Using Artificial Intelligence to Unlock Crowdfunding Success for Small Businesses

TL;DR

The paper addresses the high failure rate of small-business crowdfunding campaigns by introducing an AI-driven pipeline that combines a LightGBM predictor with 168 features, including 11 GPT-derived textual cues, to forecast outcomes and reveal actionable narrative factors. It validates the approach with counterfactual GPT-4 revisions and a randomized online experiment, finding that textual features drive roughly $80\%$ of predictive power and that AI-generated narrative refinements can raise funding probability by about $11.9\%$ on average and are highly preferred by potential donors ($83\%$). These findings demonstrate a feasible, equity-conscious AI co-pilot for crowdfunding, enabling small businesses to optimize descriptions before launch and potentially reduce funding inequities across regions and groups.

Abstract

While small businesses are increasingly turning to online crowdfunding platforms for essential funding, over 40% of these campaigns may fail to raise any money, especially those from low socio-economic areas. We utilize the latest advancements in AI technology to identify crucial factors that influence the success of crowdfunding campaigns and to improve their fundraising outcomes by strategically optimizing these factors. Our best-performing machine learning model accurately predicts the fundraising outcomes of 81.0% of campaigns, primarily based on their textual descriptions. Interpreting the machine learning model allows us to provide actionable suggestions on improving the textual description before launching a campaign. We demonstrate that by augmenting just three aspects of the narrative using a large language model, a campaign becomes more preferable to 83% human evaluators, and its likelihood of securing financial support increases by 11.9%. Our research uncovers the effective strategies for crafting descriptions for small business fundraising campaigns and opens up a new realm in integrating large language models into crowdfunding methodologies.

Figures (7)

• Figure 1: Overview of the Distribution and Funded Rates of Small Business Crowdfunding Campaigns. The left panel (a) displays the geographical distribution at the county level with the size of circle representing the total number of campaigns and the color describing the average percentage of funded campaigns. The right panels (b) and (c) showcase the relationship between local socio-economic status and the likelihood of receiving funds at the city level. (b) indicates that the percentage of residents' with higher education in a city is positively related to the funded probability. Similarly, (c) suggests a positive correlation between the city median household income and the percentage of campaigns receiving funds.
• Figure 2: Summary and Examples of Feature Importance. (a) displays the proportion of importance for each group of features in the LightGBM prediction model. (b) showcases the average marginal effect with delta-method (SE in error-bars), where features in the same group are depicted with the same color. This figure indicates significant influence of textual features: they account for 80.33% of the decisive power among all feature groups, including local demographics, pandemic shock, and GoFundMe campaign configurations.
• Figure 3: Predicted Funding Probability of Original and GPT-augmented Campaign Descriptions in Offline Simulation. Subplot (a): the average predicted funding probability of all 500 campaigns significantly increased from 33.18% to 45.10%, with the funding probability of the originally unfunded campaigns rising from 25.09% to 41.13% (with error bars representing standard errors). Subplot (b): a scatterplot comparing the predicted likelihood of funding before and after ChatGPT augmentation of all 500 campaigns. Originally unfunded campaigns received a higher boosting in the likelihood of funding than originally funded campaigns.
• Figure 4: Proportions of participant preferences between pairwise comparisons of GPT-augmented, GPT-extended, and original campaign descriptions, with error bars representing standard errors. 83% (or 82%) participants preferred GPT-augmented campaigns over the original (or GPT-extended) version, and GPT-extended campaigns are favored by 61% participants when compared to the originals.
• Figure S1: Demographic disparities in crowdfunding outcome. Subplots (a) to (e) depicts the relationships between city demographics and the percentage of funded campaigns in the city. Specifically, (a) shows that campaigns from regions with a higher percentage of individuals under 18 years old are less funded. (b) indicates that small businesses in areas with a greater proportion of foreign-born citizens tend to achieve higher funded percentages. (c) reveals that areas with more African American residents see lower funding success for small businesses. (d) demonstrates that businesses in densely populated areas are more likely to receive funding. (e) suggests that lower funding rates are observed in areas with more population in poverty. (f) describes the percentage of funded campaigns for male and female campaign organizers (at the campaign level) and uncovers that campaigns organized by female are less likely to be funded, highlighting a gender disparity.