Improve
Generate and select countermeasures
Improve turns verified causes into solutions. Brainstorming and benchmarking generate options; a weighted solution selection matrix scored against impact, cost, risk, and time prevents choosing the first idea. Pilot on a limited scale before full deployment, and run an FMEA on the proposed design to catch failure modes before they reach the customer.
Lean countermeasures
Lean tools include 5S, setup reduction, kanban, cellular layout, standard work, and poka-yoke (error proofing) applied at the source so a defect cannot pass.
Design of experiments
DOE studies several factors at once and estimates interactions, which one-factor-at-a-time testing cannot do. Know the vocabulary: factors, levels, response, treatment combinations, main effect, interaction, replication, randomization, and blocking. A full two-level factorial has 2 raised to the k runs, so three factors need eight runs; adding three to five center points tests for curvature. Fractional factorials (2^(k-p)) screen many factors cheaply but alias effects. A resolution IV design keeps main effects clear of two-factor interactions, though those interactions alias each other.
Traps
Failing to randomize run order, confounding a factor with a nuisance variable such as shift or operator, claiming an interaction without replication, and jumping to an optimization design before screening.
Worked example
Three factors at two levels, temperature, pressure, and catalyst, need a full 2^3 design of 8 runs. If budget allows only 4 runs, use a 2^(3-1) half fraction, accepting that each main effect is aliased with a two-factor interaction. Randomize the run order and replicate before trusting small effects.