DMAIC

Thinking Tools4.1 Root Causes and the Case for Going BackAlso known as Six Sigma Improvement Cycle, Define, Measure, Analyze, Improve, Control

From Six Sigma at Motorola, developed by Bill Smith (1986) and Mikel J. Harry (The Nature of Six Sigma Quality, 1988).

Runs a process improvement through five stages (define, measure, analyze, improve, control), ending with monitoring that keeps the gain from slipping back.

Five stages, each run on a guess or on the data

At each stage of the wrong-invoice project, pick the lead's move: the quick one or the one DMAIC asks for. The data's answer shows, with a verdict, and the count under the panel tallies the stages run on evidence. Illustrative.

Project lead

What the data shows

Project lead

What the data shows

Select a stage to read it, or play the sequence.Define. Customers are complaining about wrong invoices. Pick how the lead writes the project's goal.Pick a move to see the reply.Four months later nobody can say whether invoicing is fixed: the goal had no number to reach.A wish: nothing to measure success against.The goal has a rate, a date, and a boundary: receivables only, not purchasing.A defined problem: success can be checked.The charter. The scope keeps the project small enough to finish and the target makes it possible to say when it has.Measure. The billing team says errors are rare. Pick how the lead finds the baseline.Pick a move to see the reply.A few a week, they think, mostly from the new clerk.A belief, and it undercounts: the team only sees the errors customers report.26 of 400 are wrong: 6.5 percent, three times what the team believed.A baseline: the number every later stage is judged against.The baseline. The count also records each wrong invoice's source, which the next stage needs.Analyze. Everyone points at the new clerk. Pick the lead's next move.Pick a move to see the reply.Her error rate was already the lowest in the team. The 6.5 percent does not move.A name, not a cause: the fix misses the process.21 came from orders a salesperson changed after the invoice was drafted. Keying slips: 2.A root cause the data supports.The root cause. Four in five wrong invoices began with a change to the order after billing had started.Improve. The fix is a hold: no invoice until the order is locked. Pick how to introduce it.Pick a move to see the reply.Quarter end hits in week two, invoices pile up, and three branches quietly drop the hold.Untested at scale: when it wobbles nobody can tell why.Pilot branch: 1.4 percent wrong over 380 invoices. Cash arrives two days later, which finance accepts.A proven change, with its cost known.The pilot. The same measure as the baseline, so the drop from 6.5 percent is a fair comparison.Control. The hold is now live at every branch. Pick how the lead closes the project.Pick a move to see the reply.Six months later: 5.8 percent. Sales found a way around the hold at quarter end and nobody was looking.No control: the gain slipped back.Six months later: 1.6 percent. The trigger tripped once, in week 14, and traced to one rep's workaround.A control: drift becomes a signal someone answers.Six months on. The project closed when the billing manager owned the chart and the written procedure.

Stages run on the data: 0 of 5Stages run on the data: 1 of 5Stages run on the data: 2 of 5Stages run on the data: 3 of 5Stages run on the data: 4 of 5Stages run on the data: 5 of 5

  1. Define (Project lead). Customers are complaining about wrong invoices. Pick how the lead writes the project's goal.
    • Fix our invoicing problems. Four months later nobody can say whether invoicing is fixed: the goal had no number to reach. A wish: nothing to measure success against.
    • Cut wrong invoices in receivables from today's rate to under 2 percent by June. The goal has a rate, a date, and a boundary: receivables only, not purchasing. A defined problem: success can be checked.
  2. The charter (What the data shows). The scope keeps the project small enough to finish and the target makes it possible to say when it has.
  3. Measure (Project lead). The billing team says errors are rare. Pick how the lead finds the baseline.
    • Ask the billing team how often it happens. A few a week, they think, mostly from the new clerk. A belief, and it undercounts: the team only sees the errors customers report.
    • Pull 400 invoices from last quarter and check every one. 26 of 400 are wrong: 6.5 percent, three times what the team believed. A baseline: the number every later stage is judged against.
  4. The baseline (What the data shows). The count also records each wrong invoice's source, which the next stage needs.
  5. Analyze (Project lead). Everyone points at the new clerk. Pick the lead's next move.
    • Retrain the new clerk and watch her work. Her error rate was already the lowest in the team. The 6.5 percent does not move. A name, not a cause: the fix misses the process.
    • Sort the 26 errors by where each began and test each cause. 21 came from orders a salesperson changed after the invoice was drafted. Keying slips: 2. A root cause the data supports.
  6. The root cause (What the data shows). Four in five wrong invoices began with a change to the order after billing had started.
  7. Improve (Project lead). The fix is a hold: no invoice until the order is locked. Pick how to introduce it.
    • Switch all six branches over on Monday. Quarter end hits in week two, invoices pile up, and three branches quietly drop the hold. Untested at scale: when it wobbles nobody can tell why.
    • Pilot the hold at one branch for four weeks, measured as before. Pilot branch: 1.4 percent wrong over 380 invoices. Cash arrives two days later, which finance accepts. A proven change, with its cost known.
  8. The pilot (What the data shows). The same measure as the baseline, so the drop from 6.5 percent is a fair comparison.
  9. Control (Project lead). The hold is now live at every branch. Pick how the lead closes the project.
    • Announce the result and close the project. Six months later: 5.8 percent. Sales found a way around the hold at quarter end and nobody was looking. No control: the gain slipped back.
    • Chart a weekly sample of 50 invoices; the billing manager owns it; a trigger at 3 percent. Six months later: 1.6 percent. The trigger tripped once, in week 14, and traced to one rep's workaround. A control: drift becomes a signal someone answers.
  10. Six months on (What the data shows). The project closed when the billing manager owned the chart and the written procedure.

An example

A distributor of industrial parts was losing customers over wrong invoices. The finance director asked an analyst to run it as a DMAIC project with a four-month deadline.

She defined it narrowly: cut wrong invoices from accounts receivable to under 2 percent by June. The billing team guessed errors were rare, so she measured: of 400 invoices from the last quarter, 26 were wrong, 6.5 percent. Everyone blamed a new clerk. Sorting the 26 by source told another story: 21 came from orders a salesperson amended after the invoice was drafted, and only two were keying slips. The fix she piloted at one branch was a hold, with no invoice drafted until the order was locked. Over four weeks the pilot branch ran at 1.4 percent wrong. Before closing the project she set up control: a weekly sample of 50 invoices on a chart, owned by the billing manager, with a trigger at 3 percent. Six months later the rate was 1.6 percent.

Why it works

Each stage refuses to let the next one start on a guess. Define pins the problem to a number and a boundary, so success can be checked. Measure replaces what people believe about the process with what it does, which is often far from the belief. Analyze tests causes against the data before anyone spends money on a fix, which is how the clerk escaped blame. Improve tries the change small, where a failure is cheap and a success has proof attached.

Control is the stage most improvement efforts skip, and the reason DMAIC gains last. A process left alone drifts back to its old habits as people find workarounds. A sampled measure with an owner and a trigger turns that drift into a signal someone has to answer, so the fix stays fixed after the project team moves on.

The thinking behind it

Bill Smith Bill Smith, a quality engineer at Motorola, introduced Six Sigma there in 1986: a program for cutting defects to a few per million opportunities. The improvement cycle at its center, developed by Mikel Harry, began as four stages, measure, analyze, improve, and control. The define stage that makes it DMAIC was added when General Electric adopted Six Sigma in the mid-1990s, with no single named author.

Mikel J. Harry Mikel J. Harry, a Motorola engineer often called the father of Six Sigma, developed the method with Smith and wrote The Nature of Six Sigma Quality (1988). He explained the statistics behind the program's famous target of 3.4 defects per million, which assumes a process drifts by 1.5 standard deviations over time. That shift rests on his tolerancing argument about a stack of discs; it was never drawn from measured process data.

How to run it

  1. Define the problem as a number. Write the current rate, the target, the date, the customer it matters to, and the boundary of the process in scope.
  2. Measure before anyone fixes. Sample the process as it runs today and count, even when the team is sure it knows the answer.
  3. Split the defects by source. Sort every defect by where it began, and test each suspected cause against the counts.
  4. Confirm the root cause. Keep the cause that explains most of the defects and that the data, not seniority, points to.
  5. Pilot the fix small. Try the change at one site or on one shift, with the same measure as the baseline, for long enough to see a real difference.
  6. Set the control. Give the measure an owner, a sampling schedule, a chart, and a trigger level that starts a response.
  7. Hand it to the process owner. Close the project only when the people who run the process own the control and the written procedure.

Where it goes wrong

The project starts at improve. A team that already knows the fix skips measure and analyze, rolls the change out, and cannot tell later whether it worked. Ask for the baseline number before discussing any solution.

Analyze lands on a person. Blaming the newest clerk or the night shift feels like a cause and is usually a symptom of the process. If the proposed cause is a name, split the defects by source again.

Control is a slide. A dashboard nobody owns, with no trigger, reports the drift back to 5 percent to no one. The control is real only if a named person samples it on a schedule and must act when the trigger trips.

When to use something else

  • PDCA Cycle. When the problem is small, the cause is fairly clear, and quick repeated trials will learn faster than a four-month project.
  • Five Whys. When a single incident needs its cause traced in an hour, with no data set to analyze.
  • Control Chart. When the process is stable and the need is only to watch it and tell ordinary variation from a real change.

Where it sits in SOLVE eX

Go to the source

  • Six Sigma quality program, Motorola, 1986
  • The Nature of Six Sigma Quality, Motorola University Press, 1988

Back to the catalog