How zeb Built an Automated ML-Driven Revenue Prediction System for Hedge Fund Investment Decisions

Challenge: Manual revenue analysis and scalability limitations

Our client needed to predict company revenues for the current and next three future quarters across 796 entities to inform investment decisions. The core complexity was that future business metrics did not yet exist, making direct revenue prediction impossible without first forecasting those metrics. Existing processes relied on manual analysis that could not scale to cover approximately 720 KPIs per entity, resulting in incomplete coverage, delayed insights, and inconsistent prediction quality.

At the same time, predictions had to outperform market benchmarks (consensus) to deliver genuine investment value. Data arrived daily from third-party vendors with varying update frequencies, requiring handling of stale and missing data without compromising prediction timeliness.

Solution: Implementing a dual-pipeline ML architecture for automated revenue prediction

To address these challenges, our experts designed and implemented an automated ML-driven system that enabled large-scale, consistent revenue prediction across all entities.

• Real-Time Metric Forecasting Enablement: We implemented Prophet models to forecast 180 days of future business metrics across 796 entities. This ensured that carry-forward logic triggered retraining only for updated metrics, optimizing compute while maintaining data freshness.
• Quarterly Revenue Prediction Framework: Forecasted metrics were fed into multiple regression models, including Ridge, Linear Regression, XGBoost, and LightGBM, across multiple feature engineering scenarios to generate diverse prediction outputs for each KPI.
• Ensemble-Based Prediction Optimization: Generated multiple predictions per KPI and ranked them using MAPE, hit rate against benchmarks, and efficacy scoring. Top-performing predictions were selected and combined using Z-score weighting to produce the final revenue prediction.
• Advanced Feature Engineering Approach: Developed multiple feature engineering scenarios incorporating imputation, normalization, correlation-based selection, and benchmark weighting to ensure strong predictive signals across entity-KPI combinations.
• Prediction Validation and Monitoring: Implemented validation checks covering data consistency, prediction boundaries, and pipeline accuracy during each run. Performance tracking mechanisms ensured continuous monitoring of model effectiveness.
• Scalable ML Architecture Design: Designed a dual-pipeline architecture capable of supporting additional entities, KPIs, and feature scenarios without disrupting existing workflows.

Benefits: Scalable predictions, improved accuracy, and automated operations

• 80% quarterly YoY prediction accuracy delivered, ensuring reliable forecasting outcomes.
• 90% improvement in time efficiency, driven by automated daily and quarterly pipelines.
• Extended forecasting horizon enabled predictions for the current and next three future quarters.
• Consistent benchmark outperformance achieved against market prediction.
• Enhanced data quality ensured through integrated validation checks before dashboard reporting.

Looking to automate your investment analytics?

Zeb enables organizations to build scalable ML systems that deliver reliable predictions at scale. Collaborate with our experts to design data-driven architectures that support faster and more informed investment decisions.