Make the work better without making people run faster
Back to: White Belt
A correct order arrives late. A fast order contains the wrong item. Neither result meets the need. Improvement means making the whole job work better: the right result, when needed, without unnecessary effort or risk. That matters at a kitchen counter, a repair shop and a large company.
Start with what actually happens. A customer places an order, someone finds the item, someone packs it and someone delivers it. These connected steps form a process: inputs become an output. The next person who needs that output is your customer too. Ask them what they need. Don’t decide for them.
Then turn that need into something you can check. For a label, that means it matches the item ordered and the address it ships to. If you can’t check it, you can’t fix it.
Two useful ways to look at the same problem
Suppose a packing team walks across the room for labels, then sometimes chooses the wrong label. Removing unnecessary walking can improve the movement of work, called flow. Understanding why wrong labels occur can improve correctness and consistency.
Lean focuses on customer value, removing waste and improving flow. It includes quality and respect for the people doing the work. Six Sigma uses evidence to understand and reduce harmful variation and defects. A defect is a result that fails a defined requirement, such as the wrong delivery address. The methods can work together. There is no rule that one must always come first.
Do not remove an essential check just to beat a stopwatch. Look for a better way to prevent the mistake, and verify that the change works. Making one station faster while creating rework elsewhere can make the whole process worse.
Where the ideas came from
Lean and Six Sigma were not invented together. Their different roots help you choose a method for the job instead of chasing a label.
- 1924, statistical process control: Walter Shewhart proposed the control chart, a way to separate routine variation from signals worth investigating. His work helped establish data-based process control. ASQ’s Shewhart history.
- Mid-20th century, the Toyota Production System: Toyota built its approach over decades around jidoka (detect and respond to an abnormality, including stopping the process) and Just-in-Time (make what is needed, when it is needed). This is a major root of later Lean thinking. Toyota’s explanation.
- 1986, Six Sigma: Motorola developed its Six Sigma quality process to measure and reduce defects and variation. It drew on earlier quality and statistical thinking, but it was not the same program as Toyota’s. Motorola’s timeline.
- Early 1990s, the method spreads: Mikel Harry and Richard Schroeder developed Six Sigma deployment ideas at ABB in 1993 and later worked with AlliedSignal. This expanded the approach from defect measures to project selection, leadership support and business results. Dates vary among historical accounts for AlliedSignal’s initial adoption, so treat this as a period, not a single launch day. Harry’s historical account; 1999 industry reporting.
- 1995, wider adoption: GE brought Six Sigma into its management program, helping spread its use beyond Motorola. GE’s history.
Check your understanding: Match the control chart, jidoka and Just-in-Time, the 1986 Six Sigma program and the 1995 wider adoption to Shewhart, Toyota, Motorola and GE. Model answer: Shewhart, Toyota, Motorola and GE in that order. These are documented milestones, not proof that one organization invented every idea used today.
Use it tomorrow: If the problem is wasted movement, look at flow. If the result varies and misses a requirement, define the defect and study the pattern. If both happen, check both without assuming one branded method must come first. History explains the tools; the work decides the order.
The average can hide the problem
Here is a practice example, not an actual customer result. Two delivery routes each average 30 minutes. One takes 28, 29, 30, 31 and 32 minutes. The other takes 10, 20, 30, 40 and 50 minutes. If customers were promised delivery within 35 minutes, the second route missed that requirement twice.
The results spread differently. That spread is variation. Not every difference is harmful; compare it with the requirement and the conditions. Averages help, but they do not tell the whole story. Ask what differed between the trips before blaming a driver. Road conditions, order readiness or an unclear route may matter.
See the spread on one shared scale
Minutes 10 20 30 40 50 Scale | | | | | Route A ..................ooooo.................. Route B o.........o.........o.........o.........o Promise .........................|...............
Text equivalent: Route A took 28, 29, 30, 31 and 32 minutes. Route B took 10, 20, 30, 40 and 50 minutes. Both average 30 minutes. With a promise of 35 minutes or less, A missed zero of five and B missed two of five. The common axis matters because it reveals that the same average can hide very different customer experiences.
Read the diagram: Which route is more predictable against the 35-minute promise, and what does an average of 30 hide? Model answer: A is more consistent within this sample. B missed the promise twice. Five trips are too few to establish long-term performance, explain a cause or predict next week’s results; check route conditions and collect more comparable observations before changing the process.
What a sigma number can and cannot say
Sigma (standard deviation) describes spread. A sigma level is a different score that some Six Sigma methods use to express estimated defect performance. First define the customer’s requirement and what counts as one opportunity for a defect. Count actual defects, units checked and opportunities per unit. Never compare scores made with different opportunity definitions.
Worked practice, not a real result: In the packing example in Lesson 3, suppose exactly 10 of 100 orders have one wrong item and each order has one agreed opportunity for that kind of error. There are 100 x 1 = 100 opportunities. Defects per million opportunities, or DPMO, is (10 / 100) x 1,000,000 = 100,000 DPMO. The observed wrong-order rate is 10%. This is not a forecast that exactly 100,000 of the next million orders will fail.
Some conversion tables report this example as roughly 2.8 sigma: first use the one-sided standard normal score for 90% success, about 1.28, and then add the conventional 1.5-sigma shift allowance. That addition is a modeling choice, not a measured shift in these orders. Without that convention, the corresponding one-sided normal score is about 1.28. The famous 3.4 DPMO for a six-sigma process also uses a shifted-normal convention, not a guaranteed defect count. This sample does not establish stable performance, a normal distribution, independence or the future rate.
Decision check: Would it be fair to call ten wrong orders out of a hundred “2.8 sigma” without stating the one-opportunity rule and shift assumption? No. Show the raw count and rate first, identify the model and assumptions, and study whether the process holds over time. See the CSSC White Belt Body of Knowledge and JMP’s explanation of the 1.5-sigma convention.
Check what zero really means
Try it before reading the answer. Two fictional teams check the same one requirement per order. Team A finds no mistakes in 10 orders. Team B finds no mistakes in 1,000 orders. Can either team claim the process will never make a mistake? Which result gives more evidence, and what would you still need to know?
Worked answer. Neither team can claim zero future mistakes or an infinite sigma level. Report the actual checks: A found 0 in 10; B found 0 in 1,000. B checked more work. Before comparing results, confirm that the requirement, inspection method, work mix and time window were comparable. A sample with no observed defects cannot establish a future failure rate or explain why mistakes did not appear.
Use it tomorrow. When somebody says, “We had zero defects, so the process is perfect,” ask: “Zero out of how many, checked how, and over what period?” A small clean sample is a start, not a guarantee.
Try it: define a result worth improving
Choose a familiar low-risk task: finding homework, packing a lunch or handling a routine request. If you cannot observe real work, use the packing example. Write four lines:
- Who needs the result, and what counts as correct?
- What are the main steps from request to finished result?
- What will you count or time, including mistakes?
- Who can explain the difficulties you cannot see?
Model answer: The customer needs the correct item and label by the agreed time. Observe picking, label selection, packing and handoff. Count completed orders, wrong labels and search time. Ask the packer when labels are hardest to distinguish. That is a starting plan, not proof of a cause.
Keep these notes. Next, you will learn to separate customer value, necessary protections and avoidable waste before deciding what to change.
