DMAIC: the five phases, with deliverables and gate questions

DMAIC stands for Define, Measure, Analyze, Improve and Control: the fixed rhythm Six Sigma uses to improve an existing process. The key is the order — you only step in once the cause has been shown with data, and you're only done once the improvement holds up without a project team. A well-scoped DMAIC project typically takes 6 to 16 weeks.

Want to know where DMAIC comes from and when to use it? Read Lean Six Sigma. The methods behind it are covered in Six Sigma and Lean.

Define — which problem are we solving?

Goal: one problem, in numbers, with an owner. Without a sharp Define, every following phase turns into a debate.

Deliverables:

• a problem statement with a number (current state, target, timeframe);
• a customer requirement (CTQ) that makes "good" and "bad" objective;
• scope: the start and end point of the process, and what's explicitly excluded;
• a sponsor and team members with time they can actually spend on it.

Gate question: do we know which problem we're solving and what it's worth?

Pitfall: a problem statement that's actually a solution ("we need a new system"). Turn it back into the outcome that isn't right.

Measure — can we trust the numbers?

Goal: map the process as it really runs, and set a baseline to fall back on later.

Deliverables:

• a process map of the current way of working, built with the people doing the work;
• a measurement definition: what we measure, who measures, how often, in what unit;
• a measurement system analysis: do two people measuring the same case get the same result?
• a baseline over a representative period (usually 2 to 4 weeks);
• an initial Lean round: obvious waste you can already remove.

Gate question: do we trust the numbers we're about to base decisions on?

Pitfall: endless measuring. Put a deadline on this phase; four good weeks beat six months of noise.

Analyze — what's the real cause?

Goal: move from guesses to proven causes. This is the phase where most assumptions fall away.

Deliverables:

• a cause overview (fishbone) and a pareto of the biggest contributors;
• data analysis per suspected cause, with a test for whether the difference is real;
• a substantiated root cause, backed by the data.

Gate question: is the cause proven rather than assumed?

Pitfall: reading correlation as causation. Two things moving together don't necessarily cause each other; decide in advance what outcome would disprove your assumption.

Improve — does the solution actually work?

Goal: a solution that measurably shifts the baseline, tested small first.

Deliverables:

• solutions developed with the team, chosen for impact and feasibility;
• a pilot or test on part of the volume, using the same measurement as the baseline;
• a risk check: what could get worse elsewhere as a result?
• a rollout with a clear date and owner.

Gate question: is the baseline measurably moving in the right direction?

Pitfall: introducing everything at once. Then, if things improve, you won't know which part worked — and if things get worse, you won't know what to undo.

Control — does it stick?

Goal: hold on to the gain without a project team. This is the most commonly skipped, and most costly to skip, phase.

Deliverables:

• standard work: the new way documented, short and easy to find;
• monitoring: one number, a fixed frequency, visible to the team;
• an agreement on what happens if the number goes out of range;
• handover to an owner in the line and a review moment after three months.

Gate question: does the gain hold once the project team is gone?

DMAIC or DMADV?

Use DMAIC when the process exists and the outcome isn't good enough. If the process doesn't exist yet, or needs a full redesign, use DMADV instead: Define, Measure, Analyze, Design, Verify. Same rhythm, but with designing instead of improving.

DMAIC in a few practical rules

• One process, one outcome measure per project. Two goals means no clear conclusion.
• Pick the measure in Define and never change it along the way; how to choose one is covered in Choosing KPIs.
• Go back if a gate isn't passed. DMAIC is a cycle, not a one-way street.
• Plan Control before Improve. Decide who will maintain the standard before the end, not at it.
• Keep it visible: a Gantt chart per phase works better than a project plan nobody opens.

Sources

• ASQ — DMAIC
• ASQ — Six Sigma
• ISO 13053-1: Quantitative methods in process improvement — Six Sigma