- The inventory Planning team wanted to improve Supply Chain efficiency by replacing their existing siloed ML/AI process with an advanced touchless ML Platform in Azure.
- The requirement of innovative solutions to improve the existing forecast accuracy for key high-value parts & change the associated business processes.
The Before state
Inefficient forecast system for Inventory Planning of Personal Systems services resulting in
- Delayed forecast results – 5 days effort from 3 data scientists to reach the inventory team.
- Delays / Failure in decision making leading to excess products worth $15M per year.
- Failure to meet the SLAs.
The LatentView Solution
Operationalization of ML/DL models by building a reliable, automated forecasting system to generate forecasts faster.
- Migration of 12 forecast algorithms (100k ML/AI models for 24k parts) from Azure DSVM to Azure Databricks backed by Apache Spark.
- PySpark and SparkR APIs for distributed computing & hence faster computation
- Scalable framework using On-Demand clusters.
- Seamless orchestration of Data & ML Pipelines using Azure Data Factory.
- Constant Monitor, Data Audit & Version Control using MLFlow integration.
The After state
Helped the business to improve Supply Chain efficiency with timely decisions using faster forecast results.
- 8x faster computation
- 5x Reduction in Infrastructure Cost
- A scalable framework to continually serve the future scope
- Automated workflow integrated with the business operations
- Up to 1.5% improvement in Scientific models for M1 month
- Monthly data refresh automatically triggered
- Spark Upgrade
- Supported migration of additional 30+ traditional algorithms from DSVM to Databricks
- MLflow API to ensure Data & Model Audit & Governance
- Key vault to access sensitive information and ensure data security