Smart Storage Forecasting: Optimizing Reorders Based on Tank Usage Patterns

Smart Storage Forecasting
Optimizing Reorders Based on
Tank Usage Patterns

The Challenge

A leading specialist in water treatment needed a predictive system that determines the optimal reorder time for tank-stored chemicals, balancing complex constraints. These include batch-only ordering, strict tank level limits, and a 28-day delivery lead time. The system must accurately forecast consumption patterns to trigger timely reorders, preventing stockouts while avoiding excess inventory, thus optimizing operational efficiency and cost-effectiveness.

Hypothesis

  • Implementing a combination of statistical techniques and rolling window analysis on tank sensor data will accurately predict optimal reorder times for a majority of tanks.
  • This approach effectively balance inventory levels, accommodate batch ordering constraints, and account for the 28-day lead time, resulting in improved operational efficiency and cost savings across a large proportion of tanks with sufficient historical data.

Execution

  • Phase 1: Analyze tank inventory data using statistical techniques and rolling window analysis to develop initial reorder prediction model.
  • Phase 2: Integrate tank delivery data to refine the model and improve prediction accuracy.
  • Phase 3: Combine sensor and delivery data to further reduce errors and optimize the reorder prediction system.

Outcomes

  • Identified critical data pipeline issues affecting model accuracy.
  • Uncovered lack of correlation between sensor and delivery data, challenging initial assumptions.
  • Concluded need for data pipeline refinement and exploration of alternative prediction approaches.
  • Custom code to calculate reorder points based on daily usage.
  • Feasibility analysis to identify tanks with “Not Ok” behavior.

Project Highlights

42%

increase in reorder prediction accuracy.