Sensitivity Analysis in Network Optimization
Reasons for sensitivity analysis
Optimization, by definition, provides the single best answer given a prescribed set of inputs and a specific objective. The power of algorithms makes it easy to return an optimal answer to a relatively complex problem. For instance, network optimization is a fantastic tool for providing the single best answer for a problem such as minimizing cost under known conditions, including demand, supply, and transportation and labor rates.
However, network optimization is deterministic, and reality can be stochastic or uncertain. Markets shift, and costs fluctuate. Despite the intentions and rigor of forecasting and budgeting, plan is seldom an exact match with actual. Moreover, a solution which is optimal under certain circumstances could be suboptimal under even slightly different conditions. As an example, automation of specific production process could significantly lower the cost per unit; however, the capital investment in automation is only justified under higher demand. Is the demand plan accurate? (by the way, LLamasoft has tools for this, as well!)
Algorithms and software are quite good at performing explicitly what the user requests, but is the modeler asking all the right questions? When the stakes of a decision-making are high, it is important to understand the sensitivity of a solution to any input variables which are largely unknown or subject to change.
Applying sensitivity analysis in network optimization
In network optimization, sensitivity analysis should be applied to variables which have some uncertainty due to volatility (e.g. demand) and lack organizational control (e.g. fuel price), or tie to key assumptions in the model (e.g. production yield).
In LLamasoft’s Supply Chain Guru, scenario management readily facilitates the intentional varying of prescribed inputs to evaluate and compare resultant outputs. Multiple scenarios can be used to enable understanding of sensitivity of outputs to certain inputs.
For an even more thorough validation of network design and policies, you can apply simulation to the network model within Supply Chain Guru. Simulation enables the running of multiple replications of the network design while sampling predefined probability distributions for any components of the network where variability is important to consider.
Identifying robust solutions and decision-making under uncertainty
When the sensitivity of a solution to certain variables is understood, its robustness can be determined.
Robust solutions have the greatest probability of performance (e.g. cost minimization) across a spectrum of various inputs. This can also be thought of as the slope of a line of single best answers, given different inputs. In fact, the output of Supply Chain Guru scenarios can be graphed precisely in this manner. The shape of the solution curve tells a story which can provide rich insight into robustness, trade-offs, and ‘best value’ during the decision-making process.
While providing multiple answers seemingly puts the burden back on the decision-makers, it, in fact, equips them with a deeper and multi-dimensional understanding of the potential implications of that decision. This is the harmony between powerful software and adept business analysis!
Like this article? There is more to explore! If you are a current customer, please visit our support site for more practitioner insights. Future customers can request a demo here.
Read more blogs by Dan Kogan!