In-Depth Analysis of PFMEA Scoring Standards —— Practical Guidelines for S/O/D Scoring and RPN Optimization

1. Why Scoring Is More Important Than Filling Out Forms

Many companies have been implementing PFMEA for years, and the forms have become increasingly thick, yet the effectiveness of risk management has not improved but rather declined. During audits, it is common to see the following scenario in PFMEA files: severity (S) is consistently rated 8 or 9, occurrence (O) hovers between 3 and 4, and detection (D) is uniformly rated 5. The calculated RPN (Risk Priority Number) is either almost entirely over 100, requiring rectification, or all below 50, needing no attention—both extremes indicate that the scoring has not truly reflected the actual risk status of the process.

The core value of PFMEA does not lie in completing a form but in the logical judgment of S, O, and D scores, which helps the team focus on the risks that truly need priority control. Improper use of scoring standards can turn PFMEA into a meaningless game of numbers. The scoring process is essentially about the team reaching a consensus on the "likelihood of failure, the severity of its consequences, and our ability to prevent it." Disagreements in scoring often stem from differences in perception, and resolving these differences is the most valuable part of PFMEA.

2. Core Logic of S/O/D Scoring

Severity (S) measures the impact of a failure mode on the customer, including the end user and internal downstream processes. The AIAG-VDA manual divides S into 10 levels, with a critical threshold above 8—any failure involving safety or regulatory compliance must be rated 9 or 10; those leading to major functional loss are rated 7 to 8; minor functional damage is rated 5 to 6; customer-perceptible but non-functional issues are rated 2 to 4; and no impact is rated 1. A common mistake in practice is the subjective exaggeration of "S," rating all appearance defects 7 or higher, leading to a proliferation of high S scores and a dispersion of resources.

Occurrence (O) is the frequency or probability of a failure mode occurring under current control measures, also rated on a scale of 1 to 10. O scores should be based on historical data—such as PPM data from similar production lines, failure rate records from similar processes, and customer complaint statistics over the past 12 months. O scores without data support often tend to be conservative (either too low or too high), leading to inaccurate RPNs. The key principle is that O is not determined by feeling but by prevention—more effective preventive controls result in a lower O. There is also an often-overlooked interaction between S and O: for failure modes with high S, O scores should be more stringent, as the same frequency of occurrence results in greater risk due to the severity of the consequences.

Detection (D) measures the ability of current detection methods to capture a failure before it reaches the customer. D=1 indicates almost 100% reliable detection (such as an automatic poka-yoke system), while D=10 means there are no detection methods or they are completely ineffective. The biggest misconception in practice is linking D to "inspection frequency"—full inspection of each item does not necessarily mean D is 1. If the detection method itself is unreliable (such as visual inspection), D should be rated 6 or higher even with full inspection. When determining D, ask yourself two questions: Can the detection method identify the failure after it occurs but before it escapes? How reliable is the detection method itself?

3. RPN Optimization: Which to Reduce First and How to Decide

RPN (Risk Priority Number) = S × O × D, which is the most widely known indicator for prioritizing risks. However, relying solely on RPN for prioritization has two major pitfalls. First, the multiplicative effect of RPN can cause a high S × low O × low D combination to be surpassed by a medium S × medium O × medium D combination, even though the former is fundamentally more critical to address. Second, the difficulty and cost of improving the three dimensions vary significantly—reducing S often requires redesign, reducing O can be achieved through poka-yoke and standardization, and reducing D can be done by enhancing detection capabilities.

Therefore, the correct optimization sequence is: first, look at S, then O, and finally D. For any failure mode with S ≥ 9 (safety-related), improvement measures must be formulated even if the RPN is not high; for S between 5 and 7, prioritize reducing O to minimize risk; and when O has already been reduced to a low level, consider improving detection methods to reduce D.

A practical approach: set "target values" for S, O, and D for each row in the PFMEA—require all safety-related failures with S ≥ 9 to have poka-yoke systems in place, making O ≤ 3 and D ≤ 4; control O to 4 or below and D to 6 or below for general functional failures. Rows that meet the target values can be closed, while those that do not should be continuously tracked. It is also recommended to close the loop on projects with high RPNs and the results of improvement measures: re-score after implementing measures, compare the changes in S/O/D before and after improvements, and form a "scoring—improvement—re-scoring" cycle rather than shelving the scores after they are given.

4. Scoring Calibration —— Avoiding "Two Hours of Discussion to Set One Number"

The most common scenario in PFMEA review meetings is the endless debate between engineers and production line supervisors over the O value of a failure mode. The solution is to establish two standard documents—one is a historical failure database that lists the actual occurrence frequency and corresponding O values of similar failures across production lines; the other is an S/O/D scoring reference table that provides specific examples based on the company's product characteristics. Before each PFMEA review, spend 15 minutes calibrating the scoring scale to ensure everyone aligns with the references and reduces disagreements. The scoring reference table should not be a one-time effort but should be regularly updated as new failure modes emerge and the effectiveness of improvements is verified, keeping the scoring scale always aligned with the actual process.

5. Example: Scoring Calibration for a Welding Process

In a welding process at an automotive parts factory, weld spatter caused short-circuit failures. The team's initial scoring: S=8 (major functional loss), O=5 (2 to 3 times per month), D=7 (visual full inspection), RPN=280. After calibration: S remained 8, O was adjusted to 4 based on actual data over the past six months (0.5 times per month), and D was changed to 3 due to the implementation of an automatic visual inspection system. The recalibrated RPN = 8 × 4 × 3 = 96. Scoring calibration eliminated the overestimation of O and D, allowing the team to concentrate improvement resources on truly high-risk items. More importantly, this calibration exposed the team's previous tendency to score based on impressions: the discrepancy between actual data and subjective judgment was much larger than expected.

6. Conclusion

Scoring is not the purpose of PFMEA; precise risk ranking is. Accurate scoring leads to accurate improvements; clear ranking ensures clear resource allocation. It is recommended that every company invest time in establishing its own S/O/D scoring reference system, transforming scoring from a "gut feeling" to a "standardized process." Only then can PFMEA truly become a tool for effective risk management.

Accurate scoring leads to accurate improvements; clear ranking ensures clear resource allocation.

Knowledge code: 8.3.1

Version: v20260806

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