Six Sigma: what it is and how to apply it

Six Sigma is a data-driven improvement method that reduces variation in a process so the outcome stays within the customer's requirements. Where Lean removes standing still, Six Sigma removes unpredictability. You use it when a process repeats, the outcome fluctuates, and nobody knows exactly why.

Want the bigger picture first? Read Lean Six Sigma. If your issue is mainly about lead time and waste, see Lean.

Where the name comes from

The name is statistical: at "six sigma", there are six standard deviations between the process average and the nearest specification limit. Including the usual long-term shift of 1.5 sigma, that works out to 3.4 defects per million opportunities (DPMO). The method was developed at Motorola in the 1980s, became widely known through General Electric, and is described by ASQ and standardised in ISO 13053.

Sigma levels and DPMO

Roughly: 3 sigma is about 66,807 DPMO, 4 sigma about 6,210, 5 sigma about 233, and 6 sigma about 3.4. The step from 3 to 4 sigma is cheap and delivers the biggest relative gain; going further is often not worthwhile for commercial processes. Pick the level that matches the cost of a single error.

The core concepts

• CTQ (Critical to Quality) — the customer's requirement, translated into a measurable limit. Without a CTQ, "defect" doesn't really exist.
• Defect and defective — a defect is a single flaw; a defective is a unit with at least one defect.
• DPMO — defects per million opportunities; it makes processes of different sizes comparable.
• Variation — spread around the average. Distinguish special causes (incidents) from common causes (the process itself).
• Process capability (Cp, Cpk) — how well the spread fits within requirements, and whether the average is centred too.
• Measurement system analysis (MSA) — checking whether your measurement itself is reliable. Skip it, and you'll end up analysing noise.

The rhythm: DMAIC

Six Sigma improves existing processes through DMAIC. For designing a new process, there's a variant called DMADV (also known as DFSS), where Improve and Control are replaced by Design and Verify.

The full breakdown per phase, with deliverables and gate questions, is on DMAIC.

The statistical toolkit

You don't need to master all of this to get started. A control chart and a pareto chart will take most organisations a long way. Testing whether a difference is real is exactly the same question as with an A/B test — you can use our significance calculator for that.

Belts: who does what

• Yellow belt — knows the language, takes part in a team. A few days of training.
• Green belt — leads improvement projects within their own work. Usually 5 to 10 training days plus a real project.
• Black belt — leads complex projects, masters the statistics, coaches green belts.
• Master black belt / sponsor — trains others, chooses the project portfolio, removes blockers.

Certification isn't centrally regulated: no single body issues "the" title, and quality varies widely between providers. Ask about the project behind the certificate.

When Six Sigma doesn't fit

• Not enough volume. Without repetition there's no reliable data and therefore no conclusion.
• No measurement possible. If the outcome can't be recorded objectively, it becomes an opinion with charts.
• The cause is already known. Improve straight away instead of spending months proving what everyone already knows.
• Speed is the real problem. Then Lean is faster and cheaper.
• Creative or one-off work. There, variation isn't a flaw — it's the point.

Pitfalls

• Measuring without a deadline. Put an end date on the Measure phase.
• Chasing 6σ everywhere. The step from 3σ to 4σ delivers the most; going further is often not worthwhile.
• Skipping the CTQ. Without a customer requirement, you can't define a defect or prove a result.
• Skipping Control. Without monitoring and an owner in the line, the gain slips back within a few months.

Sources

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