Why We Need to Stop Ignoring Simulation for Inventory Uncertainty
This post originally appeared in Supply Chain Brain.
Optimization is great for many things but terrible at capturing uncertainty and implementing recommendations on its own. Implementing optimization recommendations still requires a big dose of, “Well, I hope this works!”
For years, simulation was dismissed because of performance and memory requirements. Nowadays, those barriers are gone, meaning simulations can be completed at the order and shipment level, allowing businesses to calculate true service rates for proposed supply chain states.
Simulation vs. Optimization
Back in the 1990s, companies frequently said they wanted to calculate service rates — the percentage of customers that got what they requested by the time they requested it — when modeling network design problems. At the time, powerful tools used to optimize networks couldn’t predict service rate changes.
Optimization is generally used to define the best network structure and is typically most concerned with cost. Meanwhile, simulation is often used to replicate a system’s performance and to examine the impacts of changing business rules or other elements (customer ordering patterns, processing times, etc.).
Simulation is the only approach that can accurately predict service rates. LLamasoft was originally founded as a supply chain simulation company to predict the effect of network changes on service levels. It incorporated mathematical optimization and hoped companies would use both tools to evaluate supply chain changes and quantify the impact on service.
When a traditional network optimization study recommends a facility be shut down, too often, other facilities are left to pick up volume, capacity limits are pushed, warehouse spaces need to be rented and last-mile transport costs need to be elevated. But if that same network study had used simulation, the decision to close the facility likely wouldn’t have been made.