Business domain information can be a good indicator for predicting the success of startups

Startup companies have attracted increasing attention in recent years. Predicting success of startups is eagerly anticipated by policymakers aiming to drive innovation, by companies aiming to enrich their business through acquisitions, and by venture capitalists looking for high returns. Here, the difficulty of the prediction lies in the “newness” of startups. They usually have no tangible assets and/or publicly available information on which we can rely for prediction. How to measure their intangible assets, therefore, is the key challenge. Many attempts have been made, including [1] which successfully identified valuable startups by applying a centrality index to the employee mobility network. However, obtaining empirical information about employees’ mobility (or other personal/confidential attributes) is not straightforward, resulting in the potentially critical limitation of the research scope. The most basic (and non-confidential) data items available in any corporate databases are the company’s name, location, size, and business domain(s). In this study, we shed light on companies’ business domains and attempt to predict startups’ success by only using the information of how companies combine multiple domains together to form their business portfolios. We used data extracted from Crunchbase [2] with the support of our collaborator, Zuva Inc. In the data, all the companies’ businesses are described by one or the combination of about 700 business domains. We created a bipartite network of companies and business domains, and calculated the degree and average nearest neighbor for each company node. Using these values, we can position a given company’s business along the axes of diversity (degree) and ubiquity (average nearest neighbor), as shown in Figure 1. (This is an application of the method proposed in [3].) Each of the cells C-1 to C-16 in the heatmap (Fig.1) indicates the number of successful companies / the total number of companies, and the calculated success rate. The cell color also corresponds to the success rate. Here, we considered a company having succeeded if it achieved one of the following conditions within seven years after its foundation: (i) having been acquired by others, (ii) underwent an IPO, or (iii) acquired other companies. The figure suggests that companies with low ubiquity (C-13 to 16), i.e., those operating rare business(es) tend to have a higher chance of success. We further investigated the three different success types (i - iii) separately, and found that, when the diversity value is high (e.g., C-15), most companies succeeded by (iii) becoming capable enough of acquiring others. When the diversity value is low (e.g., C-13 and 14), successful companies were mostly (i) acquired by others. Then (ii) IPOs were found to be more common when both ubiquity and diversity were high (e.g., C-7, 11 and 12). We also investigated the temporal transition in these trends. This study demonstrates the potential usefulness of business domain information as a good indicator for predicting startup success. Improving the quality of business domain data is the obvious next step. We are currently working on it by focusing on company descriptions, which is another type of easily obtainable information.