Crunchbase’s Prediction Models and Performance

Crunchbase’s Predictions provide predictive company intelligence, helping you anticipate what opportunities or risks will happen next in the private market. Now you can get actionable, forward-looking conclusions without the manual analysis.

With Crunchbase’s Predictions and Insights API, you can access the following prediction models: 

Funding prediction model 

This model predicts if a private company is likely to have a fundraising event and the timeframe when that funding might happen. For more information about these data fields, refer to this guide: Funding Predictions

Exit prediction models 

This model predicts if a company is likely to experience an exit event. Specifically the model predicts the likelihood for companies to be acquired, IPO, or to close. For more information about these data fields, refer to the following guides: Acquisition Predictions, IPO Predictions, and Closure Predictions in the API

Growth prediction model 

This model predicts the future growth trajectory of a private company. For more information about these data fields, refer to this guide: Growth Predictions

Layoff prediction model 

This model predicts the likelihood of a layoff event happening at a private company. For more information about these data fields, refer to this guide: Layoff Predictions 

Probability Tiers

Crunchbase’s predictions are delivered using probability tiers to best describe the likelihood for one of these events to occur. Those tiers are as follows: 

• Very Likely: with a probability score ranging from 0.95-1.00
• Probable: with a probability score ranging from 0.66-0.95
• Uncertain: with a probability score ranging from 0.36-0.65 
• Doubtful: with a probability score ranging from 0.06-0.35
• Very Unlikely: with a probability score ranging from 0.00-0.05

Crunchbase has rigorously tested and seeks to continuously improve these models so you can integrate and depend on Crunchbase’s Predictions. To gauge and understand the model’s performance we rely on their precision and recall metrics.

Performance 

What exactly do these performance metrics mean? 

• Precision confirms out of everything we predicted as positive, how often we were correct. With high precision, it means we have few false positives. 
• Recall tells us, out of all the real-world positives, how many we actually caught. With high recall, it means we caught most of the positives.

Let’s review our precision and recall across each predictive model.

• How does Crunchbase’s funding prediction model perform? 
  • When predicting if a fundraising event will occur, the model excels, achieving 95% precision and over 99% recall on our test data. This means the model missed less than 1% of funding events that occurred on the test data.
    
    
• How does Crunchbase’s exit prediction model perform? 
  • The exit model provides predictions for three different exit events: acquisition, IPO, or company closure. 
    • When predicting an exit event, specifically acquisitions, the model demonstrated strong performance, achieving 96% precision and 95% recall on our test data. 
    • When predicting IPO events, our model provides meaningful signals, achieving 59% precision and 54% recall on our test data. 
    • Predicting company closure events is incredibly difficult to do as early warning signs (like undisclosed cash flow issues) are very limited or market downturns are unpredictable. With these real-world complexities in mind, the model achieved 18% precision and 40% recall on the test data when predicting company closures.
      
      
• How does Crunchbase’s growth prediction model perform? 
  • The model predicts company growth with 68% precision, capturing 81% recall. Our growth model detects 81% of companies in our test data showing strong signals of expansion.
    
    
• How does Crunchbase’s layoff prediction model perform?
  • Layoff events can be challenging to predict, but Crunchbase’s model captures early indicators of layoffs from the test data, with 23% precision and 27% recall. 
  • Crunchbase’s layoff prediction model is highly accurate at identifying companies unlikely to conduct layoffs from the test data, achieving 96% precision and 95% recall.   

With actionable and predictive company intelligence from Crunchbase’s Predictions, users get the first-mover advantage to private market opportunity. You can find Crunchbase Predictions and Insights on crunchbase.com or the API.