Six Sigma Project Checklist

Charter covers problem statement, business case, goal statement (with measurable target — e.g., reduce scrap from 4.2% to 1.5%), in/out of scope, project timeline, and named sponsor. Tie the goal to a CTQ the customer or plant manager actually cares about; vague charters drift for months.

Build a SIPOC (Suppliers, Inputs, Process, Outputs, Customers) to bound the project. Walk the gemba with the process owner before drafting — desk-built SIPOCs miss handoffs the floor knows about.

Name the executive sponsor, process owner, Black Belt / Green Belt lead, and 3-5 SMEs from production, quality, maintenance, and engineering. Lock in standing meeting cadence — a project without a recurring tollgate review stalls.

Translate the Voice of the Customer into Critical-to-Quality measures with target and spec limits. Examples: surface finish Ra ≤ 1.6 µm, on-time delivery ≥ 98%, first-pass yield ≥ 95%.

Sponsor reviews charter, SIPOC, CTQs, and team. No phase advance until tollgate is signed. Common gotcha: charter goal stated in activity terms ("reduce defects") not measurable terms ("reduce DPMO from 18,000 to under 6,000").

Specify each measurement: what is measured, how, by whom, sampling frequency, sample size, data type (variable vs. attribute), and where data is logged (MES, paper traveler, SPC software). Plan must support enough samples to compute Cpk reliably — typically n ≥ 30 per condition.

Conduct a Type 2 gauge R&R study (typically 3 operators × 10 parts × 2 trials) in Minitab or JMP. Acceptable: %study variation under 10%; marginal: 10–30%; reject: over 30%. Without a passing MSA every downstream conclusion is suspect.

Triggered when gauge R&R fails. Investigate calibration status, fixturing, operator technique, environmental conditions. Re-train operators or swap to a higher-resolution gauge before resuming data collection.

Run the data collection plan across enough shifts and operators to capture between-shift and between-operator variation. Stratify by shift, operator, machine, and material lot. Common gotcha: pulling only first-shift data and missing the night-shift signal.

Compute baseline Cp, Cpk, Pp, Ppk, DPMO, and process sigma. Confirm normality before using Cpk; if non-normal, transform or use percentile method. This is the number the project's success will be measured against — record it precisely.

Run an Ishikawa session with operators, setup techs, and the process engineer. Use the 6M categories: Man, Machine, Method, Material, Measurement, Mother Nature. The exercise is worthless without floor participation — engineers in a conference room produce theoretical causes, not real ones.

Pareto the fishbone causes; 5-why the top 2-3 to drive past symptoms to systemic causes. Stop when the next "why" exits the team's span of control or hits a policy choice.

Use t-tests, ANOVA, regression, or chi-square (per data type) in Minitab to confirm which suspected causes are statistically significant. Don't skip this — fishbone-only projects implement fixes for non-causes and the metric doesn't move.

For multi-factor problems, run a 2^k factorial or fractional factorial DOE. Confirm with the process owner that the experimental runs are scheduled into production without blocking customer orders.

Present the validated root causes (with p-values), the data behind them, and the proposed direction for Improve. Sponsor signs off before solutions are designed.

Generate solutions tied to validated root causes. Score with an effort/impact matrix or Pugh matrix. Favor poka-yoke (mistake-proof) solutions over training-only fixes — training fades, fixturing doesn't.

Any process change requires PFMEA revision — new failure modes, updated RPNs, revised control plan. Common gotcha: rolling out a change without re-scoring failure modes the change introduces.

Run the change for at least 2 weeks or 1,000 parts (whichever is greater) on a single shift. Compare pilot Cpk against baseline using a hypothesis test, not eyeball comparison.

Triggered when pilot doesn't hit target. Re-run the cause analysis with pilot data — usually a secondary cause that wasn't picked up in Analyze, or a poorly-implemented countermeasure. Adjust and re-pilot before scaling.

Cascade the change across shifts. Update work instructions, visual aids, and ERP/MES routing. Issue an ECN if drawings or specs change. Confirm operator acknowledgment of the new revision before run release.

Per AIAG control plan format: characteristic, spec, measurement method, sample size, frequency, control method, reaction plan. The reaction plan is the part most teams skip — "if X-bar exceeds UCL, stop the line and call the cell lead" must be explicit.

Stand up X-bar R, individuals, or p-charts (per data type) at the work center. InfinityQS, Minitab, or paper charts all work — what matters is that operators can see the chart from the machine and know the Western Electric rules.

Deliver TWI-style training on the new procedure across all shifts. Capture training acknowledgments in the LMS or paper sign-off; an OSHA or ISO auditor will ask for proof. Update the skills matrix.

30 days post-rollout, recompute Cpk and DPMO from production data. The improvement must hold under real conditions — multi-shift, multi-operator, varying material lots — not just under the pilot.

Formal handoff: process owner accepts the control plan, SPC charts, training records, and updated PFMEA. Finance signs off on validated savings. Archive the project A3 in the CI repository for replication on similar processes.