Identifying high-quality sales leads at G5

G5 knew machine learning could identify the stronger leads more likely to convert. But its product team of statisticians and behavioral scientists had no dedicated data science resources, and a from-scratch build risked being slow, expensive, and a barrier to innovation.

G5 built training sets of 100,000 lead call transcripts and their scores, stored on Amazon S3 with compute on Amazon EC2. It used H2O Word2Vec to turn the transcripts into a feature table, then H2O Driverless AI to automatically engineer features, tune, and train the model, raising lead-identification accuracy with each pass.

To put results in front of leasing agents, G5 ran the models on AWS Lambda through Driverless AI's MOJO scoring pipelines, a low-latency way to deploy and score at production scale. Model complexity stayed hidden from agents, who simply received a ranked list of high-value leads to call.

Lead scoring accuracy climbed to over 95%, and leasing agents now connect with qualified leads 85% of the time, up from the previous 14% benchmark. Feature engineering and model tuning time dropped 80%, and accuracy improved six times over traditional lead follow-up.

With models built and deployed this fast, G5's product team estimates it delivers the work of two additional senior technical staff with no dedicated data science or deployment resources. One G5 customer saved over $500,000 a year.