Forum Energy Technologies · Manufacturing · Midland, TX · 20 weeks
Demand forecasting AI deployed across 3 product lines
An oil and gas equipment manufacturer relied on 12-month rolling averages for demand forecasting, causing frequent overstock on slow-moving SKUs and stockouts on fast movers. We developed an AI-driven forecasting model that improved accuracy, lowering inventory carrying costs by $1.8M per year and reducing stockouts from 34 to 4 each quarter.
Challenge
The manufacturer held $28M in inventory over three product lines, relying on a 12-month rolling average to forecast demand. This approach ignored oil price fluctuations, drilling permits, and seasonal shifts. As a result, slow movers piled up excess stock while fast movers ran out 34 times per quarter. Each stockout cost roughly $42K in lost sales and rush shipping.
Approach
We developed a demand forecasting model using oil price futures, drilling permit data from the Texas Railroad Commission, historical demand, and seasonal trends. The model produced weekly SKU-level forecasts for all three product lines, including confidence intervals to guide safety stock levels. We then integrated these forecasts directly into the ERP’s replenishment planning module.
Outcome
We cut inventory carrying costs by $1.8 million a year by adjusting safety stock levels based on demand patterns. Stockouts dropped sharply from 34 to 4 per quarter. We lowered the inventory balance from $28 million to $21.4 million without increasing stockout risk. The new forecasting model improved 12-week accuracy to 84 of every 100, up from 61 using the old rolling average approach.
How we tracked oil price swings to guess what the market would do next
Alright, here’s the deal. Before, we were smoothing out those wild swings in oil prices with a 12-month rolling average. It worked, but it felt like we were always a few steps behind. So, we tried something different, using WTI crude futures as an early warning system. That little change? It gave us a jump on demand shifts about 8 to 12 weeks before the orders actually popped up. Pretty cool to catch those trends way ahead of time.
We Dug Into Texas Railroad Commission Drilling Permits to Spot What’s Really Going On
We went straight to the source and pulled drilling permit data from the Texas Railroad Commission, because that’s a killer indicator for equipment demand. I built a pipeline that automatically grabs and updates this data every week, then feeds it into our demand model. The payoff? Our 12-week forecast jumped by 11 points in accuracy. Not bad at all.
We plugged the ERP system in to make restocking a whole lot easier.
We hooked the demand forecasts right into the ERP’s replenishment system. Basically, whenever stock fell below the safety level the model suggested, the system would automatically kick off purchase orders. No more us having to do it by hand. It kept the buying perfectly in line with what the forecast showed, every single time.
Results
- $1.8M Annual inventory carrying cost saved
- 34 to 4 Quarterly stockout incidents
- $6.6M Inventory balance reduction
- 61 correct of 100 to 84 correct of 100 forecasts 12-week forecast accuracy
We had $28 million stuck in inventory but still kept running out of the parts customers needed. Thinklytics built a model using oil prices and drilling permits to predict demand. Since then, we’ve cut stockouts from 34 down to 4 per quarter and freed up $6.6 million in working capital.