. GALI's own framing is that the measured benefits come from a combination of the ability to select high-potential ventures, the market signal that selection provides, and the programming itself, and that selection "is in and of itself a service to the entrepreneurial ecosystem."¹ A program that picks well and does little else is still doing something useful for the investors downstream of it. It is just doing a different thing than it claims in its brochure.

Program-level variance is the fact to plan around. The 5-to-78 percent range in follow-on funding means a partnership decision needs that program's own numbers, not the category's.

Teaching is the component with the best causal evidence behind it, and the cash-and-desk bundle has a published null. A program deciding where to spend its next euro has a defensible answer there.

And measure the thing the research measures. Follow-on financing within a year of graduating, revenue change against the ventures you rejected, and time to a verdict are all trackable by any program willing to keep in touch with the applicants it turned down. Most do not, which is why most program evaluation is graduate testimonials. The comparison that makes the evidence above possible is the one nearly every operator is already sitting on and throwing away: what happened to the companies you said no to. Pynn exists partly because that data is scattered across inboxes and spreadsheets rather than held anywhere it could be read back.

Sources

1.      Matthew Guttentag, Abigayle Davidson and Victoria Hume, Does Acceleration Work? Five Years of Evidence from the Global Accelerator Learning Initiative, Emory University and the Aspen Network of Development Entrepreneurs, May 2021. https://www.galidata.org/assets/report/pdf/Does%20Acceleration%20Work_EN.pdf

2.      Global Accelerator Learning Initiative, Initial Insights from GALI. https://www.galidata.org/insights/

3.      Nicky Khaki, Ease Off on the Accelerators: Why GALI's Latest Study on Accelerator Programs May Be Overstating Their Impact, NextBillion, 21 December 2018. https://nextbillion.net/accelerators-overstating-their-impact/

4.      Susan Cohen and Yael V. Hochberg, Accelerating Startups: The Seed Accelerator Phenomenon, working paper, March 2014. https://leeds-faculty.colorado.edu/bhagat/Accelerators-Start-Ups.pdf

5.      Benjamin L. Hallen, Susan L. Cohen and Christopher B. Bingham, Do Accelerators Work? If So, How?, Organization Science, volume 31, number 2, March-April 2020, pages 378-414. https://pubsonline.informs.org/doi/abs/10.1287/orsc.2019.1304

6.      Juanita Gonzalez-Uribe and Michael Leatherbee, The Effects of Business Accelerators on Venture Performance: Evidence from Start-Up Chile, The Review of Financial Studies, volume 31, number 4, 2018, pages 1566-1603. https://academic.oup.com/rfs/article-abstract/31/4/1566/4104437

7.      Saurabh A. Lall, Li-Wei Chen and Peter W. Roberts, Are we accelerating equity investment into impact-oriented ventures?, World Development, volume 131, 2020, article 104952. https://www.sciencedirect.com/science/article/abs/pii/S0305750X20300784

8.      Sandy Yu, How Do Accelerators Impact the Performance of High-Technology Ventures?, Management Science, volume 66, number 2, 2020, pages 530-552. https://pubsonline.informs.org/doi/10.1287/mnsc.2018.3256

9.      Chien Sheng Richard Chan, Pankaj C. Patel and Phillip H. Phan, Do differences among accelerators explain differences in the performance of member ventures? Evidence from 117 accelerators in 22 countries, Strategic Entrepreneurship Journal, volume 14, number 2, 2020, pages 224-239. https://sms.onlinelibrary.wiley.com/doi/abs/10.1002/sej.1351

10.   Nikolaus Seitz, Martina Buratti, Erik E. Lehmann and Julie Kurrle, A meta-analysis towards the effectiveness of startup accelerators, The Journal of Technology Transfer, published online 18 August 2025. https://link.springer.com/article/10.1007/s10961-025-10218-6

11.   Pietro Santoleri, Andrea Mina, Alberto Di Minin and Irene Martelli, The Causal Effects of R&D Grants: Evidence from a Regression Discontinuity, The Review of Economics and Statistics, volume 106, number 6, 2024, pages 1495-1510. https://direct.mit.edu/rest/article-abstract/106/6/1495/112422/The-Causal-Effects-of-R-amp-D-Grants-Evidence-from

12.   Jonathan Bone, Juanita Gonzalez-Uribe, Christopher Haley and Henry Lahr, The Impact of Business Accelerators and Incubators in the UK, Nesta for the UK Department for Business, Energy and Industrial Strategy, BEIS Research Paper Number 2019/009, October 2019. https://assets.publishing.service.gov.uk/media/5da6eb24e5274a5cae34c00c/The_impact_of_business_accelerators_and_incubators_in_the_UK.pdf