What is loss analysis and how do you use it in production?

This guide goes through how loss analysis works in practice, which questions it should be able to answer, which pitfalls make the analysis misleading, and how factories have used it to reduce downtime by over 70%.

Why is the OEE figure not enough without a loss analysis?

The OEE figure is not enough, as it only shows that you have a fever without indicating what diagnosis is causing it. In the same way, an OEE of 58% shows that 42% of the planned time did not become value-creating production, but it says nothing about the cause: breakdowns, changeovers, material shortages, speed losses, or scrap. Without that knowledge, it is impossible to choose the right action, and improvement work becomes guesswork.

The loss analysis breaks down the gap. It is based on a loss model where every lost hour gets an address: a category, a cause, a location, and a time. The basis is the TPM tradition's six major losses: breakdowns, changeovers, idling and minor stoppages, reduced speed, defects in normal production, and defects during startup. Added to this are micro-stops, the short interruptions of seconds that individually are insignificant but collectively can account for a large part of the performance losses.

An important insight changes how the analysis should be built: data from factories that have worked with OEE for a long time shows that around 80% of downtime causes are not machine-related. They are about materials, planning, waiting for orders, communication, and staffing. This means that automatic machine data is never enough for a complete analysis. The operators' coding of causes is what gives losses their addresses.

What questions should a loss analysis be able to answer?

A good loss analysis should be able to answer concrete prioritization questions in a few minutes, without specialist effort. The most important questions are these:

What are our three largest losses in hours and money? The basic question. The answer guides where improvement resources should be deployed.

Which losses are frequent but short, and which are rare but expensive? The two types require different actions. A hundred micro-stops a day are tackled differently than one breakdown a month.

Why does the same article perform differently on two different lines? The difference between the lines often contains a ready-made solution: the better line's way of working can be standardized to the poorer line's.

Which stoppage causes have increased over the last month? Trends catch problems that are on their way to becoming expensive before they actually have.

Which technical problems recur on the same machine part? Recurring failures on the same component represent a maintenance pattern pointing to the root cause, not symptoms.

How does the outcome differ between shifts? Differences between shifts are rarely about effort, but often about ways of working, and ways of working can be standardized.

At Sibbhultsverken, it was this type of breakdown that made a difference. By systematically analyzing and addressing the largest loss categories, they were able to reduce technical stoppages by 73% and unplanned stoppages by 63% in just 12 months. OEE rose by 19.4% in one cell and 40% in another, and net production could be increased by 15% without extra shifts.

What is required for the loss data to be reliable?

Loss data can be trusted when three things are in place: consistent definitions, simple coding for operators, and visibility of bad data.

Consistent definitions. Planned production time, changeovers, planned maintenance, and ideal cycle time must be handled in the same way over time and between lines. Otherwise, the numbers cannot be compared, and the analysis is built on too shaky grounds. A documented OEE policy, where the team has defined what is measured and how, is the simplest safeguard.

Simple coding. If it takes longer to code a stoppage than the stoppage itself lasts, the coding will not happen. The interface on the shop floor should allow coding in seconds, with clear cause codes that the operators themselves have helped design. Orkla Nidar eliminated the "Other" category from their reporting when operators were given tools that made it easy to code the correct cause directly. A growing "Other" category is otherwise the surest sign that the analysis is losing its basis.

Visible data quality. The system itself should show where the data is lacking: uncoded stoppages, activity outside the schedule, unrealistic speed patterns. Then the shortcomings can be resolved ongoingly instead of being discovered only after someone has already made a decision on incorrect information.

How do you go from analysis to action?

From analysis to action, you proceed via prioritization, root cause, and follow-up, in a cycle that is repeated. The analysis in itself does not improve anything. The value arises when it guides action.

A proven workflow looks like this:

Prioritize by value. Choose the two or three losses that cost the most, not the ten that are visible. Doing a few things with proper resources beats doing many without.

Seek the root cause. Feel free to start with a fishbone diagram (called Ishikawa within Lean) to brainstorm and identify causes. All suggestions are welcome. Then the probable causes are prioritized, and the most likely ones are investigated using the "five whys" method to find underlying causes. This method generates ideas and focuses on the most likely ones to find the root cause. 

For complex problems, a structured DMAIC project may be needed, as at Bostik where the analysis of changeover data led to 70% shorter changeover times.

Address with owner and date. An action without an owner remains on the list.

Follow up in the statistics. The addressed loss should disappear or decrease clearly in the data. If it does not, the problem was not solved fundamentally and a new round is needed. The follow-up takes place in the same system as the analysis, which makes it almost free.

Barilla Wasa in Filipstad shows the big picture. Through systematic work with losses, the utilization rate rose from 75% to 94%, downtime decreased by 22%, product waste decreased by 15%, and net production increased by 15%. At the same time, CO2 consumption fell by 28%, because lost machine time also means lost energy. The loss analysis supported the entire chain, from prioritization to proven effect.

What are the most common pitfalls in loss analysis?

The most common pitfalls are analyzing too rarely, getting stuck in the overall figure, comparing the wrong things, and forgetting the small losses.

Analysis once a quarter. Then the problem has time to grow expensive before it is discovered. The loss analysis must be integrated into the weekly rhythm, with the largest losses as a standing item in the improvement meeting.

Staring at the overall OEE. The total figure moves slowly and hides underlying opposing movements when some causes decrease while others increase. A line can improve changeovers while micro-stops increase and the overall total stands still. Analyze the parts, not just the sum.

Comparing lines and factories directly. Different definitions and different conditions make direct comparisons misleading. Compare each unit with its own history, and use differences between units as a source of questions, not as a grade.

Ignoring micro-stops. They are barely visible individually and are not captured manually, but together they can constitute one of the largest loss categories. Automatic measurement is the only way to make them visible.

How does Good Solutions work with loss analysis?

The loss analysis is the core of the platform from Good Solutions, built on its own loss model that gives every lost hour an address. The reports break down losses by stoppage cause, cause group, line, article, shift, and trend over time. The timeline shows the 24-hour day visually, with downtime, volume, and OEE in the same view, and stoppages can be coded directly in the view. The operator tool makes coding fast enough to be done in the moment, which is the prerequisite for the analysis to cover the roughly 80% of stoppage causes that machine data cannot explain.

Around the analysis are the ways of working. The operational deployment ensures definitions and data quality right from the start. The OEE policy workshop documents what is measured and how it is measured, so that the numbers remains comparable. The training courses give operators and superusers the common understanding that keeps coding consistent. And because the platform is built to drive continuous improvements, not just measure, the analysis leads on to prioritization, action, and follow-up in the same tool.

The results for customers come precisely this way. Sibbhultsverken reduced technical stoppages by 73%. Kopparbergs reduced quality deviations by 68%. In all cases, the work began in the loss analysis and ended in proven effect.

FAQ

What is the difference between OEE and loss analysis?
OEE is the figure, the loss analysis is the explanation. OEE shows what proportion of the planned time became value-creating production. The loss analysis breaks down the rest by cause, location, and time, so that improvement work knows where to target.

What loss categories should we use?
Start with the six major losses, with micro-stops as a category of their own, and adapt the cause codes to your business together with the operators. The codes should be few enough to be fast to choose and clear enough not to be confused. A growing "Other" category is a signal that the structure needs to be reviewed.

How often should we do a loss analysis?
Ongoingly. The largest losses should be a standing item in the weekly improvement meeting, and deviations should be visible in the daily management. Quarterly analyses are too sparse, as problems have time to become expensive between the occasions.

Can we do a loss analysis in Excel?
On a small scale, as a start, yes. But manual collection misses the micro-stops, coding becomes incomplete, and the analysis work takes time that should be spent on actions. An OEE system collects data automatically, keeps it consistent, and does breakdowns in minutes.

How do we know the analysis is leading us right?
Through follow-up. When a prioritized loss is addressed, it should decrease clearly in the statistics. If it does, the analysis is correct. Redo the analysis if the action has targeted a symptom instead of the root cause. Then the analysis needs to be deepened, preferably with the operators' knowledge as a complement to the data.