Decisions
half a day

Failure modes and effects analysis

Score each failure mode on severity, likelihood and detectability, then work the highest products first.

Time cost
half a day
Output
A ranked table with a risk priority number per failure mode.
Steps
6

Use when

  • A process has many possible failure points and you must rank rather than list them.
  • Failures differ in how easily they would be noticed, not only in how bad they are.
  • The same analysis must be handed to someone else and re-run later.

Do not use when

  • The system is novel and the failure modes are unknown. FMEA ranks a list; it does not generate one.
  • There are fewer than about six failure modes, where ranking adds nothing over reading them.

Inputs required

  • A decomposed process or plan
  • Failure modes per step
  • Three 1–10 scales agreed in advance

Procedure

  1. 01

    Break the plan into steps

    Each step must be something that can fail independently. Granularity decides the quality of everything downstream.

  2. 02

    List failure modes per step

    How could this step fail? Not why yet — what the failure looks like from outside.

  3. 03

    Score severity

    1 to 10, on the consequence if it happens. Anchor the ends explicitly before scoring anything, or the scale drifts between steps.

  4. 04

    Score likelihood and detectability

    Likelihood of occurring, and — this is the one people skip — how likely you are to catch it before it does damage. High detectability scores low, since being caught is good.

  5. 05

    Compute and rank

    Multiply the three. The risk priority number is the ranking. Work from the top.

  6. 06

    Treat detectability as an action

    A high-severity failure you cannot detect is a different problem from one you can. Adding detection is often cheaper than reducing likelihood, and it changes the score.

Characteristic failure mode

Treating the risk priority number as a measurement. It is an ordinal score multiplied three times, so a 336 is not meaningfully worse than a 320. Use it to sort, not to threshold.

Worked example

A clinic reviews its medication-dispensing process.

  1. 01Eleven steps, twenty-three failure modes.
  2. 02One mode scores severity 9, likelihood 3, detectability 8 — nobody would notice before harm.
  3. 03Its product, 216, ranks second overall despite a low likelihood.

Result

The top action is adding a check, not reducing the failure rate. Detection was the cheap lever and it only surfaced because detectability was scored separately.

Where to go next