Machine learning reveals factors for successful crowdfunding

Modern crowdfunding has grown from relatively humble beginnings in the late 1990s to a multi-billion dollar market for funding all sorts of early innovations. The platform Kickstarter alone went from $276 million pledged in 2012 to $7.8 billion in 2024. There are even professional project designers to help craft that winning proposal.

With these types of interests, it is of the utmost importance that the pitch is good.

Machine learning helps with that. Researchers at the University of Toronto’s Rotman School of Management tested four different types of this artificial intelligence application, including Deep Learning. Machine learning not only outperformed conventional statistical methods in predicting whether a crowdfunding campaign would reach its goal, it also identified which elements helped the most and how.

The findings were published in the Journal for Business Venturing Design.

“Running a crowdfunding campaign is expensive and can fail,” says Ramy Elitzur, a professor of accounting at the Rotman School. “Our analysis shows project creators how to increase their chances of success, or whether they should pursue a different project financing strategy.”

Kickstarter is an all-or-nothing platform, meaning that project creators don’t receive money unless they reach their fundraising goal. After analyzing more than 100,000 Kickstarter projects, Prof. Elitzur, Prof. David Soberman, Canada’s national chair of strategic marketing, and other researchers found that the size of the campaign’s financial goal accounts for more than half of a project’s success. The creator’s social capital, the number of reward options offered, and the length of the campaign were also important factors.

Machine learning also drilled down into the details of how much and for how long. The probability of a project’s success remained fairly constant up to a fundraising goal of $100,000, but then began to decline, with a sharper drop above $133,300. However, standard conventional regression models predict that the probability of success drops consistently as the monetary goal increases. That’s because these models tend to identify “linear” relationships, where influencing factors and outcomes move in only one direction.

Crowdfunding, like many things, is more complex, with multiple variables interacting with each other and the outcome.

“One of the things that machine learning does is model all the possible interactions between variables,” says Prof. Elitzur. “It gives us the direct effect on the outcome of each variable and the total effect of its interaction with other variables.”

Standard regression showed that success increased as social capital increased (measured by the number of comments a project generated), but machine learning showed that success actually increased up to around 750 comments and then leveled off.

The results also showed that the optimal campaign duration was 10 to 15 days and that the number of reward options for projects had a moderately positive effect on success up to about 15 reward options, then a slightly negative effect between 15 and 20 options, followed by a positive effect in waves between 20 and 50 reward options, and finally a plateau after 50 options.

When the textual analysis capabilities of machine learning were put to use (something standard numerical methods cannot do), they were able to extend beyond Kickstarter’s top 15 project classifications and identify “gadgets” as the least successful project type.

It turns out that creators looking for that winning proposal should stay away from flyable World War II aircraft. In that case, “the cards are stacked against you and you have a lower chance of success than in any other domain,” says Prof. Elitzur, who is currently applying the same methods to predict the success of high-tech startups.

Furthermore, the textual analysis capabilities of the models show that, as with real estate, the location of the project matters.

The study was co-authored by Noam Katz of Ben Gurion University in Israel and Perri Mutath of The Israel Innovation Authority.