A Pilot That Looks Too Good to Book
A team piloted a fixture change and revised work instructions on one assembly cell. Defects fell from 3.4% to 0.9% over three weeks. The cell ran with a dedicated engineer on site, on one shift, using a single expedited lot of incoming components. Leadership wants the savings booked into this quarter's gain-sharing plan, which requires three months of sustained results. The control chart shows the shift in level, but the last four points are drifting upward. The key operator retires in six weeks, the controlled SOP still shows the old fixture, and the supervisor proposes quarterly retraining instead of adding a sensor that would physically prevent a mis-loaded part.
1. Should the team declare the project closed and book the savings? What evidence is missing?
Read the worked answer
No. This is a validated improvement in one configuration, not a controlled process.
Missing evidence:
- Representativeness. Dedicated engineering support, a single shift and a single expedited component lot each plausibly contribute to the result. None of those is the normal operating condition, so the 0.9% is confounded.
- Stability at the new level. Four consecutive rising points is a run signal, not noise. The process is drifting back up before the handoff, which is exactly what a control chart is for catching.
- Institutionalization. The SOP still shows the old fixture, so anyone trained from the document will undo the change. The retiring operator takes tacit know-how out the door in six weeks.
- Timing. Gain-sharing requires three months of sustained results; the pilot is three weeks.
What to do: extend the run to all shifts with normal component lots and no dedicated engineer; update the SOP and train against it; re-baseline and confirm stability with the control chart; hand ownership to the process owner with a control plan; then have Finance validate the sustained rate before booking.
The right message upward: "I can give you a defensible number in three months, or a number now that we will have to withdraw."
2. Sensor (error-proofing) versus quarterly retraining: how should the team decide?
Read the worked answer
Judge the controls by what they actually do. Training asks an operator to notice and remember; error-proofing makes the wrong action physically impossible or self-evident. Retraining's effectiveness decays, cannot be verified at the point of use, and does nothing about the new or temporary operator - a real exposure here given the retirement.
Score the failure mode in FMEA terms (severity x occurrence x detection) for a mis-loaded part:
- High severity (scrap, rework, downstream failure or safety) plus low detection means the sensor wins almost regardless of cost.
- Low severity and self-detecting at the next station may justify the cheaper procedural control plus a visual aid.
Also weigh practicality: sensor cost, cycle-time impact, and whether it can be bypassed or defeated - a poka-yoke that operators tape over is worse than none because it creates false confidence.
Decision: select the control that removes or blocks the error rather than detects it, provided cost and cycle time allow. Whatever is chosen, this is countermeasure selection, so it belongs in the Improve phase with a verification run, not in a control plan that assumes it already works. Retraining can supplement but should not be the primary control for a high-severity failure.
3. Write the core of the control plan: what is monitored, on what chart, how often, by whom, and what happens when it signals.
Read the worked answer
Metric and chart: defect rate for the cell. Defects are attribute data, so use a p-chart if the inspected quantity varies by shift or lot (or np/c if the sample is constant). An X-bar/R chart is for variable measurements and does not apply here; do not chart a rate on a variables chart.
Frequency and ownership: per shift or per lot, plotted by the cell lead on the floor - a Yellow Belt task, close to the process so signals are seen within hours, not at month-end.
Reaction plan (written, visible, rehearsed):
- Contain suspect units produced since the last in-control point; quarantine, do not ship.
- Notify the process owner the same shift; record time, chart point and suspected cause on an occurrence log.
- Verify the fixture is the current revision and the sensor has not been bypassed - the most common failure of a poka-yoke.
- Two consecutive out-of-control points, or any point outside limits: escalate, revert to the last known good setup, and open a follow-up action.
Standardization and audit: SOP reissued at the new revision, training records for every operator including the replacement for the retiring one, weekly audit of fixture presence and settings, and a scheduled review of the control chart monthly for three months and then quarterly. Savings are released against this plan, not against the pilot.