Using AI To Predict The Success Of Startups

Startups have a notoriously high failure rate, which can make investing in them an extremely risky business.  Research from Santa Clara University’s Leavey School of Business proposes using AI to lessen the risk and better predict whether a startup might succeed or fail.

The researchers trained a machine-learning model on over 1 million companies and they believe it was eventually able to accurately determine whether a startup would thrive or fail.  Indeed, the researchers argue that their model has an accuracy of up to 90%.

“This research shows how ensembles of non-linear machine-learning models applied to big data have huge potential to map large feature sets to business outcomes, something that is unachievable with traditional linear regression models,” the researchers say.

Picking winners

A large part of the team’s success was due to their ability to develop an ensemble of models that were able to work together, thus outperforming any one model on its own.  Each model worked to try and classify the startup and assign it a probability of success or failure.

For instance, the models might predict that a startup has a 75% chance of being acquired or achieving an IPO, whereas it has just 25% of failing.

The models were trained on data gained from the Crunchbase platform, which has detailed information on startups from around the world.  This data was combined with information gathered from the US Patent and Trademark Office’s patent database.

Suffice to say, the crowdsourced nature of the Crunchbase database meant that information was missing on many startups, but the researchers were able to use this in their models.  Indeed, it transpired that this was one of the key aspects of accurately determining whether the business would succeed or fail.

“The models generate a level of accuracy, precision and recall that exceeds other similar studies. Investors can use this to quickly evaluate prospects, raise potential red flags and make more informed decisions on the composition of their portfolios,” the researchers conclude.

Facebooktwitterredditpinterestlinkedinmail