Case study · Manufacturing
Forecast demand instead of reacting to it.
We benchmarked forecasting approaches and deployed DeepAR+ for an engine and power-solutions manufacturer, cutting inventory holding costs by 17%.
Engagement detail
The problem
A leading manufacturer of engines and power solutions needed better demand forecasting to get ahead of inventory. Demand swings, long material lead times and competitive pressure meant both stockouts and excess inventory were eating into margins and customer satisfaction. The existing process couldn't reliably capture seasonality or cross-product-line demand relationships, so planning and procurement were reacting to demand rather than anticipating it.
What we built
We benchmarked multiple forecasting approaches — ARIMA, LSTM neural networks and other machine-learning methods — on MAPE%, RMSE and MAD. DeepAR+ was selected for its ability to learn seasonal patterns and related time series across product lines at once. Source ERP data lands in a cloud data lake, is transformed through a serverless ETL layer and loaded into a warehouse for modelling; the model runs on a managed ML platform, and monthly forecasts flow into a BI semantic layer for planning and procurement.
Impact
- 17% reduction in inventory holding costs.
- Monthly forecasts feeding directly into production planning and procurement.
- Planned enhancement: real-time marketing inputs and customer-forecasted sales to sharpen accuracy further.
On sourcing. Every fact and figure here comes from Sail Analytics' project record for this engagement. Client names are withheld where they were not released for publication.
The measured result
What the client measured.
−17%
Inventory holding cost
Reduction from forecasting.
Monthly
Forecast cadence
Into planning and procurement.
Have a problem like this?
Tell us the systems you're working with and what you can't currently see.