Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Fused Large Language Model for Predicting Startup Success

Abdurahman Maarouf, Stefan Feuerriegel, Nicolas Pröllochs

TL;DR

This paper tackles predicting startup success by leveraging both structured fundamentals and unstructured textual self-descriptions from VC platform profiles. It introduces a fused large language model that uses BERT-based embeddings for text together with traditional variables, and evaluates multiple final classifiers under a time-aware, out-of-sample framework on Crunchbase data. The results show that textual self-descriptions add significant predictive power, achieving an AUROC around 0.83 and a 7.23x ROI when combined with fundamentals, with robustness across sectors and events. Practically, the approach provides investors with a data-driven screening tool that enhances decision-making while illustrating the operational value of large language models in business analytics.

Abstract

Investors are continuously seeking profitable investment opportunities in startups and, hence, for effective decision-making, need to predict a startup's probability of success. Nowadays, investors can use not only various fundamental information about a startup (e.g., the age of the startup, the number of founders, and the business sector) but also textual description of a startup's innovation and business model, which is widely available through online venture capital (VC) platforms such as Crunchbase. To support the decision-making of investors, we develop a machine learning approach with the aim of locating successful startups on VC platforms. Specifically, we develop, train, and evaluate a tailored, fused large language model to predict startup success. Thereby, we assess to what extent self-descriptions on VC platforms are predictive of startup success. Using 20,172 online profiles from Crunchbase, we find that our fused large language model can predict startup success, with textual self-descriptions being responsible for a significant part of the predictive power. Our work provides a decision support tool for investors to find profitable investment opportunities.

A Fused Large Language Model for Predicting Startup Success

TL;DR

This paper tackles predicting startup success by leveraging both structured fundamentals and unstructured textual self-descriptions from VC platform profiles. It introduces a fused large language model that uses BERT-based embeddings for text together with traditional variables, and evaluates multiple final classifiers under a time-aware, out-of-sample framework on Crunchbase data. The results show that textual self-descriptions add significant predictive power, achieving an AUROC around 0.83 and a 7.23x ROI when combined with fundamentals, with robustness across sectors and events. Practically, the approach provides investors with a data-driven screening tool that enhances decision-making while illustrating the operational value of large language models in business analytics.

Abstract

Investors are continuously seeking profitable investment opportunities in startups and, hence, for effective decision-making, need to predict a startup's probability of success. Nowadays, investors can use not only various fundamental information about a startup (e.g., the age of the startup, the number of founders, and the business sector) but also textual description of a startup's innovation and business model, which is widely available through online venture capital (VC) platforms such as Crunchbase. To support the decision-making of investors, we develop a machine learning approach with the aim of locating successful startups on VC platforms. Specifically, we develop, train, and evaluate a tailored, fused large language model to predict startup success. Thereby, we assess to what extent self-descriptions on VC platforms are predictive of startup success. Using 20,172 online profiles from Crunchbase, we find that our fused large language model can predict startup success, with textual self-descriptions being responsible for a significant part of the predictive power. Our work provides a decision support tool for investors to find profitable investment opportunities.
Paper Structure (38 sections, 2 equations, 3 figures, 15 tables)

This paper contains 38 sections, 2 equations, 3 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Venture Capital
  4. Predicting Startup Success
  5. Machine Learning in Business Analytics
  6. Empirical Model
  7. Research Question
  8. Fused Machine Learning Approach
  9. Large Language Model (BERT) as embedding generator.
  10. Baseline text representations.
  11. Final Machine Learning Classifier.
  12. Performance Metrics
  13. Implementation Details
  14. Empirical Setting
  15. Online Profiles on Crunchbase
  16. ...and 23 more sections

Figures (3)

  • Figure 1: Our machine learning approach based on a tailored, fused large language model for predicting startup success.
  • Figure 2: Schematic overview of a large language model (here: BERT).
  • Figure 3: Left: Summary plot of the SHAP values for our fused large language model predictions. Each dot represents a SHAP value for a feature across different samples, where color indicates a high (red) to low (blue) feature value. Right: Bar plot of the mean (absolute) of SHAP values across all samples. The features are sorted by their feature importance.