Improve Phase
Purpose
Improve turns verified causes into changes that move the metric, then proves the change worked.
Approach
Generate candidate solutions (brainstorming, benchmarking, supplier input), evaluate them against criteria (impact, cost, risk, time, ease), and select with a weighted solution-selection matrix rather than by loudest voice. Then pilot on a limited scale, confirm the effect with data, and only then roll out.
Lean tools live here: 5S (sort, set in order, shine, standardize, sustain), kaizen events, value stream mapping to expose non-value-added steps, kanban pull, SMED for changeover time, and poka-yoke (mistake-proofing). In the poka-yoke hierarchy, prevention (making the error impossible) beats detection (catching it afterward); a fixture that accepts only the correct part is prevention, an alarm is detection.
DOE (design of experiments) changes several factors together so main effects and interactions can be estimated efficiently. One-factor-at-a-time testing cannot reveal interactions and usually costs more runs.
PDCA (Plan-Do-Check-Act) is the improvement cycle; a pilot is a Plan-Do-Check before a full Act.
Worked example
A 2² experiment varies temperature (low/high) and pressure (low/high). The main effect of temperature = average response at high temperature minus average response at low temperature, averaged across both pressure levels. If the four responses are 10, 12, 14, 16, with high temperature giving 14 and 16, the temperature effect = 15 − 11 = 4.
Traps
Skipping the pilot, treating detection as mistake-proofing, and forgetting to update the FMEA and control plan when the process changes.