An action is only worth listing if it changes a score. In a process FMEA, the scores you can realistically move are occurrence and detection. Occurrence falls when you remove or control the cause: a fixture that sets insertion depth, a locked parameter, a better tool-change rule. Detection falls when you add or improve a control that catches the mode: a 100 percent leak test, a go/no-go gauge, a vision check.
Severity almost never changes from a process action, because it describes what the effect does to the user. Only a product or process design change that alters the effect itself moves severity, for example a connector design that cannot be assembled short. If a team claims a training session reduced severity, the row needs another look.
After the action is implemented and shown to work, re-score the row with the same scales and recalculate RPN = S × O × D. Keep the before and after side by side. Thresholds such as "act on anything above 100" are a useful starting filter, but they are arbitrary. Never nudge a score down to slip under a line, and never assume a row at 95 is safe because the cutoff is 100.
Standard Procedure
- 1For each priority row, decide whether the action targets the cause (occurrence) or the control (detection), or both.
- 2Implement the action and collect enough data to show it works.
- 3Re-score occurrence and detection honestly; leave severity unchanged unless the design changed.
- 4Recalculate the RPN and record before and after values with the evidence.
- 5Review rows just below any threshold on their own merits rather than treating the threshold as a pass line.
Lesson Summary
- ✓Lower occurrence by fixing the cause; lower detection by adding a reliable control.
- ✓Re-score after proving the action, and never adjust scores just to clear a threshold.