Operations Deliver

DMAIC

DMAIC is define, measure, analyze, improve, control: the Six Sigma cycle for reducing variation in a process.

The questionWhat variation or defect will we reduce in this process, what causes it, and how will we keep the gain?

Define, Measure, Analyze, Improve, Control. The design counterpart is DMADV (define, measure, analyze, design, verify), also called Design for Six Sigma.

Process · Mikel Harry, Motorola, and GE under Jack Welch, 1986 to 1995

DefineCharter, customer, and process boundary
MeasureTrusted baseline of the defect
AnalyzeCause, tested against data
ImproveTrial the fix at limited scale
ControlStandard, owner, and reaction plan

One pass of DMAIC. Control holds the gain and can start the next project.

When to use it

A process has a repeatable defect or a wide spread in cycle time, yield, or errors, and the team cannot say which cause is largest.
A sponsor will fund a fix only if the baseline, the cause, and the result can be shown in numbers.
Several improvement ideas are in play, and you need a sequence that measures first and changes second.
A gain from a workshop faded within months because nobody owned the new standard.

When to avoid it

The process does not yet exist, or the work is a new product. That is a design problem. Use DMADV, Stage-Gate, or a discovery method.
The cause is already known and the fix is mandated, such as a recall or a regulation. Implement it.
You cannot get trustworthy data in the time you have. DMAIC without a measurement system is a long opinion.
The issue is organization design, strategy, or a one-off failure. A variation-reduction cycle will not answer it.

Overview

DMAIC is a five-phase project cycle for cutting variation in a process that already runs. You define a countable defect, measure a trusted baseline, analyze until one or two causes survive the data, trial a fix, and then lock the gain with a standard, an owner, and a reaction plan. It supports the choice of which cause to attack, and whether a proposed system or layout has earned the spend.

The work that teams skip is Measure and Control. Without a measurement-system check, Analyze is a workshop. Without Control, a good trial month is not an improved process. DMAIC is heavier than PDCA and the wrong tool for designing something that does not exist yet. Use it when the process is stable enough to measure and the sponsor will only move on numbers.

The five phases

Define

Name the problem, the customer, and the process boundary. Write a charter with a defect definition, a goal, a scope, a sponsor, and a timeline. A SIPOC is the usual one-page view of the process. If the defect cannot be counted, stop here and rewrite it until it can.

Measure

Document how the work actually runs and collect a baseline of the defect. Check that the measurement system is good enough before you trust the number. What comes out is a capability figure, a process map, and a data set the later phases will use.

Analyze

Find the causes that actually move the defect, using the measured data, not a workshop list. Fishbone diagrams and 5 Whys help generate candidates. Pareto charts, stratification, and designed tests decide which candidates survive. One or two causes should remain.

Improve

Design a change that attacks those causes, try it at limited scale, and compare the result with the baseline and the goal. Keep competing ideas in play until the data pick one. Do not rewrite the standard yet.

Control

Lock in what worked. Update the standard work, give it an owner, put the measure on a control chart or a standing review, and write a reaction plan for when the measure moves. Without this phase the process drifts back.

What it is not

The same as PDCA. DMAIC is a project cycle with a measurement-system check and a statistical core. PDCA is a smaller loop for a single change. Use PDCA inside Improve and Control.
A belt-training program. Belts are a staffing model. The method is the five phases and the evidence they require.
A method for designing something new. That is DMADV. Forcing DMAIC onto a blank sheet produces a fake baseline.

The steps

Write the charter and freeze the boundary

Agree the defect in one sentence, the goal, the start and end of the process, the sponsor, and the date of the first tollgate. If two people would count the defect differently, rewrite it. Output is a one-page charter.
Get a trusted baseline

Map the process as it runs, check the measurement system, and collect enough data to state the current rate and its usual range. Do not analyze causes on a number you cannot defend. Output is a baseline with the data behind it.
Reduce the cause list to what the data support

Generate candidate causes, then test them against the baseline data. Keep the one or two that move the defect. Drop the rest in writing so they do not return in Improve.
Trial the change against the goal

Run the fix at a scale you can reverse. Compare the result with the baseline and the charter goal. Adopt, adjust, or drop it before you touch the standard.
Standardize and watch

Update the procedure, name the owner, put the measure on a review, and write what happens if it moves. Set the date of the first control review before the project team disbands.

In practice

Invoice mismatches run at 3.8 percent of supplier invoices, about 190 mismatches a month on 5,000 invoices. The finance director wants a new matching system. The team writes a charter first: a mismatch is an invoice that cannot be posted on first pass, the goal is to cut that rate to below 1.5 percent in four months, and the process starts at invoice receipt and ends at posting. Measure shows the coding of mismatch reasons is inconsistent, so they fix the codes and recode six weeks of invoices. Three reasons then account for 71 percent of mismatches: quantity on the invoice disagrees with the goods-receipt record, the purchase-order price is stale, and invoices arrive without a purchase-order number.

Analyze traces the quantity mismatches to a receiving step that lets operators confirm a delivery without counting mixed-SKU pallets. Improve trials a mandatory count on mixed pallets on one dock for three weeks. Mismatches on that dock fall from 4.1 percent to 1.6 percent. Control writes the count into the receiving standard, adds a daily mismatch tally owned by the warehouse supervisor, and sets a reaction if the rate exceeds 2 percent in a week. The system purchase is deferred. The remaining two causes get their own charters.

What you leave with

A charter with a countable defect, a goal, a boundary, and a sponsor.
A baseline rate, with the measurement-system check behind it.
One or two causes that survived the data, and a list of causes that did not.
A trial result set beside the baseline and the goal.
An updated standard, a named owner, a control measure, and a reaction plan.

Failure points

Jumping to Improve

The team arrives with a preferred system or layout and uses Define to justify it. Measure and Analyze become theater, and the baseline is never used. The old defect remains.

A defect nobody can count

"Poor quality" or "slow process" is not a defect. Without an operational definition, every later number can be argued away.

Causes chosen in a workshop

A fishbone is filled in from opinion and treated as the answer. Analyze is supposed to kill most of those branches. If every bone is still on the wall at Improve, the data were never used.

Control skipped

The project presents a good month and closes. The owner of the new standard is unnamed, the measure drops off the review, and the rate returns to the old baseline.

Scope that will not fit a project

The charter covers a whole plant, a whole function, or "culture." DMAIC needs one process and one defect. A sprawling charter never reaches a trial.

The failures above are usually Leverage & IP: it was used once, and nothing was left that makes the next engagement easier.

Origin

Six Sigma began at Motorola in the mid-1980s as a way to cut process variation, measured in defects per million opportunities. Bill Smith, a Motorola engineer, is usually credited with the statistical program and the name. Mikel Harry helped turn it into a management system of projects, belts, and tollgates. Jack Welch made it mandatory at General Electric from 1995, which is how most executives first met DMAIC. The five phases are the improvement cycle for a process that already exists. New designs use DMADV.

Motorola registered Six Sigma as a trademark. The mark has since been widely licensed and used, and DMAIC itself is not licensed. Belt ranks, software, and training programs are commercial products. This note describes the five phases as they are taught in standard quality practice. It is not a substitute for a licensed Six Sigma curriculum.