Crowdfunding has evolved significantly from its humble origins in the late 1990s to a formidable multi-billion-dollar financing arena for many early-stage innovations. Kickstarter, a prominent player in this field, exemplifies this growth trajectory, escalating from $276 million pledged in 2012 to an impressive $7.8 billion in 2024. This exponential growth has given rise to a new breed of professionals: expert project designers who sculpt the perfect proposal to secure funding.
The potential of machine learning to significantly enhance the chances of crafting a successful crowdfunding campaign is a beacon of hope in this high-stakes environment. A team of researchers from the University of Toronto’s Rotman School of Management, including Professor Ramy Elitzur, has explored the efficacy of various machine learning applications, including advanced Deep Learning techniques, in the realm of crowdfunding. Their studies revealed that machine learning not only outperforms traditional statistical approaches in predicting the success of crowdfunding campaigns but also pinpoints the most influential factors contributing to this success.
Professor Ramy Elitzur, an accounting scholar at the Rotman School, highlights the significant risks involved in running crowdfunding campaigns, which can often result in failure and substantial financial losses. However, through detailed analysis, the research team, empowered by the insights provided by machine learning, offers valuable guidance for project creators on how to enhance their likelihood of success or whether they should consider alternative funding strategies.
The all-or-nothing funding model, a key feature of Kickstarter, adds a sense of urgency and pressure to the fundraising process. Professor Elitzur, along with Professor David Soberman and other colleagues, discovered that the size of the monetary goal is a predominant factor, accounting for over half of a project’s success rate. Additional critical elements include the creator’s social network strength, the diversity of rewards offered, and the duration of the campaign.
Machine learning delves deeper into the nuances of these factors, such as the optimal fundraising goal and campaign duration. Their findings challenge traditional models, which suggest a linear decline in success as the financial target increases. Machine learning demonstrates that success rates remain stable up to a $100,000 goal, beyond which the likelihood of success declines, with a more pronounced decrease beyond $133,300.
The complex interplay of multiple variables in crowdfunding is better captured by machine learning, which models all possible interactions among variables. This method allows for a comprehensive understanding of each variable’s direct and interactive effects on the outcome. For instance, while traditional regression analyses show that higher social capital increases success, machine learning reveals that this effect plateaus after approximately 750 comments. Furthermore, the optimal campaign duration was identified as 10 to 15 days, and while having up to 15 reward options generally benefits a campaign, the effect becomes slightly negative as the number increases to 20 before becoming positive again with up to 50 reward options.
An innovative application of machine learning in this study is its text analysis capability, which goes beyond the capabilities of standard numerical methods. This allowed the researchers to explore Kickstarter’s 15 main project categories and identify that projects categorised under ‘gadgets’ typically had lower success rates.
One intriguing finding relates to projects that tend to underperform significantly, such as those involving Second World War aircraft, where the odds of success are notably lower than other domains. Professor Elitzur is now applying these machine learning techniques to predict the success of high-tech startups, further demonstrating the versatility and depth of this approach.
Additionally, the study underscores that a project’s location plays a critical role, similar to the dynamics observed in real estate markets. This insight into the spatial dimensions of crowdfunding success adds another layer to the strategic considerations for aspiring entrepreneurs and innovators.
More information: Ramy Elitzur et al, The power of machine learning methods to predict crowdfunding success: Accounting for complex relationships efficiently, Journal of Business Venturing Design. DOI: 10.1016/j.jbvd.2024.100022
Journal information: Journal of Business Venturing Design Provided by University of Toronto, Rotman School of Management