The Good, the Bad, and the Ugly for Incubators and Accelerators

In 2008, Airbnb was running out of ways to stay alive.

Brian Chesky and Joe Gebbia had accumulated tens of thousands of dollars in credit-card debt for the operating costs of the business. Investors were repeatedly dismissing the idea of renting out spare rooms to strangers. At one point, the founders generated more revenue selling limited-edition Obama O’s and Cap’n McCain cereal boxes during the US presidential election than they did from Airbnb itself. The cereal was a way of buying the company a little more time, but the founders needed something better to keep the business above water.

Airbnb eventually applied to Y Combinator, but there was little evidence that it would succeed. Paul Graham (co-founder of Y Combinator) later admitted that the partners were not particularly enthusiastic about the business. Customers did not appear to want to do anything with it either. What stood out to the Y Combinator team were the founders instead. According to Graham’s account of Airbnb’s early years, they appeared unusually resourceful, intensely committed, and willing to act immediately on feedback. This won them their place in the accelerator.

The advice from Y Combinator was to get growth started immediately.

“Where are your users?” – they asked.

Most of them were in New York.

“So go to New York.” – was the reply.

Chesky and Gebbia began meeting hosts in person. They photographed apartments and helped rewrite listings. They also began observing how people used the service. The work was slow and impossible to automate, but it exposed problems that had remained invisible from behind a laptop.

Revenue began to rise, and by the end of Y Combinator, Airbnb had become credible enough to attract investment. The company that investors had dismissed would eventually become one of the most valuable businesses produced by an accelerator.

The story has since become evidence for the power of Y Combinator, and, by extension, for startup incubators and accelerators more broadly. 

But it also raises a question.

Do these programs make companies successful, or are they simply good at identifying founders who were already likely to succeed? 

In other words, the better a program becomes at selecting exceptional companies, the more difficult it becomes to separate the value of the program from the quality of the companies it admits. 

This article breaks down whether incubators and accelerators genuinely improve startups, and where they create value or fall short.

The Counterfactual Company

Every claim that an accelerator or incubator created a successful company depends on a comparison.

What would have happened to the same company without the program?

Since that exact company doesn’t exist, we can’t observe that outcome. Airbnb either joined Y Combinator, or it did not. Researchers can compare it with rejected applicants or similar startups, but no alternative Airbnb exists in which the founders made different decisions under otherwise identical conditions.

Economists refer to this missing outcome as “the counterfactual.” It is the result Airbnb would have produced had it followed another path. But because only one path occurred, the accelerator’s contribution can never be directly observed. This greatly complicates studying the effects of accelerators and incubators.

In particular, separating the effect of the program from the quality of the founders is especially difficult at prestigious accelerators and incubators. They choose from large applicant pools and deliberately admit companies they believe have unusual potential. In Airbnb’s case, Y Combinator selected the founders because of their behavior during a brief interview where they appeared unusually determined. This alone is an indicator of the potential future success of the company. 

This creates a selection effect that conventional comparisons can’t quite capture. Accelerator-backed companies may outperform because the program improves them. They may also outperform because the program chooses founders who learn quickly and can act under uncertainty.

The available research on whether these programs are impactful shows support for accelerators, but it also highlights the fact that making broad conclusions is difficult.

A 2025 meta-analysis combined 21 primary studies containing 68 effect sizes. It found a statistically significant positive overall relationship between accelerator participation and new-venture performance. 

However, the effect varied substantially across the studies. Program duration, cohort size, sponsorship model, and geographic context all influenced the results. The researchers also identified publication bias, which means studies reporting favorable accelerator effects may be more likely to appear in the published literature than studies finding weak or negative results. 

Nonetheless, accelerators appear to improve startup performance on average. But it has to be noted that a 3-month investor-backed program in Silicon Valley, a publicly funded regional accelerator, and a university incubator supporting scientific spinouts may all be grouped together here. Meanwhile, their objectives, participants, and resources can be entirely different.

Asking whether accelerators and incubators work is sort of like asking whether schools work. The answer will heavily depend on the institution and its participants.

Separating Selection from Value Added

Unsurprisingly, researchers studying schools run into similar challenges to those studying accelerators. High-performing schools may improve their students, or they may enroll children who would have performed well elsewhere. Economists therefore estimate a teacher’s or school’s “value added” by examining whether comparable students achieve different results under different instructors.

In a 2026 NBER paper “Beyond Demo Day: Sorting and Value Added in Startup Accelerators,” Youn Baek and Deepak Hegde borrowed a method used in education research and applied the framework to approximately 750,000 US startups linked to 329 accelerators. 

Their analysis attempted to separate two forces that accelerator success stories usually combine.

The first is “sorting,” meaning that better ventures are more likely to enter accelerators and to sort into programs with stronger estimated performance. The second is “value added” for each program. 

Both of these matter, they found, firstly because higher-quality founders sort into stronger accelerators. (This is also why prestigious programs typically begin with an advantage, which is then reflected by having strong portfolios, which then attracts more high-quality candidates, and so on.) 

Secondly, Baek and Hegde found that companies entering higher-value-added accelerators were more likely to reach important milestones, including raising venture capital, generating revenue, employing workers, or achieving a successful exit.

It should be noted that this was not a universal effect, however. In fact, the positive results come from a small concentration of programs. Around 60% to 80% of the programs studied produced worse outcomes than the estimated no-accelerator alternative. The positive overall effect was driven by a relatively small group of high-performing accelerators, while most programs added little value or left participating companies worse off.

This further complicates any attempt to judge how successful incubators and accelerators really are.

A program can possess an impressive portfolio while adding limited value to the median participant. It can also work effectively with less obvious companies whose outcomes would look modest beside Airbnb but are significantly better than the outcomes those companies would otherwise have achieved.

Then, we should also note that portfolio value and program value are related, but they are not equivalent. I.e., value means something different for each stakeholder.

Venture investors are rewarded for portfolio value. One enormous outcome can compensate for many failures. Meanwhile, founders face a different calculation since they care about whether participation improves the probability and quality of their own outcome.

An accelerator can therefore be an excellent investment vehicle while remaining a weak intervention for most of the founders it accepts.

So when a program is described as “successful,” the next question should be: successful for whom? For founders, the more useful question is what the program can actually change for them.

What Accelerators and Incubators can Change 

Early-stage companies never run short of problems. What a good accelerator does is help founders figure out which ones deserve attention first, and how to attack them.

This is where these programs tend to earn their keep. They’re good at spotting the constraint that’s actually holding a company back, and advice that sounds obvious in the abstract can be worth a great deal when it lands on the right company at the right moment.

The research backs this up as well. A study of Start-Up Chile found that funding and coworking space alone did little for company performance. Founders only pulled ahead when those resources came bundled with structured entrepreneurship education.

There’s also evidence about what that education should teach. A series of randomized trials across Milan, Turin, and London trained founders to treat their business idea as a hypothesis and test it the way a scientist would, while a control group relied on intuition. Trained founders shut down weak projects more readily, pivoted more decisively, and earned more revenue on average. What the training changed was how founders evaluated their own evidence.

Along the curriculum, who delivers the teaching matters as much as the content. Another experiment compared advice from experienced entrepreneurs with advice from consultants and found the entrepreneurs moved the needle more. Founders take advice more seriously when it comes from someone who has actually sat with the same decisions, which is why most programs build their mentor benches out of former founders and operators rather than professional advisors.

Programs also supply something no advisor can, which is a schedule with consequences. A demo day 12 weeks out forces decisions that founders working alone might defer for a year. Whether that pressure produces real progress or just the appearance of it is a question we’ll return to.

What none of this changes is the underlying quality of the opportunity. A program can refine how founders decide and shorten how long those decisions take. Advice, though, is only half the story. Some of what an accelerator does for a company happens before the program even starts.

Selection is Part of the Product

In his 1968 paper The Matthew Effect in Science, the sociologist Robert K. Merton observed that eminent scientists tend to receive outsized recognition for their work, and that this recognition attracts further attention and resources, which makes future success more likely. Accelerators and incubators do something similar for companies.

Getting selected is valuable in itself. When a reputable program accepts a company, the market can see it too. This helps investors know the founders survived a competitive filter. Prospective employees also read the company as less precarious. For example, job applicants were 26% more likely to click on a vacancy when told the company had backing from a top-tier investor, and investor reputation also made candidates more likely to begin an application. Customers, too, may be more willing to try an unfamiliar product from a company someone credible has already vetted.

This matters quite a lot at the earliest stages, when a company has almost nothing verifiable to point to. An accelerator’s brand can stand in for the track record the company doesn’t yet have.

From there, a feedback loop can take hold. Admission attracts investor attention, investment funds faster hiring and experimentation, and the resulting metrics make the accelerator’s original judgment look prescient. The program may have contributed little in the way of technical insight, but its endorsement changed the environment the company grew up in.

While endorsement typically lands on every company the same way, the educational aspect of the program may not.

Why the Same Incubator Program Affects Founders Differently

Founders arrive at incubators and accelerators with different amounts of experience, different access to capital, and companies at different stages of development. Some benefit from broad structure, while others need support that is narrow and specific to what they’re building.

Research in the Strategic Management Journal found that these differences shape program outcomes. Founders with more prior knowledge tended to get more from specialized accelerators, while less experienced founders often did better in generalist programs.

Programs manage this mismatch mainly through selection. This is one reason so many accelerators organize around a theme, whether fintech, climate, or biotech. By admitting companies from one industry, a program constructs a cohort whose collective knowledge runs deep rather than broad, and the cohort is a large part of what any participant receives. Founders learn from the companies around them, so who else is in the batch changes what the program can teach.

This design choice carries a measurable tradeoff too. A study of 8,580 startups across 408 accelerators found that cohorts with deeper peer knowledge predicted stronger revenue growth but weaker ability to attract external capital, and that longer development periods amplified the benefits of these knowledge-building elements. Research on corporate accelerators points the same way, with specialized programs improving speed to market and growth while making follow-on venture funding somewhat harder to secure.

Even so, no selection process produces a uniform cohort, and standardized support still helps some companies while distracting others. Weekly targets and investor rehearsals can be exactly what a first-time founder needs. The same routine costs an experienced founder time they could spend elsewhere. 

Deep-tech and regulated companies face a similar version of the problem, since a standard curriculum can push them toward commercial milestones before their technical or regulatory uncertainty is anywhere near resolved. A therapeutics company can be coached toward customer discovery when its actual gating risk is a preclinical result.

Every accelerator carries an implicit timetable for progress, and speed is only useful when it shortens the route to valid evidence.

The Psychology of Compressed Progress

Accelerators change founder behavior through structured pressure and through comparison with the other companies in the cohort.

These conditions can improve execution, and they also shape which forms of progress founders learn to value. Leon Festinger’s theory of social comparison argues that people assess their own performance by measuring it against others. Inside a cohort, founders quickly work out which companies receive praise and which metrics attract investor attention. This can create discipline, and it can also teach founders to present activity as if it were evidence.

The second force at work is the difficulty of letting go. Decision researchers call it “escalation of commitment,” the tendency to keep investing in a failing course of action precisely because so much has already been invested. Founders are unusually exposed to it. They have poured savings and years into the company, their identity is tangled up with it, and quitting can feel like a verdict on themselves rather than on the idea. Even venture capitalists, who are paid to cut losses, show measurable escalation patterns when reinvesting in deteriorating companies.

This is why an outside voice is important. A strong accelerator should help founders recognize when a company’s central premise has failed, something the founders themselves are psychologically positioned to see last. Baek and Hegde’s 2026 study found that high-value-added accelerators were more likely to accelerate the shutdown of weaker ventures. Earlier closure spares founders from committing more time and capital to a company whose prospects have already deteriorated, and it returns their most limited asset, their own working years, to better uses. 

Outside voices are only useful in the right dose, though, and programs are structurally inclined to over-prescribe them.

When Advice Becomes Interference

More mentorship does not guarantee better decisions. Founders must judge whether each adviser understands the company and whether their experience applies, and conflicting recommendations can consume time and make the business increasingly reactive.

The pattern is common enough that accelerators have a name for it. Research in Administrative Science Quarterly documented what one program called mentor whiplash, where advice from different mentors pulls a company in opposite directions at nearly the same time. One founder described knowing exactly what his company was doing before the program and having no idea a week in. The same study found the problem has a structural remedy. 

Ventures that received their consultation concentrated into a compressed period developed better business models and strategies, because a dense stream of conflicting opinions can be triangulated in a way that a slow trickle cannot. The confusion, in other words, is a stage to pass through rather than a signal to stop, but only when the program is designed for it.

Not all forms of support pull their weight either. A UK government study of accelerators and incubators found that few forms of support were consistently associated with positive outcomes across the board.

 Direct funding and help with team formation showed some of the broadest positive relationships. Advice from consultants and business developers was negatively associated with certain growth measures, while mentoring from investors or industry experts produced mixed results.

The report also found that founders’ perceptions did not always match measured outcomes. Access to customers and partners was highly valued but showed no clear relationship with the performance indicators tested. Support with business-model development was linked to weaker short-term growth and investment outcomes, possibly because founders redirected effort toward finding a more viable model.

Short-term underperformance can therefore reflect a useful correction. A company that abandons the wrong customer segment may lose revenue before improving its long-term prospects.

Programs judged primarily on immediate results may discourage such changes. A better evaluation would examine whether the accelerator improved consequential decisions, even when those decisions reduced visible progress during the cohort.

The Bad Equilibrium

Accelerator and incubator programs become harmful when they reproduce the visible features of successful programs without the underlying capability. Mentor directories and workshops are easy to copy, and so is demo day.

Effective support requires operators to understand each company well enough to identify its most consequential uncertainty. They must then connect the founder with advice or resources that can change the relevant decision. Without that diagnostic capacity, the program becomes a series of activities that may feel productive without improving the company. The distinction is hard to observe from outside, and early data from the Global Accelerator Learning Initiative points the same way, finding that the quality of a program’s partners and applicants predicted performance while mentor rosters and cohort networking showed much weaker support. The features that matter are precisely the ones a brochure cannot show.

In Akerlof’s market for lemons, sellers know the quality of what they’re offering, and buyers cannot verify it before paying. Accelerator admission has the same structure. Programs know far more about previous cohort outcomes than applicants do, and founders usually see selected alumni and aggregate funding figures rather than the experience of the median participant. A program can therefore look credible while offering little measurable benefit to most of the companies that pass through it.

The market is slow to correct this for several reasons. Outcomes take years to materialize, by which point the program has run many more cohorts. Attribution is genuinely difficult, since a program’s best results may owe more to its selection than to anything that happened inside it. And a meaningful share of programs are funded by governments, universities, or corporate sponsors who measure activity and cohorts launched, rather than what became of the companies. Supply persists whether or not the product works.

The cost to founders can still be substantial. They may surrender equity or pay fees while diverting months from customer work. One successful alumnus can support years of marketing, and the accelerator retains the reputational value of that association even when its contribution was limited. The founder bears a different cost, since equity transferred early becomes progressively more valuable if the company succeeds.

There are partial defenses. Founders can ask for the full list of companies from the last three cohorts rather than the showcased ones, and talk to the ones that didn’t raise. They can ask what the program believes their company’s binding constraint is and judge whether the answer reflects any real diagnosis. A program that cannot name what it would change about a specific company is describing a curriculum, and a curriculum is the part that was easy to copy.

The exchange can be worthwhile when the program provides capital and improves important decisions, and its endorsement may strengthen the company’s position with investors. Those benefits remain uncertain when the founder signs the agreement, while the ownership cost is fixed.

How to Tell Whether a Program Adds Value

Headline outcomes reveal little about an accelerator. Portfolio valuation, funds raised, and acquisitions blend the quality of the companies admitted and the value created after admission. A program’s marketing has every incentive to leave the two entangled.

The way to separate them is to ask the counterfactual question. Not what alumni achieved, but what they would have achieved anyway. This is how the more careful research in this field works, comparing admitted companies against similar applicants who just missed the cut, and a founder can run an informal version of the same test. Look at the strongest companies in a recent cohort and ask honestly whether they were already strong when they arrived. A program whose successes were visibly succeeding beforehand is demonstrating its selection, which benefits the program more than it benefits the next applicant.

A more informative question is what changed during the program. Did founders resolve a critical uncertainty earlier than they otherwise would have? Did they make better decisions? Did the program connect them with people who materially altered the company’s trajectory? These outcomes are harder to measure, but they are closer to the intervention itself.

Alumni behavior is one of the more honest signals available, because it reflects what people do rather than what anyone claims. Do former participants return as mentors or invest in later cohorts? Do second-time founders, who have options and know exactly what the program was worth the first time, come back or send their peers? Repeat engagement from the people with the best information is difficult to fake.

So is a program’s relationship with its own failures. An accelerator that talks only about its winners either selects perfectly or is not being straight about the distribution. A better sign is a program that can discuss the companies that shut down and what the program did once the premise had failed. Given the evidence that strong accelerators wind down weak ventures faster, a portfolio with no closures is less reassuring than it looks.

Founders can also probe the diagnostic capacity directly. Ask for the full list of companies from the last few cohorts rather than the showcased ones, and talk to some that didn’t raise. Ask what the program believes your company’s binding constraint is, and judge whether the answer reflects any real engagement with the business. A program that cannot name what it would change about a specific company is describing a curriculum, and a curriculum is the part that was easy to copy. The same applies to the program’s own account of itself. If what it expects to change is limited to visibility, community, or fundraising opportunities, it may be describing an experience rather than an intervention. Strong accelerators can explain which founder decisions they aim to improve and which uncertainties they are designed to resolve.

Equally important is recognizing when acceleration is the wrong tool. A company waiting for clinical trial data or semiconductor fabrication is unlikely to benefit from being pushed toward milestones designed for software startups. A program that understands its own limitations is often more credible than one claiming to accelerate every business.

Did Y Combinator Make Airbnb?

Returning to where this article began, the case shows selection and intervention working together. Y Combinator identified founders with unusual determination, and that same quality made them more likely to act on uncomfortable advice. The program then redirected their attention from scalable solutions to the actual constraint, put credible investors in front of them, and lent the company its endorsement.

While the advice sounded obvious, it was valuable at that moment. It came from someone the founders had reason to believe, and it resolved the company’s most consequential uncertainty. In addition, the program’s endorsement changed how investors read a business they had previously dismissed.

Airbnb’s founders did the work, but Y Combinator changed the conditions around it. Without the founders, the advice to visit New York would have meant little. Without the program, they may have kept solving the wrong problem until the money ran out.

The good incubators and accelerators are therefore neither fortune-tellers nor startup factories. They are systems for recognizing potential and changing the way founders confront their uncertainties.

The bad ones pick winners and think they made them.

The ugly ones know they didn’t and say it anyway.

Continue the Conversation

At Arcanum Ventures, we spend our days with founders working through exactly these kinds of decisions, usually well before it’s clear how things will turn out.

We share what we learn through our podcasts, videos and original writing. Join us as a speaker, tune in, or keep pace with what’s moving across tech.

If you are building something difficult and want a second set of experienced eyes, we would love to hear from you.