Skip to content
LeanSuite

Practical guide · Kaizen and continuous improvement

DMAIC: the five phases, the tools and gate for each, an example, and DMAIC vs PDCA, 8D and A3

Short answer

DMAIC (define, measure, analyze, improve, control) is a five-phase method for fixing a process problem when you do not yet know the cause and have to find it in data. Each phase has its own tools and an exit gate: the team shows the output, and the sponsor signs it off, before the next phase starts.

DMAIC is said duh-MAY-ick. Define writes the problem as a number. Measure checks the gauge and sets a baseline. Analyze finds the causes the data supports. Improve tests a change against the baseline. Control writes the new method into the standard and hands it to the process owner. It is the core method of Six Sigma, though plants use it without belts. If you already know the fix, make the fix. Against PDCA, DMAIC adds a measurement phase and gates. Against 8D, it targets a chronic problem, not one event.

Last reviewed:

What is DMAIC? What the five letters stand for

DMAIC stands for define, measure, analyze, improve and control. The order is the point: a team may not start on a fix before it has a baseline and a cause the data supports, or close before the gain has held. The Lean Enterprise Institute's lexicon calls it a structured approach to removing defects and variation.

It is the project method inside Six Sigma, so belts and tollgates surround it. You do not need a belt: a charter, a good gauge, plain charts and a sponsor who holds the gates will do.

Some companies add a first step, Recognize, to decide whether the problem deserves a project (RDMAIC). DMADV, also called DFSS, is the sister method for something that does not exist yet.

DMAIC and DMADV side by side
DMAICDMADV
Use it forA process that already runsA new product or process design
PhasesDefine, measure, analyze, improve, controlDefine, measure, analyze, design, verify
Last two phasesImprove the process, then control it to hold the gainDesign the new one, then verify it meets requirements

Source: Wikipedia, Six Sigma.

Six Sigma is also tied to 3.4 defects per million opportunities, but that is not part of DMAIC. Wikipedia says the aim is to improve every process, not always to reach 3.4. The DPMO calculator puts a defect rate on that scale.

The five DMAIC phases: tools and the exit gate for each

The five phases of DMAIC are define, measure, analyze, improve and control. Define sets the problem and goal, Measure takes a baseline you can trust, Analyze finds the cause in data, Improve tests a fix, Control makes it stick. Each phase ends at a gate. If the team cannot show the gate, the phase is not done, however many meetings it has held.

The five DMAIC phases, each with the tools and the gate that closes it

  1. Define

    What is the problem, how big is it, what is the goal?

    Main toolsProject charter, SIPOC, voice of the customer, critical to quality (CTQ)
    Exit gateCharter signed by the sponsor: problem and goal as numbers, scope, team.

  2. Measure

    How does the process perform now, and can the numbers be trusted?

    Main toolsData collection plan, check sheet, gauge R&R, process map
    Exit gateA baseline for the charter's metric, long enough to include normal swings.

  3. Analyze

    What causes the gap between the baseline and the goal?

    Main toolsPareto chart, fishbone, 5 Whys, scatter plot, hypothesis tests
    Exit gateEach cause to act on is backed by data. Ruled-out causes are written down.

  4. Improve

    What change removes the cause, and does it work?

    Main toolsBrainstorming, FMEA, a pilot run as a PDCA cycle, design of experiments
    Exit gateA pilot result against the baseline, and a rollout plan.

  5. Control

    How do we keep the gain?

    Main toolsStandard work, control plan, control chart, response plan, audits
    Exit gateNew standard in use, result held for the agreed period, owner has taken over.
Read left to right. A phase is finished when its gate can be shown, not when its meetings are over. If a gate cannot be shown, the team goes back a step and does not carry on.
  1. 1

    Define: write the problem as a number

    State what is wrong, where, since when, how big and what it costs. Leave out cause and fix. The project charter holds the problem, goal, scope and team, and a SIPOC draws the edges of the process. Ask what the customer needs (critical to quality).

  2. 2

    Measure: check the gauge, then take the baseline

    If two people measure the same part and disagree, a baseline built on their numbers is noise. Run a gauge R&R study first (the gage R&R study sheet is one way), then collect the baseline over several weeks, not one day. The process capability calculator fits a measured output with limits.

  3. 3

    Analyze: split the data until a pattern shows

    Split it by machine, shift, product, supplier or tool age. A Pareto chart picks the biggest category. A fishbone or 5 Whys lists candidate causes, and the data accepts or rejects each. A cause nobody tested is a guess.

  4. 4

    Improve: test the simplest change small

    Pick the simplest change that removes a confirmed cause and pilot it, a PDCA cycle inside DMAIC. An FMEA on the change asks what it could break. Compare the pilot with the baseline, counted the same way.

  5. 5

    Control: make the gain stay

    Gains slip away here. Write the new method into standard work, put the check on a control plan, chart the metric on a control chart so drift shows early, and agree who acts when the number moves. Close after the result has held for the charter's period.

Can the team pass the Measure gate?

  1. Does a second person get the same reading on the same part?

    If no

    Fix the gauge or how it is used, then run the check again. Do not collect a baseline yet.

  2. Does the baseline run long enough to include normal swings, such as shifts, tool changes and material lots?

    If no

    Keep collecting. One good or bad day is not a baseline.

  3. Is the metric counted the way the charter counts it?

    If no

    Recount, or have the sponsor restate the charter number, so the goal and the baseline are the same measure.

  4. Every answer right

    The sponsor signs the baseline and the team starts Analyze.

Three questions in order. If a second person does not get the same reading, fix the gauge. If the baseline is too short to show normal swings, keep collecting. If the metric is counted differently from the charter, recount. Answer all three well and the baseline is signed and Analyze starts.

The gate is three questions. A no sends the team back to the step that failed, not forward.

A worked DMAIC example from a machining cell

Illustrative example. Cell 4 machines aluminum valve bodies. Scrap has been "around six percent" for a year, everyone has a theory, and the plant manager asks for a project.

Illustrative: a valve body scrap project through the five phases

Example numbers

  • Define

    What the team did
    Charter from 13 weeks of records: 975 of 16,250 valve bodies scrapped (6.0 percent) at $38 a part, $37,050 a quarter. Goal: 3.0 percent.
    What it found or decided
    In scope: cell 4, all shifts. Out: the casting supplier.

  • Measure

    What the team did
    Two operators measured the same parts with the dial bore gauge and differed by up to 0.02 mm on a 0.05 mm tolerance band. An air gauge cut that to 0.004 mm. Then every scrapped part was logged for four weeks.
    What it found or decided
    300 scrapped in 5,000 parts, 6.0 percent, with a reason for each (log below).

  • Analyze

    What the team did
    Plotted bore diameter against parts since the last insert change. Inserts were changed every 600 parts.
    What it found or decided
    147 of the 171 oversize bores came from parts 451 to 600 of an insert's life. Oversize rates were about equal on all three shifts, so operators were ruled out.

  • Improve

    What the team did
    Insert changed every 450 parts for four weeks. An FMEA on the change added a check that each new insert is seated.
    What it found or decided
    Oversize bores fell from 171 to 18 in 5,000 parts. Total scrap fell from 300 to 147, which is 2.94 percent.

  • Control

    What the team did
    450-part limit set in the tool-life counter, standard work and control plan. Weekly chart of oversize bores per 5,000 parts, a response plan, handover to the supervisor.
    What it found or decided
    Close when scrap has held at or below 3.0 percent for eight weeks.
Each phase answers the one before it: the gauge is checked before the baseline, the baseline names the defect, and the data names the cause.

The sums: 975 / 16,250 is 6.0 percent, and 975 x $38 is $37,050. The four-week log covers 5,000 parts, 1,250 a week, and 300 were scrapped, also 6.0 percent.

Illustrative: the four-week scrap log from the Measure phase
Scrap reasonPartsShareCumulative share
Bore oversize17157%57%
Porosity6321%78%
Thread damage3612%90%
Surface scratch186%96%
Other124%100%
Total300100%100%

Illustrative: the four-week scrap log from the Measure phase, as a Pareto chart

Example numbers
  • Bore oversize: 171, 57% of the total, cumulative 57%.
  • Porosity: 63, 21% of the total, cumulative 78%.
  • Thread damage: 36, 12% of the total, cumulative 90%.
  • Surface scratch: 18, 6% of the total, cumulative 96%.
  • Other: 12, 4% of the total, cumulative 100%.

Bars for scrapped parts in four weeks: bore oversize 171, porosity 63, thread damage 36, surface scratch 18, other 12, with the cumulative line at 57, 78, 90, 96 and 100 percent.

Bore oversize is 171 of 300, 57 percent of the scrap. It is the only bar the project goes after, which is why the goal of 3.0 percent can be reached without touching the other four.

Bore oversize is 171 / 300, or 57 percent of the scrap, and 171 / 5,000, or 3.42 points of the 6.0. Parts 451 to 600 are 150 of every 600, which is 25 percent of production, yet they gave 147 of the 171 oversize bores, or 86 percent. After the change the other reasons stayed at 129 parts (63 + 36 + 18 + 12), so scrap was 129 + 18 = 147 parts, or 2.94 percent, just inside the goal. That is 153 fewer scrapped parts per 5,000, worth 153 x $38 = $5,814. A year is 13 blocks of 5,000 parts, so the saving is 13 x $5,814 = $75,582, less about $430 for the 36 extra inserts at $12 each.

DMAIC vs PDCA, 8D and A3: which one to use

The four overlap. What separates them is its size and how much proof it needs.

DMAIC next to PDCA, 8D and A3

  • DMAIC

    What it is
    Five phases with gates
    Best for
    A chronic problem, cause unknown, findable in data
    Typical length
    Months
    Proof it asks for
    A checked gauge, a baseline, tested causes

  • PDCA

    What it is
    Four steps in a loop: plan, do, check, act
    Best for
    A change to test, or the daily loop of small improvements
    Typical length
    Days to weeks, repeated
    Proof it asks for
    Enough to compare before and after

  • 8D

    What it is
    Eight steps in a team report, with containment
    Best for
    One failure or complaint that needs containment
    Typical length
    Days to weeks
    Proof it asks for
    A verified root cause and fix

  • A3

    What it is
    One page that tells the problem-solving story
    Best for
    A problem one owner can work through, with a coach
    Typical length
    Weeks
    Proof it asks for
    Whatever the story needs
The four methods answer different sizes of problem. DMAIC asks the most proof and takes the longest, and it often contains the others.

They fit together. A DMAIC project runs PDCA cycles inside Improve, and an A3 can write it up. A complaint mid-project usually gets its own 8D report. See the PDCA examples, the A3 guide and problem solving methods compared. The lean method navigator suggests one.

When not to use DMAIC

DMAIC is slow on purpose. It pays when the cause is unknown and the data can find it. On a problem you can already fix it wastes weeks, which is why Businessmap's DMAIC page warns it can be heavy for simple, obvious problems. A missing guard needs repair and a line in the standard, not a charter.

Which method for which situation

  • Situation

    You already know the cause and the fix

    Use

    Make the fix, check that it worked, update the standard

    A project adds weeks, not information.

  • Situation

    One event or complaint, containment needed now

    Use

    8D or a short problem report

    The urgent work is containing this event.

  • Situation

    Cause unknown, one owner, a few weeks

    Use

    5 Whys or an A3

    Enough structure for one person's problem.

  • Situation

    Cause unknown, output varies, data exists, several functions involved

    Use

    DMAIC

    The job the method was built for.

  • Situation

    The product or process does not exist yet

    Use

    DMADV or a design process

    DMAIC improves a process already running.

  • Situation

    There is no way to measure the problem

    Use

    Build the measurement first

    Measure cannot set a baseline without one.

DMAIC is the right answer in one row of six. In the others a smaller method, a fix, or a measurement comes first.

Two questions decide it: can you write the problem as a number, and is the cause still unknown? If both are yes, write a charter.

Common DMAIC mistakes

Five DMAIC mistakes and what to do about each

  • Mistake

    A solution written into the charter

    What goes wrong: Analyze becomes a hunt for proof of the idea.

    Do this instead

    State the problem as a number. Let Analyze find the cause.

  • Mistake

    Skipping the measurement check

    What goes wrong: The baseline is partly gauge error, so a real gain hides in noise.

    Do this instead

    Have two people measure the same parts first.

  • Mistake

    A scope that covers the plant

    What goes wrong: Too many causes: the project stalls in Analyze.

    Do this instead

    One process, one product family, one defect type.

  • Mistake

    Improve before Analyze

    What goes wrong: A favorite fix, with the cause untested.

    Do this instead

    No solutions until a cause has data behind it.

  • Mistake

    No real Control phase

    What goes wrong: The gain drifts away and nobody sees it go.

    Do this instead

    Standard work, control plan, response plan, hold period, owner.

Most of the fixes come to one rule: do not move to the next phase on a belief. Move on a gate that was shown.

DMAIC templates, tools and software

Start with the DMAIC project charter, a free form with a tollgate checklist, and the DMAIC storyboard, which lays the five phases out as five columns on one board. The DMAIC examples outline 16 projects. For numbers, try the first pass yield calculator and the Pareto chart maker.

In LeanSuite, Kaizen & Projects manages every improvement project in one visual hub, from a small kaizen event to a full DMAIC project. Lean Studio has a DMAIC storyboard among its ready-made boards.

Continuous Improvement Software

Run a DMAIC project in LeanSuite

Use A3, DMAIC, PDCA, PDSA or just-do-it templates, or define your own method and phases.

See Continuous Improvement SoftwareBest continuous improvement software

Go deeper

This page is the short version. These guides cover each part in full.

All our articles on this subject are in the kaizen and continuous improvement topic.

Free templates and tools

Key terms

Lean glossary

Doing this in LeanSuite

See it on your own lines

Bring the problem you are working on to a 45-minute demo. Or look around with sample data first, or pilot LeanSuite on your own lines for 90 days, with the setup done by our team.

Book a 45-minute demoClick through the interactive demoStart a 90-day pilot

FAQ

Questions about dMAIC

  1. Define, measure, analyze, improve and control. Define writes the problem, goal and scope, and Measure checks the gauge and sets a baseline. Analyze finds the causes the data supports, Improve tests a change against the baseline, and Control writes the new standard and hands it to the process owner. Each phase ends at a gate the sponsor signs off.

  2. DMAIC is a five-phase method for solving a process problem when the cause is unknown. It comes from Six Sigma. Each phase has tools and a gate, so the team measures a baseline, proves the cause with data, tests a fix against the baseline and then makes the gain last.

  3. Duh-MAY-ick, with the stress on the middle syllable.

  4. DMAIC improves a process that already runs. DMADV (define, measure, analyze, design, verify) is for a product or process that does not exist yet, so its last two phases design and verify something new instead of improving and controlling something old. Pick by asking whether there is a running process to measure.

  5. We have no survey figure for how many plants use it, so we will not quote one. What does still apply is the discipline: write the problem as a number, check the gauge, take a baseline, test the cause and keep the result. Plants run that with or without belts, and DMAIC is one way to organise it.

  6. A machining cell scrapped 6.0 percent of valve bodies. Measure showed 57 percent of the scrap was oversize bores. Analyze showed 86 percent of those came from the last quarter of each cutting insert's life. Improve changed the insert every 450 parts instead of 600, cutting scrap to 2.94 percent. Control put the limit in the tool-life counter and the standard work. The figures are illustrative.

  7. PDCA is a short loop of plan, do, check and act, suited to testing a change and to daily improvement. DMAIC adds a define phase, a measurement check, a baseline, tested causes and a control phase with gates, which suits a chronic problem whose cause is unknown. A DMAIC project runs PDCA cycles inside its Improve phase.

  8. Do not use it when you already know the cause and the fix: make the fix and update the standard. Do not use it for a single event that needs containment now, where an 8D fits, or when the problem cannot be measured. It is also the wrong tool for a product or process that does not exist yet, where DMADV or a design process fits.

  9. No. Belt training helps with the statistics in Measure and Analyze, but a team can run DMAIC with a charter, a checked measurement, plain charts such as a Pareto and a run chart, and a sponsor who holds the gates.

Related guides and tools

More method guides

Or browse the topics, the full library and the lean glossary.

Rather see it on your own problem?

Bring the problem you are stuck on, and we will show you on a call how LeanSuite handles it.

Schedule Demo

Pass it on

We want LeanSuite to be the best place on the internet for lean help. If this was useful, send it to someone on your team or in your network who needs it.