Western Digital makes thousands of products across a global Silicon Operations network, and each product must be qualified at a site before it can be made there. Qualification takes months and hundreds of engineering hours.
The obvious choices are bad: qualify each product at only one site, leaving no fallback, or qualify it everywhere, spending a fortune on redundant qualification.
The move was a mixed-integer stochastic program with multiple stages of recourse. It decides which sites qualify each product, weighing both expected demand and demand uncertainty, solved with a greedy heuristic plus a linear programming solver.
Deployed in the first quarter of 2015 and run as a quarterly process, it streamlined the decision and produced substantial savings from fewer shortages and less wasted qualification effort.
Pre-qualify sites to shift demand when it spikes
Qualification is slow and costly, so you cannot wait to see demand before deciding.
Demand is uncertain, so the choice must weigh expected demand and the risk of a shortfall.
Multiple stages of recourse let the plan revise site choices as uncertainty resolves.
Having alternate qualified sites turns a potential shortage into a routing change.
When you cannot react quickly, buy the option to react. Pre-qualifying a second source is slow and expensive, but it converts a would-be shortage into a routine re-routing instead of a lost sale.
Western Digital deployed the model from the first quarter of 2015 and institutionalized a quarterly site-qualification review, streamlining the decision process and crediting the tool with substantial savings in lost sales and qualification effort.
FOLLOW THE EVIDENCE
The sources
- Optimizing Site Qualification Across the Supply Network at Western Digital EconPapers / INFORMS Interfaces · econpapers.repec.org
- Optimizing Site Qualification Across the Supply Network at Western Digital INFORMS Interfaces · pubsonline.informs.org