ML based Attach Rate Forecasting

The Customer

This large High-Tech customer based in Silicon Valley provides unified data storge services to the world’s biggest clouds. The customer plans at two levels: Configure To Order (CTO) platforms and Make To Stock (MTS) sub-assemblies.

The Problem

The customer had a high degree of forecast error at MTS sub-assemblies level because of highly configurable platforms. Although the Planners would translate Platform level Plan into Sub-Assembly level Plans using Planner maintained Attach Rates, these rates were not in-sync with Sales Order Bookings and hence led to high error in forecasted vs. needed sub-assemblies.

The Solution

GitaCloud Data Scientists analyzed Sales Order Bookings (History and Open Orders) to understand the linkages between specific configurations of a Platform and the resulting attach rates for sub-assemblies. Attach Rate variability and trends were also assessed to understand forecastability using attach rates. Machine Learning based forecasts for Attach Rates were then used to drive conversion of Platform Forecast to Sub-Assembly Forecasts.

The Value

GitaCloud team was able to demonstrate 20% error reduction using Sales Order History based ML generated Attach Rates compared to Planner Forecast for Sub-Assemblies using static Attach Rates.