Dive Brief:

• A more than decade-long inventory management improvement process at Intel generated over $1.3 billion in increased gross profits from 2014 to 2017, according to a paper published last year in INFORMS Journal on Applied Analytics by a team of researchers from Intel and a professor from University of Tennessee.
• The new inventory management model has largely automated inventory planning and uses historical data from similar products to help predict demand for SKUs that have not generated enough of their own demand data to inform a forecasting model.
• "Intel’s total finished-goods inventory subsequently decreased by $321 million by the end of 2013 and by an additional $280 million by the end of 2014, contributing to a 30% ($600 million) reduction in total finished-goods inventory for Intel within two years of implementation," the paper reads.

Dive Insight:

The "multiechelon inventory optimization" model turned an inventory management process that previously relied on "rule-of-thumb heuristics and spreadsheets" into one where planners accept an algorithm's automatically generated inventory targets 99.5% of the time, the researchers said.

As the researchers point out, the importance of getting inventory levels right was something Intel, a company that has been manufacturing for decades, understood quite well.

"Since Intel’s first device rolled off the manufacturing line, holding too much inventory has wasted capacity and holding too little has risked revenue," the researchers wrote. The effort got some push back in the early days with comments suggesting it wouldn't work or the current method was functioning fine. But when a former factory manager was appointed as the VP of supply chain, "he felt there must be a better and more scientific way to set inventory targets than just planner rule-of-thumb estimates applied across all products," the paper reads.

The automation of the process has allowed planners to focus on more complicated inventory issues within the company, the paper said.

"All facets of calculating the inventory targets (e.g., processing data inputs, calculating targets, publishing summary statistics) were automated, freeing planners to handle only the exceptions — the truly problematic SKUs on allocation or newly launched in uncertain markets," the researchers wrote.

Intel began piloting the new inventory model in 2005 with its boxed CPU channel, which sells microprocessors in retail-branded boxes through distributors and local computer manufacturers. It resulted in an 11% reduction in inventory investment while total demand was satisfied on time more than 90% of the time, the paper said.

But in the pilot, workers would still manually override the algorithm about 50% of the time. The researchers tweaked the algorithm to improve its fit before rolling it out more widely across the company.

The pilot only ran on some of the SKUs within the boxed CPU business, which provided the researchers with a control. The new method satisfied demand on time at a higher level than the old method while keeping lower inventory levels. Notably, the SKUs using the new model also "avoided the wild fluctuations in inventory that the control group experienced during the turbulent markets of the Great Recession," the paper said.

In 2011, Intel began rolling out the model to its vendor-managed inventory, which was a year-long effort.

As for forecasting the demand for new products, Intel began mapping products to tie current items to its most-similar predecessor. The company relied on a number of variables to make the comparison including price, division code, market segment and microprocessor brand. The company only uses a product's data to forecast demand if it has 22 weeks of demand data. If it doesn't, then it forecasts the demand using historical data from other products.

Planners were able to override the algorithm's suggestions throughout the pilot, but the frequency of this trended down throughout the project as the model improved.