Does your AI use Machine Learning (ML)?

Yes. We use advanced machine learning to keep our Sifters improving.

How accurate are your Sifters?

On average, our Sifters F1 accuracy score is 95%, with our latest Sifters at 97% or higher. We use Precision, Recall, and F1 to measure the performance of our Sifters. For an in-depth look at these metrics and surrounding concepts, read this nice blog entitled, "4 things you need to know about AI: accuracy, precision, recall and F1 scores," from Lawtomated. A Sifter must hit a minimum precision score of 90%, a minimum recall score of 95%, and a minimum F1 of 93% to be published within our products. The average F1 for all of our Sifters is just above 95% and rising (Sifters improve over time). The Sifters coming out of our Sifter Factory in the last year are starting at 97%+. These F1 scores materially outpace human performance. That doesn't mean that the Sifters are "smarter" then people, but it does mean that a collection of Sifters is more likely to accurately classify and analyze contract concepts then a human.

Do Sifters get smarter over time?

Yes. Sifters are consistently improved with the help of Sifter Trainer reports and focused research and development efforts from our Data Science and Content Development teams. When we started in 2013, our Sifters were performing at F1 levels around 90%. Now they exceed 97%. We will continue to improve the performance of our Sifters as our user base grows and our technology advances.

Do Sifters make mistakes?

Yes. Just like people, our Sifters make errors. Clients and team members report Sifter errors through our Sifter Trainer, a feature of both LegalSifter Review and LegalSifter Organize. Once our Content Development team confirms the accuracy of the report, we will retrain the Sifter that made the error within 5-10 business days and release the improved Sifter back into production.