What is root cause analysis?

The discipline of asking why something went wrong until the answer stops moving, so that fixing the root removes the symptoms above it.

When should a group do root cause analysis instead of an analyst?

When the causes are spread across the people involved and no single analyst holds them all. Structured Democratic Dialogue collects causes from everyone and lets everyone judge the links at an 80% bar.

How large should a root cause analysis group be?

Twelve to twenty. In simulation panels of fifteen find 4.1 of five known causes and panels of fifty 3.6, because a fixed 80% bar gets harder to reach as a panel grows. Larger rooms are split into groups of fifteen and their maps merged.

Root cause analysis template: seven stages a group can run on paper

Root cause analysis template

For a group of twelve to twenty people who share a problem. Runs on a wall of index cards in a day, or online over a week. Each stage has one rule that keeps it honest. This is the template this site runs; the numbers are the ones measured to work.

0 · The question

One sentence, about causes, not solutions. "What is driving delivery delays across the region?" — not "how do we fix delays". Write it where everyone can see it for the whole session.

1 · Generate 45–90 min

Each person writes causes on cards, alone, one cause per card, before reading anyone else's. Rule: write first, read second — a card written after reading the wall is not independent. Expect 50–100 cards from fifteen people.

2 · Clarify 60–120 min

Each author says what their card means; others may ask what it means. Rule: no judgement — nobody says whether a card is right, important or better than another. Merge two cards only when both authors agree they say the same thing.

3 · Cluster 45 min

Group cards that share a theme. Rule: a cluster forms when two people independently put the same cards together. Name clusters last. The names help people read the whole wall and vote across themes; they are never themselves voted on or structured — cards are.

4 · Select 15 min

Five stickers each, one per card, in silence. Take the top fifteen. Rule: hold five places for cards that got no stickers but that someone can back with a source — a cause nobody voted for can still be the root.

5 · Structure 2–4 hours; the heart of it

For each pair of shortlisted cards, ask: does A significantly influence B? Rule: a link is settled only when 80% of those voting agree; a pair that cannot reach 80% either way is recorded as contested, never as "no". Skip pairs already implied (if A→B and B→C are settled, A→C is). With fifteen cards that is 210 questions before skipping; in practice 60–90.

6 · Map 30 min

Draw the links. Put a card below every card it influences. The cards at the bottom with nothing below them are the root causes. List the contested pairs beside the map.

7 · Actions 60 min

Propose actions. For each action and each root, ask whether the action moves that root, at the same 80%. Rule: an action that moves no root is not a root-cause action, whatever it fixes.

The numbers behind the rules

Twelve to twenty people, because a fixed 80% bar gets harder to reach as a panel grows once the panel is right less than 80% of the time — in simulation, panels of fifteen find 4.1 of five known causes, panels of fifty 3.6. Eighty percent rather than a simple majority, because a simple majority certifies a link when eight of fifteen lean that way, and that is where wrong links come from: settling pairs by simple majority finds 2.7 of five causes, settling them at 80% — with a re-vote, and a split recorded as contested rather than forced — finds 4.3. Five held seats, because a room that ranks purely by what it feels finds none of the real causes by voting and 2.9 of five with the seats.

For more than twenty people, split into groups of fifteen and merge the maps — this site does that for any number of groups. Methods compared · A worked example