Predictive Maintenance
IGS Energy's legacy on-premises platform and manual solar monitoring process could not reliably handle terabyte-scale forecasting data or accurately detect rooftop-solar underperformance, creating cost, complexity, and customer-experience risk.
Reported impact
IGS cut forecasting-model training costs by 75%, reduced forecast generation from about half an hour to minutes for hundreds of thousands of customers, and improved solar anomaly detection so customer issues can be identified with less manual effort and fewer false results.
IGS Energy rebuilt its demand-forecasting and solar-performance analytics on Snowflake so data science could work directly on the large energy datasets that drive commodity purchasing and customer service. Weather signals, historical energy consumption, account attributes, and contract dates flow into Snowflake, where the company stores and queries 40 to 50 billion rows of data that can exceed a terabyte. Instead of training hundreds of thousands of customer-level models in Databricks, the team now trains a single unified forecasting model in Snowflake, cutting training overhead while preserving accuracy. For solar operations, IGS used Streamlit and Snowpark to prototype an anomaly-detection application that combines weather data with solar-array specifications to flag underperforming rooftop systems. Marketplace datasets can be joined in place, keeping forecasting, feature engineering, application development, and stakeholder review inside one governed Snowflake environment.
Weather data, historical usage, account details, and contract dates land in Snowflake, where IGS stores terabyte-scale energy data and trains a unified demand-forecasting model instead of per-customer models. For rooftop solar, Streamlit and Snowpark use weather inputs and array specifications to score performance anomalies, and the resulting forecasts or alerts are consumed by commodity-purchasing teams and customer-service operations.
Moving from hundreds of thousands of individual forecasting models in Databricks to one unified model in Snowflake helped IGS achieve 75% cost savings in training — without sacrificing accuracy.
Previously, the process to train all these models and generate predictions took a half hour.
Thanks to Streamlit’s ease of use, Shah’s team created an initial anomaly detection application for internal stakeholders in just two weeks for testing.
The data produced in this process could add up to 40 to 50 billion rows and exceed a terabyte.
The unified model on Snowflake is super quick; we’re talking minutes to generate forecasts for hundreds of thousands of customers.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Snowflake Sample Modified · Aug 7, 2026 Uniper Volatile energy markets demanded faster access to trading, risk, and operational data than Uniper's aging reporting platform and fragmented systems could provide. 18.5 gigawatts of power-generating capacity Snowflake · Snowflake CoCo · Snowflake CoWork Snowflake Sample Archived · Jul 16, 2025 Expand Energy replace siloed warehouse, Hadoop, and SAP systems that could not combine financial, IoT, and OT data fast enough for self-service analytics, drilling decisions, forecasting, and M&A integration. 1,000+ Minimum estimated hours of labor saved annually Cortex Analyst · Snowflake · Snowpark Snowflake Sample Modified · Sep 4, 2026 Eutelsat Group GEO and LEO satellite operations generated petabyte-scale telemetry that needed unified storage, cross-domain access, and AI-driven outage prediction without adding major platform-management overhead. $2M savings in just six months Apache Iceberg · Cortex AI · Cortex Analyst 