Sales Opportunity Predictive Forecasting
Sep
2024
Ongoing
Problem / Purpose
The Sales Team relied on qualitative labeling and Salesforce Einstein scores to predict opportunity closures, both of which were not sufficiently accurate.
Solution
Used DBT and Dagster to build a Snowflake data pipeline sourced from Salesforce. Created two models in DataRobot using a Random Forest Classifier—one for the first half of the sales process and one for the second half. Ensured proper data partitioning to avoid leakage. Scheduled the model training and scoring to run weekly, storing results in Snowflake (with snapshot history) and loading predictions into Salesforce.
Key Achievements / Impact
Reduced reliance on inaccurate manual or Einstein-based forecasts. Achieved 90–95% accuracy in predicting whether an opportunity would close in the current or following month, significantly improving forecasting reliability for both Sales and executive leadership.
Key Technologies / Tools Used
SQL, Snowflake, DBT, Dagster, DataRobot, Salesforce, ML Models (Classifier), End-to-End Data Pipeline Design, Model Scoring Automation
Role
Data Scientist
ProService Hawaii