Most accelerators make atartups worse. What do the good ones do?

Every accelerator makes a version of the same offer: capital, mentorship, a network, three months of support, and materially better odds of survival. Evidence suggests that little of it actually works.

In April, Youn Baek and Deepak Hegde of NYU Stern published a working paper through the National Bureau of Economic Research examining nearly 750,000 American startups across 329 programs. Between 60 and 80 percent of accelerators, they found, leave the companies that join them worse off than if they had never applied. A smaller group does the opposite, raising funding, growth and exit rates by a wide margin. Among them, Y Combinator, Techstars and Endless Frontier Labs.

The study establishes which programs work, but it does not explain why. For that, we asked founder and product-market fit expert Yann Goarin.

Goarin spent a decade at Google and YouTube, where he launched more than twenty products in Europe and the United States, and has since led product and marketing at several venture-backed startups. He founded Zag Labs in 2023, an advisory firm that has helped more than a hundred early-stage companies go to market and accelerate their path to product-market fit. He developed the “PMF System”, a method that treats product-market fit as a problem-solving process rather than an event or a vibe. He is currently Founder in Residence at AAXIS, where he leads the enterprise technology firm’s venture-building work. He also mentors and judges at five accelerator programs across the US (Techstars, gener8tor, FoundersBoost, Expert Dojo, and USC’s Iovine and Young Academy), which gives him a unique perspective on how different programs support their founders.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

Most accelerators take equity in exchange for a check and three months of support, and their return depends on whether a few companies in each cohort raise at scale or exit. What they offer founders is leverage in several forms: capital, introductions to investors and customers, brand recognition, and knowledge.

Like top universities, the best accelerators attract and select the best founders. Even so, the odds of success are very low. Building a category-defining, venture-backed company is incredibly difficult, and luck and timing decide a great deal of it. But it is not magic. There is a method to the madness, and that method, Goarin claims, is either not taught or not taught well.

Research shows that knowledge is the form of leverage that appears to matter most. Susan Cohen, Benjamin Hallen and Christopher Bingham, who spent years studying the original American accelerator programs, found that where accelerators do improve their companies, the primary driver is what those companies learned inside them. But it is also the hardest to scale.

Goarin remembers one client engagement, a seed-stage AI startup that had built a video production platform. Its founders had come through one of the world’s most selective accelerators. It raised $4 million and within twelve months passed $1.2 million in annual recurring revenue. However, churn was running above 30 percent. The response was to sell harder and build faster, adding features as customers asked for them, and investors supported that on the view that revenue was the number that mattered most.

What the founders failed to realize was that the three segments they were selling to (small marketing agencies, independent video creators, and boutique production companies) were not a cohesive market. While they appeared to need faster and cheaper video production, they differed in how much video they produced, how polished it had to be, how it fit in their workflow, and where it was distributed. The product tried to stretch across all three, and served none of them well. Customers left faster than sales could replace them. After cutting half the team and pivoting, they failed to secure a bridge round and ran out of runway.

Goarin came in near the end, too late to change the outcome. “I assumed that founders coming out of a program like that would be better at testing their assumptions and diagnosing their issues. I was wrong. They were just as clueless as most of the others I advise.”

Around that time he started mentoring at Techstars. That’s where he saw an opportunity to address the problem at scale. From inside a program, it becomes clear how knowledge actually reaches founders, and what never does.

The programs that do teach tend to teach in fragments: a product expert teaches product, a sales executive covers sales, someone who has raised four rounds helps with fundraising. Founders are expected to assemble them into a working company. Most fail. There is something odd in that, viewed from outside. Accelerators and venture funds spend enormous effort on selection, screening thousands of applicants to find the few worth backing, and then just hope they figure it out.

What goes untaught is product-market fit itself, i.e., the correct assembly of these fragmented pieces that ultimately leads to widespread demand for something people badly need, delivered profitably every time. There are two reasons it does not appear on syllabuses. Product-market fit is not understood as a discipline in its own right, so there is no settled body of practice to teach from. And the mentorship model recruits subject matter experts by function, so PMF, which sits between and over the functions, isn’t owned by anybody. Until now.

What Goarin teaches in these programs runs end-to-end, and his objective is straightforward: avoid building something nobody wants.

“Accelerators give founders access and funding, and of course that matters,” says Goarin. “But where they can have an even bigger impact is teaching first-time founders to operate like second-time founders. That means going beyond the surface-level material and breaking down the mechanics of startups.”

User experience went through the same thing. Usability testing, information architecture and interaction design were practiced separately for years before the field recognized them as one discipline and created roles for people who worked across all of them. Naming it is what made it possible to teach.

The case for teaching product-market fit as its own subject is getting stronger. As technology levels the playing field on building and execution, what separates companies is judgment: Is this problem worth solving? Is this the right customer segment? Can I deliver my solution repeatably and profitably? Is it time to pivot? None of those questions can be answered well without knowing what to look at, and that is what Goarin focuses on.

“In the early days only three things matter,” Goarin claims. “Speed of learning, speed of decision-making, speed of execution. A startup is a learning machine before it is anything else, and learning is the part founders struggle with the most. Building is fast and cheap now, so the temptation is to ship something and see if it sticks. But that’s how you end up with a product in search of a problem. That’s how you end up in pivot hell.”

His work has been expanding. He was a Lead Mentor at Techstars for the Spring 2025 and Spring 2026 cohorts and a judge in Mentor Magic, the program’s week of back-to-back mentoring and evaluation sessions. He has advised two gener8tor cohorts and judged USC’s Venture Showcase. He is in discussions with other top programs in the United States and Europe.

Top accelerator entry requirements have been rising. Joshua Lu, who runs Speedrun, told TechCrunch this year that because AI has made building and testing so much faster, the program now expects market validation or early traction before it will admit a company. That created a new market of programs beneath the accelerators. The best of them are focusing on education, and have invested accordingly. FoundersBoost, one of the world’s best pre-accelerators, brought Goarin in to strengthen its programming and asked him to teach its last two cohorts.

The gap is about to matter more. AI is accelerating a trend already underway, in which smaller and smaller teams, working alongside swarms of agents, can perform like much larger companies. That does not reduce the value of knowing what to build. Rather, it concentrates it. Judgment, pattern recognition, knowing what to focus on and when, the confidence to make a decision and move: these have always been the unfair advantage, but are ever more critical in the AI age.

“Fundraising used to be something most founders didn’t understand,” says Goarin. “Now every program teaches it. Product-market fit is more complex, but it is a subject, and I expect it will be taught the same way before long.”

Baek and Hegde could not say what separates the accelerators that work from the ones that do not. If the answer is what they teach, the programs that work it out first will be the ones worth applying to.