Finding Errors in Advance: FMEA Simplified for Small Companies

Why errors usually only show up once they have happened

In many companies, dealing with errors follows a fixed pattern. A customer complains, the team looks for the cause, a rule or checklist is added. That has a price: every improvement presupposes damage. For rare but expensive errors, that is a bad trade.

Failure mode and effects analysis, FMEA for short, aims to reverse this order. This leads to the guiding question of this article: what does FMEA offer a small company, and how can it be simplified so that it is useful without turning into a body of forms? To answer it, we first clarify the terms and the classic rating, then examine the criticism of the risk priority number and the research findings, weigh two perspectives, and finally describe a one-page FMEA.

A customer complaint is an FMEA the customer has carried out for you, at the most expensive possible moment.

What is an FMEA? Terms and origin

FMEA, failure mode and effects analysis, is a method by which a team collects, rates, and reduces through measures the possible errors of a product or process in advance. According to the review by Aleksić, Tadić, Komatina, and colleagues (2025), it was introduced in aviation in the 1960s, among other things for the flight control system of naval aircraft at the manufacturer Grumman. In the automotive industry it is now mandatory under the industry standard IATF 16949 and is prepared there by a dedicated team.

Three terms carry the method. The failure mode describes what can go wrong, for example wrong delivery date in the quote. The effect describes what follows for the customer or the company, for example contractual penalty, annoyed customer. The cause describes why the error arises, for example the date is entered without checking with production.

Traditionally, each error is rated on three characteristics, usually on a scale of 1 to 10. Severity indicates how serious the effect is. Occurrence indicates how likely the cause is. Detection indicates how likely the error is to go unnoticed before it reaches the customer. A high detection value is therefore bad. The product of the three values is called the risk priority number, RPN for short, and lies between 1 and 1,000.

The thinking model: rating in three dimensions and the weakness of multiplication

The idea behind FMEA is convincing. A risk is not only a matter of probability, but also of severity and of whether the error is noticed in time. The third characteristic is the method's real contribution. It connects with the Swiss cheese model described in the article on errors as process signals: every check in the process is a slice, and detection asks how many holes it has.

Multiplying the three values, however, has a well-known weakness. Very different risks can produce the same number. An error with severity 10, occurrence 2, and detection 5 gives an RPN of 100. An error with severity 2, occurrence 10, and detection 5 also gives 100. The first can cost a customer, the second costs a few minutes of rework. According to the RPN, they are equally urgent.

Industry has drawn conclusions from this. Jeluš, Breznická, Kohutiar, and colleagues (2026) compare the earlier American standard from 2008 with the joint standard of the American and German automotive industries from 2019. The latter replaces the RPN with an action priority with the levels high, medium, and low. The level is read from a table in which severity counts first. This way, a very serious effect can no longer be calculated down by favorable values for occurrence or detection. In addition, there is a mandatory seven-step procedure.

What research shows about FMEA

Much of the research on FMEA deals with its method of calculation. Aleksić and colleagues (2025) analyzed 68 articles from 2015 to 2024 that combine FMEA with multi-criteria decision methods. The authors name as the basic problem that the three characteristics are not equally important and that ratings often arise in words such as rare or serious that cannot be cleanly translated into numbers. For a small company, this is above all a warning not to trust the number with too much precision.

How important the detection characteristic is can be seen in a case from medical training. Roberts, Remy, Lavine, and colleagues (2026) analyzed 22 reported equipment failures in a simulation center over 24 months. The complete failure of a training manikin had the highest severity but reached an RPN of only 140, because it is noticed immediately. A failure of the video recording during an exam reached 378. The exam seemingly continues as normal, and the error only becomes apparent when someone looks for the recording. Going by severity alone, the center would have fixed the wrong error first.

A randomized study from a hospital suggests that a carefully applied FMEA can improve results. Fan, Guo, Ma, and colleagues (2026) compared usual care with a care process redesigned on the basis of an FMEA in 280 patients after a cardiac catheter examination. Bleeding within 24 hours occurred in 4.3 percent of the FMEA group and 13.6 percent of the comparison group. The study comes from a single center, and the effect cannot be separated from the simultaneously improved collaboration.

Numerical model or conversation structure? Two perspectives

There are two readings of where the value of an FMEA lies. The first sees it as a numerical model: risks are measured, ranked, and steered with thresholds. That is how it works in the automotive industry and in many food companies. Samimi, Aslani, Shariatifar, and colleagues (2026), for example, describe a croissant production plant in Iran in which the RPN of the failure modes ranged from 24 to 240 and all risks above 64 were included in the hygiene plan.

The second reading sees it as a conversation structure: a team goes through a process step by step and asks the same four questions at each step. The benefit arises because knowledge from different heads comes together and assumptions are voiced. The number is then only an aid for setting the order of measures.

For small companies I consider the second reading the more useful one. The criticism of the RPN shows that the precision of the number is often only apparent. The case from the simulation center, on the other hand, shows that the question about detection yields real insights. A simplified FMEA therefore keeps the four questions and the three characteristics, but replaces the multiplication with a simple rule modeled on the action priority.

The one-page FMEA: how to do it in two hours

The one-page FMEA is a simplified version for a single process. It suits processes in which an error is expensive and rare, such as costing a large contract or the handoff to a subcontractor.

1. Set process and team (10 minutes)

   One process with five to eight steps. Three to five people who know it from different angles, including someone who performs it.

2. Ask for each step what can go wrong (40 minutes)

   Collect failure mode, effect, and cause for each step. No rating in this phase, only collecting.

3. Rate severity (15 minutes)

   Each failure mode gets high, medium, or low. High means customer, money, or safety seriously affected.

4. Check detection (20 minutes)

   For each failure mode of high severity, ask: at which point would we notice the error before it reaches the customer? If there is none, that is a finding.

5. Set measures (25 minutes)

   Errors with high severity and poor detection take priority. For each: one measure, one responsible person, one date.

6. Review after three months (10 minutes)

   Have the measures been implemented? Did one of the errors occur anyway? The page is adjusted.

The measures often end up in a checklist at a fixed checkpoint, as described in the article on checklists in the company, or in a revised template. For individual large projects, the pre-mortem from the article on cognitive biases of entrepreneurs is a useful complement: you imagine the project has failed and write down the reasons.

Where FMEA reaches its limits

FMEA only finds errors that someone can imagine. Completely new processes or rare chains of several circumstances often remain undetected. Nor is it a substitute for analyzing errors that actually occurred. A complaints register, as provided for in the article on quality management without certification bureaucracy, shows which assumptions of the FMEA were wrong. The two complement each other: FMEA looks ahead, the register looks back. How the risks of the whole company are kept in a maintained list is described in the article on risk management for small companies.

The studies cited come mainly from hospitals, industry, and food production. The transfer to office work and services is based on the logic of the method, not on measured effects.

Conclusion: what does FMEA offer a small company?

The question at the start was
FMEA brings a small company and how it can be simplified. Its value lies in four questions asked for each step of a process: what can go wrong, what follows from it, why does it happen, and would we notice it in time? The last question in particular brings insights that are missing without the method, as the case of the unnoticed recording failure shows.

As shown, the classic risk priority number, by contrast, is weaker than its precision suggests. Industry itself has replaced it with a ranking in which severity counts first. For the small company this means a one-page FMEA: one process, one team, two hours, three to five measures.

First steps for this week

1. Choose a process in which an error would be expensive and that occurs rarely enough that you do not learn from routine.
2. Invite three to five people to a two-hour meeting, including someone who performs the process.
3. Collect failure mode, effect, and cause for each step, without rating.
4. Rate severity as high, medium, or low and check detection for all high-severity errors.
5. For errors with high severity and poor detection, set one measure each with a name and date.