Smart Storage Forecasting Using AI in Tank-Managed Systems

Quick Summary

Challenge

A global water treatment company struggled to optimize chemical reorders due to delayed lead times, batch-only constraints, and unreliable usage data across tanks.

Solution

Tatras Data developed a predictive forecasting system to model tank depletion rates and trigger timely reorders.

Result

42% increase in reorder prediction accuracy.

Tech Stack

AI: ML: Rolling window analytics Statistical modeling Anomaly detection | Data & Retrieval: Tank sensor logsDB Delivery schedules Historical usage patterns | Dev: Python Pandas Custom forecasting logic | Security: On-premise data processing with access-controlled outputs

The Challenge

The company’s chemical reorder cycle was under strain. Tanks across sites had strict minimum and maximum fill levels. Deliveries took 28 days. And, orders could only be placed in fixed batch sizes. But reorder decisions were based on guesswork, not on real consumption data. Some tanks would run dangerously low. While others sat overstocked. Reorders were often late, sometimes too early. The impact rippled across logistics, procurement, and plant operations. What the client needed was a way to predict when each tank would hit its reorder threshold, days or weeks in advance.

A Day in the Life: Before Our Solution

Each morning, operations managers scanned dozens of tank dashboards. Some tanks had dipped below safety levels. Others hadn’t been touched in days. The refill cycle was anyone’s guess. Manual checks, spreadsheet-based trackers, and inconsistent sensor data made the job harder. Even when a tank looked low, managers hesitated. Was it truly time to reorder? Would the batch meet the supplier minimum? Would it arrive in time? The result: late orders, emergency deliveries, and bloated stock at other sites. All while trying to meet compliance standards and avoid downtime.

Pain Points:

Stockouts triggered costly emergency deliveries
Overstocking led to wasted inventory and storage cost
28-day delivery window created planning pressure
Sensor and delivery data lacked synchronization
No system-wide visibility or intelligent reorder logic

Solution

1. Core Innovation

Tatras delivered a multi-phase system that aligned tank-level consumption data with delivery schedules and operational constraints:

Rolling Window Forecasting: Predicted future tank usage based on recent consumption patterns.
Sensor Fusion: Merged real-time fill-level data with historical trends and delivery history
Custom Reorder Logic: Accounted for batch constraints, lead times, and site-specific behavior.
Anomaly Flagging: Surfaced tanks with unreliable patterns for manual review.
Integrated Output: Routed reorder triggers into existing procurement and inventory systems.

2. Key Features

Daily predictive reorder scores per tank
Constraint-aware scheduling (batch size, lead time, max/min fill levels)
Sensor + delivery data fusion for improved accuracy
"Not OK" tank flagging based on anomaly thresholds
Configurable for tank-specific usage behaviors

3. Workflow Integration

The forecasting engine runs nightly across tank telemetry logs and flags tanks nearing reorder risk thresholds. Predicted depletion dates are visualized weeks before the actual event. This gives procurement teams time to schedule deliveries that match business needs and supplier timelines, while also catching data pipeline errors that could compromise operations.

Outcomes

📈 42% improvement in reorder prediction accuracy

🧯 Reduced stockouts, especially in high-volume tanks

🔍 Identified and addressed gaps in sensor-data quality

🤝 Stronger coordination between operations and procurement