The five phases
- Define. State the problem, the customer requirement, the goal and the scope, usually in a project charter with a process map (SIPOC).
- Measure. Establish how the process performs now: the measure, its data collection plan, the measurement system's reliability and the baseline capability.
- Analyse. Find the root causes of the gap using the data: Pareto charts, fishbone diagrams, hypothesis tests, regression, process analysis.
- Improve. Generate, test and implement solutions, often with pilots or designed experiments, and confirm the measure moved.
- Control. Make the gain stick: control charts, updated standard work, training, a response plan if the measure drifts, and handover to the process owner.
DMAIC, PDCA and A3
All three are the scientific method applied to a problem. PDCA is the shortest and suits daily problems solved by the team in days. A3 adds a one-page story and coaching and suits problems that take weeks. DMAIC is the most rigorous and statistical and suits problems of variation that take months and a trained belt. The mistake is to use DMAIC for everything, which starves daily problems of attention, or PDCA for everything, which leaves chronic variation untouched.
When DMAIC is the right choice
- The cause is unknown and not obvious from observation.
- The problem is about variation or capability rather than a single failure.
- Data exists or can be collected.
- The gain is worth weeks of project time.
DMAIC and daily management
DMAIC projects come from the board. Measures that stay red week after week despite daily countermeasures are the candidates; the project's progress is reviewed at the Tier 2 or Tier 3 meeting; and the control phase ends with the measure back on the board with a new standard behind it.