Why we predict successful startups?

play_arrowPredicting Successful Organizations

Our Vision

Why we predict successful startups?

The impact of startups on the world’s economy shows no sign of slowing down, as the creation of these companies continues to boom. This has derived in a significant increase within the Insurtech eld, surprisingly so, when taking into account an unprecedented situation such as this year’s COVID-9 crisis. As described by Deloitte’s 2020 Insurance M&A Midyear Update, “Insurtechs may have a greater willingness to partner or be acquired due to the uncertain economic outlook, and carriers seeking to quickly respond to a rapidly changing world (such as an increased need for virtual or touchless claims) may find Insurtech investment a more appealing allocation of capital than investing in the current low-interest-rate environment”.

For these companies or individuals looking to invest in startups, having a vast knowledge on the scope around which these enterprises are developed is a key requirement. Along those lines, last year we developed a small study inspired by the Thesis by Francisco Ramadas da Silva Predicting Startup Success with Machine Learning, that, throughout the appliance of data mining and machine learning technologies, implemented a predictive model that classified whether a startup was likely to be successful or not, using data related to Insurance startups extracted from the webpage inttrend.com.

That model was tested on the Insurtech dataset, collected by the Insurance staff, and insightful information was extracted from the predictions. Not only did it achieve a True Positive Rate (TPR) of 70% and a False Positive Rate of 6%, but it also predicted a list of unsuccessful startups at that moment that were to be successful in the future. Since that study was launched, its consistency has been empirically verified; at least two of the startups included in the predicted shortlist have received an Initial Public Offering (IPO): US car insurance company Root Insurance Co. and digital property/casualty insurance marketplace CoverHound.

This year, continuing on from the previous study, a similar analysis has been carried out. However, several modifications, along with an updated and extended data sample, have been introduced as to enhance its robustness. Firstly, the sample size has been significantly increased (the dataset has been more than doubled and not only data resulting from Insurtech has been implemented to train the model). Furthermore, we have broadened our definition of what it means for a startup to be successful, and have also added a detailed category analysis and classification of these companies, including information on their founders.

For these companies or individuals looking to discover relevant startups or business models, having a vast knowledge on the market is not enough. Creating a specific knowledge on the key variables in which these enterprises are developed is a key requirement.

Richard Calvo. Head of Insurtech