Science reveals the tipping point between success and failure

From Thomas Edison to Silicon Valley startups, those that rise to the top have learned how to treat past failures.

Science reveals the tipping point between success and failure
[Photo: Martin Sanchez/Unsplash]

Almost every winner starts a loser.


J.K. Rowling, Henry Ford, and Thomas Edison all failed, at least initially, often disastrously. But each found their way to outsized, world-changing success. As did many others.

But what about the countless others across fields who toiled for decades only to sink into the vast ocean of “never-were?”

The reality is that only some histories of failure lead to success, begging the question of which failure patterns ultimately yield results. Conventional explanations center on luck or differences between winners and losers based on learning strategies or individual characteristics.

But it’s not so simple.

Our new research published in Nature with PhD student Yian Yin and postdoc Yang Wang at Kellogg School’s Center for Science of Science and Innovation reveals a tipping point that separates failure dynamics into “stagnation” and “progression” regions. Specifically, we show that two people who are equally lucky, learn the same amount, and stick it out for similar periods could experience fundamentally different progression patterns and outcomes—leading to breakthrough success or ongoing failure.

We find consistent empirical support for these patterns across domains including scientists, startups, and terrorist groups, with important practical implications.


Scientists, startups, and terrorists

To understand failure patterns, we took a big data approach, studying populations in radically disparate domains.

Specifically, we looked at all R01 grant applications submitted to the National Institutes of Health (NIH; the predominant funding program in the largest funder for biomedical research) from 1985 to 2015—776,721 applications by 139,091 investigators—to understand the path to science funding (or not) for individual investigators.

We also studied startup investment records from the National Venture Capital Association (58,111 startup companies launched by 253,579 innovators), including every VC-backed startup from 1970 to 2016, and whether the business had a successful exit (IPO, high-value M&A).

Finally, we looked at the nontraditional research domain of terrorism, using information from the Global Terrorism Database: 170,350 terrorist attacks by 3,178 terrorist organizations from 1970 to 2017. We tracked each organization’s pattern of success (at least one fatality) and failure among attacks.

Modeling failure

We built a simple model of how successive attempts build on one another.

Key to the model is viewing failure not as negative but offering aspirants two crucial assets: experience and feedback. Each attempt includes multiple components. A business startup requires a viable idea, revenue model, operational practices, leadership team, and others. Failure in this domain, such as a delayed IPO, sheds light on whether to reuse certain components—like sticking with the same revenue model and team—through feedback on the original components and why they failed.


Consider two extremes for how people treat past failures. At one end, people ignore all past experience, effectively starting anew each time. At the other, people sift through past failures for feedback on specific components, strategically retaining and replacing. Consequently, the latter group’s performance/efficiency systematically improves, resembling what’s known as Wright’s Law (progress improves with experience). They fail increasingly faster, refining their path to success.

The tipping point: Success or stagnation

Our model shows that how much we benefit from failures does not follow a smooth path but an abrupt one characterized by tipping points separating stagnation and success regions—similar to physical-state transitions for water.

In stagnation, people learn little from past failures of any number, such as a grant applicant who considers only the most recent proposal. Those in this regime hit an early, terminal saturation point, failing to engage in intelligent improvement. In many cases, they throw out prior attempts altogether, though some components may be on target. The stagnation region is similar to water’s physical phase of ice—although the temperature may rise from -40 degrees C to -10 degrees C (symbolizing learning from more failures), ice remains solid.

But learning from just a few more failures may lead one past the tipping point (like the abrupt 00 degrees C transition from ice to water), entering the regime of progression and success.

Here, aspirants use feedback to engage in “intelligent improvement” yielding incrementally better attempts and, ultimately, victory. Those who eventually succeed fail much faster—and increasingly fast with every attempt—reusing previous components efficiently, rather than discarding the baby with the bathwater as their stagnant peers do.

Importantly, the now-popular “tipping point” concept suggests that small changes can make a big difference. In our model, the difference between those who succeed and fail may be virtually indistinguishable regarding learning styles and productivity. Water at -1 degree C and 1 degree C has negligibly different temperatures but one is solid and the other fluid. Of two equally promising and lucky entrepreneurs, one may enjoy a wildly successful IPO while the other does not.


Our model finds that people across fields—terrorists, entrepreneurs, scientists—predictably diverge into the success and stagnation groups described here, though the groups are of similar size and initial characteristics.

Working smarter, not harder

Michael Jordan famously said, “I failed over and over and over again, and that is why I succeed.” But our results suggest success is not guaranteed. Indeed, our findings underscore working smarter, not harder. Among equally capable people, those attending to what needs to be improved are more likely to fail faster and refine components, eventually yielding success. Those that start over each time won’t get far. So aim for intelligent improvement by incorporating feedback carefully.

Moreover, our results suggest that Silicon Valley’s “fail fast” mantra is prescriptive and diagnostic. If you’re not failing faster over time, you may be stuck in the stagnation region, making misguided improvements or none at all. We found stagnation-group members worked more, not less than their successful counterparts—aiming their efforts at the wrong areas. VCs evaluating tech startups’ potential or government security agencies monitoring terrorist-group threats should also heed this insight. Groups making fast, strategic changes are more likely to have a breakthrough.

We are just beginning to understand this domain and don’t yet know all the mechanisms involved. There may be unmeasured differences between success and failure groups. We find, for example, women are more likely to follow a stagnation regime, likely due to cultural expectations.

Still, our findings have large implications for the pursuit of success. Those who are able to foster systematic, intelligent, fast improvement are much more likely to achieve victory, while others move inevitably toward the oblivion of the “also-ran.”

Dashun Wang is an associate professor at Kellogg School of Management and leads the Center for Science of Science and Innovation.


James Evans is a professor at the University of Chicago and leads the Knowledge Lab.