PFMEA sessions routinely stall on the first failure mode. A facilitator asks for a severity rating, the senior design engineer offers a six, and the room clusters around that number. Forty minutes later, the cross-functional team has debated a single subjective rating without resolving the underlying engineering disagreement. The resulting document reflects social dynamics rather than analytical rigour.
The core problem is unstructured subjectivity. Risk assessment in IATF 16949 and AS9100 environments demands human judgement, but traditional consensus-building corrupts that judgement. Three mechanisms drive the failure: anchoring bias, authority gradients, and ambiguous scale definitions. The first person to speak sets the anchor. The highest-ranking person shifts the consensus. Ambiguous definitions let people argue past each other using different frameworks.
I have audited plants where the PFMEA detection ratings for automated optical inspection were rated at 2 out of 10. The inspection machine itself was excellent. But the operators routinely overrode the system's reject signals when they disagreed, which happened frequently. The effective detection rating was closer to 7. Traditional FMEA facilitation never surfaced this because the production supervisor deferred to the quality manager's optimistic initial assessment.
The Three Failures of Traditional Risk Consensus
Anchoring is the dominant bias in any group estimation. When the most experienced engineer states a number first, every subsequent estimate gravitates toward it. This is documented cognitive behaviour, not a character flaw. The room adjusts upward or downward from the anchor rather than calculating independently. In PFMEA, this means the final severity, occurrence, or detection rating often reflects the first speaker's assumptions rather than the team's combined knowledge.
Authority gradients compound the anchoring problem. A production operator who runs the line daily knows which sensors give false rejects and which control plans are routinely skipped. But when the plant manager or quality director suggests a detection rating of 3, the operator rarely contradicts them in a public setting. The number goes into the FMEA. Six months later, a defect escapes undetected, and the root cause investigation reveals the rating was optimistic by a factor of three.
Definition drift is the third failure. Severity means one thing to a design engineer evaluating end-user safety and another to a process engineer evaluating scrap cost at the station. Occurrence scales without historical data devolve into guessing. Detection ratings conflate having a gauge with actually using it correctly and acting on the result. When team members use different definitions, consensus is meaningless. They agree on a number while disagreeing on the underlying reality.
Traditional FMEA vs Quality Poker Dynamics
Traditional discussion
- First speaker sets an anchor that frames all debate
- Authority gradient suppresses operator dissent
- Circular debate over ambiguous definitions wastes time
- Final score reflects negotiation, not engineering
Quality Poker
- All estimates submitted simultaneously and silently
- Every card carries identical visual weight on reveal
- Outliers drive targeted discussion of hidden knowledge
- Score reflects independent judgement plus structured debate
The Mechanics of Simultaneous Estimation
Quality Poker borrows its structure from Planning Poker in agile software development. The protocol replaces sequential opinion-giving with simultaneous silent voting. Before the session, print the severity, occurrence, and detection scales on large posters with full descriptors, examples, and boundary conditions. Every participant must reference the same definition map throughout the session.
The facilitator reads the failure mode, its effect, and its cause. No opinions, no context-setting, no leading questions. Each team member selects a card from a deck numbered 1 through 10 and places it face-down on the table. No discussion, no body language, no eye contact. On the facilitator's signal, every card flips simultaneously. The distribution is immediately visible to the entire room.
A tight cluster signals agreement. A wide spread signals the need for discussion. A bimodal split signals that team members are likely using different definitions. The facilitator opens discussion not with the consensus but with the outliers. If eight people voted severity 6 and one voted 2, the discussion starts with the person who voted 2. They may have information the majority lacks. They may be using a different framework. Either way, the difference itself is the most valuable diagnostic signal in the room.
The team discusses the outlier reasoning for two to three minutes, then re-votes. The range typically narrows. If it does not, the facilitator captures both estimates and flags the item for further investigation. The cycle takes two to five minutes per rating, compared with the thirty-plus minutes common in traditional debate formats.
Card Design and the Question Mark Protocol
Standard Planning Poker cards use a modified Fibonacci sequence, but for FMEA a linear 1-10 scale maps directly to standard severity, occurrence, and detection tables. Some organisations print custom decks with their specific FMEA scale descriptors on the card backs, so a team member holding a 7 can read the exact definition of what 7 means in their context before placing it down.
Two non-numeric cards add significant value. The coffee cup card signals a needed break, preventing fatigue-driven consensus. The question mark card signals genuine uncertainty. In traditional sessions, people who lack information vote with the majority rather than admitting they do not know. The question mark card gives them permission to express uncertainty without losing face. That single admission prevents more inaccurate estimates than any amount of structured discussion, because it identifies exactly where data is missing.
Physical cards outperform digital tools. Holding a card, looking at it privately, and placing it face-down creates a tactile commitment to an independent judgement. Distributed teams can use shared spreadsheets with simultaneous reveal or dedicated estimation applications, and the core principle of independent judgement survives. But where physical co-location is possible, use physical cards.
What the Outliers Reveal
The outlier discussion is where Quality Poker delivers its highest value. When the cards reveal an outlier, the standard question is: what were you thinking? The answer almost always surfaces one of three things. The outlier voter has specific operational knowledge the rest of the team lacks. They are applying a different definition of the scale. They have witnessed a failure mode the others have not considered.
The minority estimate is not wrong. It is different, and the reason it is different is the most valuable information in the room.
In one automotive supplier case, a cross-functional team had consistently rated their automated optical inspection detection capability at 2. With Quality Poker, the votes scattered: two people at 2, four at 4, three at 6, two at 8, and one at 9. The discussion revealed that the machine itself was excellent, but the operators routinely overrode the reject signals when they disagreed. The effective detection rating was closer to 7. Two years of traditional FMEA sessions had never surfaced this insight.
This pattern repeats across industries. The most junior person in the room frequently holds the most accurate estimate because they are closest to the process. In a traditional session, their voice is the last to be heard, if it is heard at all. Simultaneous reveal gives every participant identical visual weight. The production operator's card sits next to the plant manager's card, face-up, and the difference becomes data rather than hierarchy.
Facilitation Rules and Common Anti-Patterns
The facilitator enforces five rules. No discussion before the vote. Everyone votes, with the question mark card counting as a vote. Discussion starts with outliers, not consensus. Discussion is time-boxed at two to three minutes per outlier. The facilitator does not vote, managing process rather than influencing outcome. These rules are simple to state and difficult to enforce. The first rule, silence before the vote, is the hardest and the most important.
Quality Poker Cycle per FMEA Rating
- 011. Present failure modeFacilitator reads function, failure, effect, and cause. No opinions.
- 022. Silent estimationEach member selects a card and places it face-down. No discussion.
- 033. Simultaneous revealAll cards flipped on signal. Distribution is immediately visible.
- 044. Outlier discussionLowest and highest estimates explain their reasoning first.
- 055. Re-voteTeam estimates again after hearing outlier rationale.
- 066. Converge or flagIf range narrows, record score. If not, flag for investigation.
Four anti-patterns undermine the method. The pre-meeting, where teams discuss failure modes beforehand and arrive with pre-agreed numbers, defeats the purpose. The persistent anchor, where team members signal their estimate through body language before the reveal, requires active facilitation to suppress. The false consensus, where teams treat the method as majority-rules voting, misses the point that the minority perspective is the most valuable. And the tyranny of the average, where facilitators split the difference between divergent estimates, eliminates the diagnostic signal that disagreement provides.
Averaging is the most dangerous anti-pattern. An average of severity 2 and severity 8 is not severity 5. It means the team fundamentally disagrees about the failure mode's impact, and that disagreement must be understood, not averaged away. Record the range, investigate the cause of the split, and reconvene with better data if the range does not narrow after a second vote.
Implementation Beyond PFMEA
PFMEA is the most obvious application, but the method applies to any context where a team converts subjective judgement into a shared number. Risk matrices using severity and likelihood grids benefit from simultaneous estimation. VDA 6.3 process audit scoring, where the audit team must agree on a finding's severity, benefits from preventing the loudest voice from dominating the score. Supplier risk assessments, where cross-functional teams disagree wildly about a supplier's risk profile, become more rigorous when the disagreements surface as data.
Inspector calibration is another application. Have multiple inspectors independently grade the same set of parts using Quality Poker. The disagreements reveal calibration gaps that no training programme can surface as efficiently. If three inspectors rate a part differently, the spread itself defines the training need. Design review risk assessment works the same way: independent estimates of risk for each design element highlight exactly where the team lacks confidence.
The implementation cost is negligible. The method requires index cards, a marker, printed scale definitions, and a facilitator willing to enforce silence. The first session will feel awkward. People are uncomfortable voting without first hearing what the group thinks. By the third session, the team will refuse to return to the old method. The speed gain is immediate, but the quality gain is what matters. The estimates are more defensible, more consistent, and more trusted because every team member contributed independent judgement before the social dynamics took over.
When an IATF 16949 auditor asks how the team arrived at a severity or detection rating, the answer is no longer that the team discussed it. The answer is that twelve independent estimates were submitted simultaneously, the outliers were discussed against defined scale criteria, and the team converged on a number that every member had a role in shaping. That answer holds up under scrutiny because the process is transparent, repeatable, and documented.