Focus on the data flow. Start by mapping every input–output pair in your build process. Then treat each component as a node in a directed graph, edges weighted by material throughput. Run a cost‑minimisation algorithm on that graph—linear programming works well if you can quantify waste as a penalty term. Next, identify critical paths where waste accumulates; those are your hotspots. Replace the heavy nodes with alternative modules that have a lower coefficient in the objective function, or redesign the geometry to reduce material. Finally, loop in real‑time sensor data: let the system adjust weights on the fly so the optimisation stays valid as production variances creep in. Keep the model lean—too many variables make the solver choke and blur the patterns you’re hunting for.

Good plan. Watch the variance; if the test batch shows an unexpected spike, trace the graph back to a single edge and reassess that coefficient. Keep it lean, keep it tight.