Improve: Generating, Selecting and Testing Solutions
Improve converts verified causes into changes that move the metric. First generate candidate solutions with brainstorming, benchmarking, mistake-proofing (poka-yoke) thinking and lean flow tools - 5S, standard work, cellular layout, setup reduction, kanban and pull. Then screen and select using structured criteria: an impact/effort matrix, Pugh matrix (concept selection against a datum) or weighted scoring. Selection must trace back to the verified root cause; a popular solution that does not address the cause is a trap.
The heart of Improve is the Plan-Do-Study-Act (PDSA) cycle and designed experiments (DOE). Know the vocabulary: factors, levels, response, main effect, interaction, replication, randomisation and blocking. Screening designs (fractional factorials) test many factors cheaply; full factorials and response-surface methods refine settings. A trap is treating an interaction as noise or assuming one factor at a time (OFAT) is equivalent to DOE - it is not, because OFAT misses interactions.
Also understand FMEA as a risk tool: severity × occurrence × detection gives the risk priority number (RPN), and actions target high-severity or high-RPN items. Pilots limit risk before full rollout; define success criteria and a decision rule before running the pilot.
Worked example / rule of thumb: a DOE shows temperature main effect positive and a temperature × pressure interaction significant. You cannot optimise temperature independently - choose the pair of settings that maximises the response. Rule: if an interaction is significant, interpret main effects only in light of it.