Leadfeeder uses machine learning to improve its users’ sales lead quality. New model reaches 98% accuracy in the first weeks.
Leadfeeder uses Google Analytics data to show companies visiting user websites. While the recognition of visitors is primarily based on network location data, visits made from home broadband subscriptions or mobile phones is often not correctly identified but shown as the Internet Service Provider.
Example of unwanted noise in lead list; it’s not a company
This is not useful for sales lead generation purposes and an effort is made to filter those out from user results. So far this has been done with pattern recognition models and manual verification.
However, as Leadfeeder is tracking currently over 30,000 Google Analytics accounts our data is currently at a scale it can support machine learning models. To improve our data quality we implemented machine learning models to filter ISP noise from leads.
Separating noise from a real lead is difficult and even humans make mistakes. Machine learning model we are using is already exceeding the quality of our previous process. “We implemented a machine learning model a week ago and the accuracy is already 97% and we believe it to go even higher in the upcoming weeks when the algorithm learns more” says Matthew Higgins, software engineer who was in charge of the implementation.
Data from Google Analytics and other sources Leadfeeder uses is dirty and requires a lot of categorization and clean-up in order to be useful. Leadfeeder does it both manually and automatically. “While we are constantly improving our automatic tools, machine learning is clearly the next step with our amount of data”, says Herkko Kiljunen, Leadfeeder CTO.
We’ve estimated that handling an unwanted web lead takes on average 4 seconds. If our machine learning model allows us to filter even one lead per account, the new model is saving over a week’s worth of manwork every day.
Start free 30-day trial
No credit card required.
Free leads from Google Analytics. No extra script installation needed.See for yourself
Free trial. No credit card required.