EN
Back to the archive

The encyclopedia · Strategy & Leadership · Operational decision · 2015

USG cut modeled distribution costs 4.8% by optimizing for uncertain demand

USG optimized its Durock network with chance-constrained stochastic programming, cutting modeled delivered cost about 4.8% versus a single-month plan.

USG Corporation

the move

USG makes building products and ships its Durock line to customers throughout North America from plants with capacity limits. The old approach set the network plan from demand and cost data for a single month, so any swing in demand made the plan wrong before it finished rolling out.

The team first showed that demand uncertainty, not production-cost uncertainty, was the main driver of cost variation. They then reformulated the problem as a stochastic program with chance constraints, optimizing for the 50th percentile of demand and applying penalty costs for unfulfilled constraints.

The optimized network cut theoretical delivered cost by about 4.8% versus the base case. It was implemented as sourcing rules in USG's order fulfillment system and Oracle's advanced supply-chain planning module; practical delivery concerns reduced the realized benefit, but savings remained substantial.

why it works

  • Modeling demand as a distribution instead of a point estimate keeps the plan robust to the swings that actually happen.
  • Chance constraints formalize how much service risk is acceptable instead of hiding it in a safety margin.
  • Implementing via sourcing rules in production systems meant the plan ran every day, not once.
the payoffOptimize for a demand distribution, not one monthneat

what transfers

A plan built for last month's numbers is a guess; a plan built for the demand distribution survives swings, and a modest theoretical gain is worth chasing when freight and production run at scale.

what came after

USG deployed the new sourcing rules in its order fulfillment system and Oracle's supply-chain planning module. Realized savings were below the theoretical 4.8% because of practical delivery constraints, but management still considered them substantial.

references

spotted an error? The archive wants to know.

same kind of clever