DMAIC Five-Stage Complete Map: What Should Be Delivered at Each Stage —— A Table to Understand Deliverables and Gate Criteria from Project Initiation to Closure

In 2023, an automotive parts company initiated 11 Black Belt projects, all of which "passed the five gates on schedule" by the end of the year. The average review time per gate was 28 minutes, and the reported financial benefits at project closure were 18.6 million yuan. However, the finance department only recognized 2.4 million yuan, with an approval rate of 13%.

Upon reviewing the records, the issues were clear: the gate deliverables for the M stage were a 12-row Excel sheet without sample size or gage analysis; the A stage deliverables were three fishbone diagrams and the statement "the root cause is improper operation"; the C stage had no updated control plan, only a training sign-in sheet. The projects were not poorly executed; they were too easily passed through each gate—each gate only checked whether the work was done, not whether the deliverables could support the next step.

1. Conclusion First: DMAIC Is Five Handovers, Not Five Chapters

The true meaning of the five stages is five responsibility handovers: the deliverables from the previous stage must be sufficient to allow the next stage to begin. Therefore, to determine whether a stage is complete, we do not look at how smoothly the presentation is given, but rather at whether the deliverables are present and sufficient.

This table is the backbone of the article. The following sections will detail the most common areas where each cell can be watered down.

2. D Define: Gate Deliverables Are Not Application Reports

The most common pseudo-deliverables are of three types: the charter is written as a project application (focusing entirely on significance without a baseline or scope); the SIPOC is only drawn at the department level, with S being "supplier" and P being "production department," and cross-departmental issues are still kicked around; the CTQ tree directly uses "customer satisfaction" as an indicator, which is not measurable.

At the D stage gate review, only three questions are asked: What is the current value of Y? Who measured it and how? How much financial value will improving it bring, and does finance recognize it? If any of these questions cannot be answered, the project should not proceed to the M stage—because all measurement work in the M stage is based on "Y being clearly defined." For details on the six elements of the charter and the gate criteria, refer to DMAIC Project Charter and Stage Gates, which will not be repeated here.

A practical criterion: give the charter to an unrelated person to read. Can they state "the current value, the target value, the starting line, and the end time"? If not, it is not complete.

3. M Measure: Unreliable Baseline Means All Subsequent Work Is Ineffective

The gate deliverables for the M stage are only two solid items: measurement system analysis and baseline.

First, look at the gage. If the same batch of data is measured by different people using different methods, the A stage calculations will be noise. The GR&R criteria cannot be relaxed:

Next, consider the baseline. There are three strict requirements: the sample must cover at least 25 subgroups or a continuous 3-month period; the data terms must be consistent with those of the customer and finance (customers count PPM by piece, internal counts defect rate by batch, the two numbers will never align); the sampling method must be clearly stated (continuous or random sampling, whether it includes changeovers and first articles). For the conversion relationship between terms and capability, see What Is the Difference Between Cpk and Ppk.

The most hidden pseudo-deliverable in the M stage is "using historical reports as the baseline." Historical reports typically only count incoming inspections, missing the repaired and qualified parts, thus the baseline is naturally biased, making the improvement space appear twice as large.

4. A Analyze: Root Causes Must Be Turned Off and On

The gate deliverables for the A stage are not fishbone diagrams, but root causes that can be verified on-site. The sequence must be: layering (machine/batch/shift/time) → graphical (multi-vari chart, box plot, scatter plot) → hypothesis testing or regression → on-site reproduction. Skipping the first two steps and directly running hypothesis tests is equivalent to betting on a p-value without understanding the data structure.

The standard actions for root cause verification are three, all of which are essential: on-site reproduction (manually creating defects under the assumed conditions and verifying their recurrence); turning off and on (returning the factor to normal, and the defect disappears); time sequence alignment (the time points of factor changes align with the time points of defect outbreaks). Meeting all three criteria confirms the root cause; meeting only one makes it a suspect.

Two other common errors: a significant p-value is inevitable with a sufficiently large sample size, but the effect size may be too small to justify improvement—focus on the confidence interval and actual difference, not the p-value itself; correlation is mistaken for causation, especially coincidences like "defect rate decrease" and "training completion rate increase" occurring simultaneously. For a more comprehensive methodological discussion, see Six Sigma Hypothesis Testing in Practice and SPC Overview.

5. I Improve: Solutions Must Be Attributable

The gate deliverables for the I stage are screening matrix + pilot data + poka-yoke design. Solution screening should score at least four columns: expected effect, implementation cost, risk (to other characteristics, to cycle time), and implementation period; the weights should be set before the meeting and not adjusted on-site.

Pilot design determines whether the benefits can be clearly calculated. The three most common methods are: before-and-after comparison on the same machine (requires a stable process, supported by control charts); ABAB alternation (excludes time trends); and concurrent comparison with a control line (the most expensive but most reliable). Simply stating "better after improvement" is insufficient, especially when seasonal or order structure changes are involved.

The most common directional error is treating "strengthening inspection" and "全员培训" (全员培训 = full-staff training) as improvement measures. Inspection does not reduce the defect rate; it only prevents defects from leaving the factory. Training changes people's attention, not process parameters. These actions can be retained but should not be the main measures to promise Cpk improvement. For related cases, see I Stage Solution Screening and Pilots.

6. C Control: Document Signatures Do Not Equal Control

The gate deliverables for the C stage are four items: updated control plan page (clearly stating what to control, the method, frequency, and who is responsible), standard work and SOP revisions, response plan (what to do within a certain time if the control limit is exceeded, including line stop authority), and transfer and training records.

To pass the control chart through the gate, two questions must be answered: how were the control limits calculated (using baseline period data, not guesswork or directly using old charts); who is monitoring (patrol inspection or shift leader, and how out-of-control points are recorded). A control chart without a responsible person will certainly become a wall decoration after three months.

Another item often overlooked in the C stage is a three-month tracking plan. Project closure does not mean the end; the benefits must be verified for three consecutive months after closure to confirm they are not just "review period effects." For a complete logic of solidifying results, see Six Sigma Improvement Results Solidification in Practice.

7. Gate Review: Eight Typical Nonconformities and One Rejection Rule

Turning gate reviews into presentation meetings is the most common way this method fails in domestic implementation. Reviews only check eight criteria, and hitting any one of them means the project cannot pass the gate:

1. Gate deliverables are only PPT conclusions without original data or records;
2. The baseline does not clearly state the sample size, time span, and sampling method;
3. Root causes are only qualitatively described without verification records;
4. Improvement effects are only compared before and after without excluding time trends and other changes;
5. Benefits are not confirmed by financial terms, or "loss avoidance" is directly counted as cash benefits;
6. The control plan is not updated, or it is updated without a responsible person;
7. The gate review has no rejection records—unanimous approval often means no one is truly reviewing;
8. The gate review is conducted by the project manager themselves.

Accompanying this is one rejection rule: gates must simultaneously review "deliverables from the previous stage" and "plans for the next stage," and neither can be missing; rejections must be written with specific conditions and a re-review date, not just "improve it." For experience in rebuilding project initiation mechanisms, refer to Black Belt Projects: Choosing the Wrong Problems Leads to Greater Losses the Longer They Are Worked On.

8. Case Development: From Presentation Meetings to Handover Meetings

The following year, the company made three changes without changing personnel or increasing the budget:

1. Gate Checklist: Each stage lists 6-10 deliverables, and any missing item means the project does not pass; the checklist is posted on the project board.
2. Financial Pre-Approval: The financial benefit confirmation form is signed off in the D stage, and all subsequent data is collected using the same terms to avoid reconciling with finance at the project closure.
3. Rejection Authority Transfer: The gate rejection authority is transferred from the project manager to the Black Belt supervisor and finance representative, with the project manager only responsible for reporting.

Result: In the second year, out of 11 initiated projects, 4 were cut in the D stage (all due to unclear CTQ or invalid financial terms), the remaining 7 projects had an average cycle time extended from 5 months to 7 months, and the recognized financial benefits increased from 2.4 million yuan to 6.2 million yuan. The cost was real: the "project quantity" performance metric decreased, and the head of the promotion office and the production director openly clashed at the budget meeting; in the third year, the performance metric was changed to "recognized benefits created by each Black Belt," which resolved the conflict.

In summary, the essence of this change is: the five gates are not five chapter titles, but five handovers—deliverables that cannot support the next step should not proceed.

9. Self-Inspection Checklist

• Does each stage have a fixed list of gate deliverables, and can missing items be blocked on the spot?
• Does the baseline clearly state the sample size, time span, and sampling method, and are the data terms consistent with those of the customer and finance?
• Have the root causes in the A stage been verified on-site by "turning them off and on"?
• Have the effects in the I stage excluded time trends and concurrent other changes?
• Does the control plan in the C stage have a responsible person and response plan for each item?
• In the last three gate reviews, have there been any rejection records?

One actionable sentence: the five gates are not five chapter titles, but five handovers—deliverables that cannot support the next step should not proceed.

Gates review deliverables, not words; the five gates are five handovers.

Knowledge code: 6.1.1

Version: v20261103

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping companies continuously improve their quality capabilities.