The short version
From a range of sources, we assemble a set of potential websites for each company. We call these candidates. We then scrape up to 25 pages of each candidate website. Finally, we apply a machine learning model to select the best candidate for each company. This model is trained on manually checked website matches for thousands of companies collected over the last decade. We measure the success of our model continuously. The two most important metrics are:- Precision: how often we pick the correct website, or no website (regardless of whether the company does actually have a website).
- Accuracy: how often we pick the correct website, or no website if the company does not have one.