The five phases
DMAIC structures an improvement project so that conclusions come from data. Each phase ends with a deliverable and a review before the next one starts.
- Define: what problem, for whom, with what goal and scope
- Measure: how the process performs today, and how reliable the measurement is
- Analyse: what causes the problem, proven with data
- Improve: solutions that address the causes, tested before full rollout
- Control: how the gain is kept, and who responds when it slips
Define
Write a problem statement that is specific and measurable: what is wrong, where, since when and how large. Name the customer requirement it affects as a CTQ characteristic, set a goal, and agree the scope and the people involved.
Typical tools are a project charter, a SIPOC diagram (suppliers, inputs, process, outputs, customers) and a short summary of what the customer requires.
Measure
Decide what to measure and how, then check that the measurement can be trusted. A measurement system analysis such as a gauge R&R study shows how much of the observed variation comes from the measuring itself.
Then collect data and establish the baseline: defect rate, DPMO, capability (Cp, Cpk, Pp, Ppk), yield or OEE, depending on the problem. Plot the data over time before drawing conclusions.
Analyse
Look for the causes, then confirm them with data instead of opinion. Pareto charts show which categories dominate, fishbone diagrams and the five whys help to generate hypotheses, and stratification by machine, shift, material, tool or operator shows where the problem concentrates.
Correlation and regression can identify associations. Check for competing explanations such as a material change coinciding with a shift change. Where practical, vary one suspected factor in a controlled trial and check whether the result follows it.
Improve
Generate solutions aimed at the verified causes and choose between them on effect, cost and risk. Test on a small scale first, with a pilot or a designed experiment, and compare against the baseline.
Prepare the implementation: instructions, training, and what happens to the old way of working.
Control
Make the gain stick. Update the control plan and standard work, put the critical characteristic on a control chart, set limits and a clear response plan, and hand the process to its owner together with the evidence of the improvement.
A short review after a few months confirms that the gain held. If it did not, the project returns to analysis instead of starting again from zero.
A worked example
Suppose 18 of 500 turned parts fail the diameter check. The team aims to bring scrap on this characteristic below 2% within a quarter. A gauge study checks the measurement and a time plot shows that the diameter climbs as each tool ages.
On comparable runs with the same material, the diameter returns towards target after the insert is replaced. The team pilots an earlier inspection. Six of the next 500 parts fail, or 1.2% versus the 3.6% baseline. The shift lead owns the reaction plan and tracks further runs to see whether the change holds across tools and lots.