From measurement to root cause: measure then explain

Two different questions: where, and why

Lastingly improving an OEE means answering two distinct questions, and not confusing them. The first is ‘where and when does performance drop?’: that’s the question of measurement. The second is ‘why does it drop?’: that’s the question of root cause. These are two complementary steps, and neglecting either leaves the problem open. (OEE, Overall Equipment Effectiveness, is the English name for what French manufacturers call TRS.)

Confusing the two is a frequent source of failure. Some believe measuring is enough to understand; others seek the cause without having a reliable measurement to hand. Yet measurement and explanation don’t substitute for one another: they follow on. Measurement makes the losses visible, cause analysis explains them. One prepares the ground for the other, and it’s their articulation that produces a lasting gain. Measure without explaining and you log the same drop forever; explain without measuring and you argue about a problem nobody has actually pinned down.

Measure first: make the losses visible

The first step is always measurement. As long as a loss isn’t measured, it doesn’t exist for analysis: you can’t seek the cause of what you can’t see. Measuring real OEE, continuously, machine by machine, to the second, is therefore the indispensable starting point of any root-cause approach. It’s what turns a vague intuition into a precise, located fact.

This measurement must be complete and objective to be useful. A partial manual reading, which misses micro-stops and under-pace, provides a gap-ridden base on which cause analysis will stumble. An automatic measurement, which captures every loss and timestamps it, offers instead a rich material, where each loss is placed in time and tied to a context. It’s that quality of measurement that conditions everything that follows. Build your analysis on an incomplete reading, and the conclusions will be as incomplete as the data.

Measure with a standalone layer, no heavy project

The advantage of a standalone measurement layer is that it provides this foundation without imposing a heavy project. Real OEE is displayed continuously, machine by machine, with no MES project, a sensor fitted in under an hour and usable data within 48 hours. You thus quickly have the material needed for analysis, on old machines as much as recent ones.

This speed matters, because it prevents the measurement phase from becoming a project in itself that endlessly delays the search for cause. You measure fast, completely, and you can start analysing almost immediately. Measurement isn’t an interminable prerequisite: it’s a foundation you lay quickly in order to build, on top of it, the understanding of causes. The sooner the data flows, the sooner the team can move past arguing about what really happened and start asking why it happened.

Explain next: from context to cause

Once the losses are measured and located, the explanation can begin. It consists of tying each loss to its production context: does this stoppage recur with a particular material batch, in a given time window, with a team, or for a combination of settings? It’s by cross-referencing the measured loss with these factors that you trace back from the symptom to the real cause. The pattern that the eye misses on a single shift often becomes obvious once weeks of timestamped events are laid side by side.

This contextual analysis is what separates a lasting fix from a patch. Acting on a stoppage without understanding its origin exposes you to seeing it return; identifying that it’s the conjunction of a batch and a setting that causes it means being able to eliminate it at the source. Measurement provides the facts; context analysis gives them meaning and points to the right lever for action. Without that step, you are left treating the visible event rather than the conditions that keep producing it.

When advanced analysis takes over

For simple causes, human analysis of the data is often enough: a trained eye spots the correlation between a loss and a factor. But some causes are multi-parameter and too subtle to untangle by hand. That’s where advanced analysis, assisted by machine learning, takes over, exploring large volumes of data to surface combinations a human wouldn’t see.

It’s important to place this contribution correctly: advanced analysis doesn’t replace measurement, it extends it. It only makes sense on reliable, complete data, and for problems that simple analysis hasn’t solved. So you don’t start with ML: you start by measuring, you treat the accessible causes, and you reserve advanced analysis for the complex cases that justify it, once the measurement foundation is solidly in place. Reaching for advanced models before the basic measurement is trustworthy only dresses up an unreliable input in a sophisticated output.

The order of the steps never reverses

One principle structures the whole approach: you can only seek the cause of losses you’ve first made visible. That order never reverses. Trying to analyse causes without reliable measurement is building on sand; wanting to deploy predictive tools without first having measured real performance is putting the cart before the horse. Measurement always precedes explanation.

Respecting this order spares you many disappointments. Many ambitious analysis or predictive projects fail because they skipped the measurement step, or rested on partial data. By first laying a solid measurement foundation, you make sure everything built on top of it, from simple analysis to ML, rests on a reliable reality. The most advanced model is only ever as trustworthy as the measurement underneath it. Hutchinson improved its OEE from 42% to 75% with the same headcount and machines, sensor installed in under an hour.

A gradual, coherent approach

The right trajectory is therefore gradual and coherent: measure real OEE to see the losses, act on the accessible causes through context analysis, then mobilise advanced analysis for the residual complex cases. Each step rests on the previous one, and none skips measurement. It’s that coherence that produces lasting gains rather than passing fads. Skip a rung and the whole ladder wobbles; respect the order and each level reinforces the one below it.

This progression also has the advantage of being accessible. You don’t need to wait until you have a sophisticated analytics platform to start: you begin with measurement, which creates value immediately, and you ramp up in sophistication at the pace of real needs. More than 450 plants across 30+ countries already monitor their OEE to the second with TeepTrak. Root cause isn’t a starting point, it’s the outcome of an approach that begins, always, by measuring.

Key takeaways

Measuring says where and when OEE drops; the search for root cause says why. The two steps are complementary and follow on in an order that never reverses: you only seek the cause of losses first made visible. You measure real OEE with no heavy project, you tie losses to their context to explain them, and you reserve advanced ML-assisted analysis for the complex cases.

Measurement is the foundation of any lasting approach, and it’s by laying it first – quickly and completely – that you make sure everything built on top rests on a reliable reality rather than on guesswork. Hutchinson improved its OEE from 42% to 75% with the same headcount and machines, sensor installed in under an hour.

FAQ

Is measuring OEE enough to improve it lastingly?
No. Measuring says where and when performance drops, but you then have to explain why, by searching for the root cause. Both steps are needed: without explanation, you treat symptoms and the problem comes back.

Where should you start?
With real-time measurement, complete and objective: you can only seek the cause of losses you’ve first made visible. Measurement is the indispensable foundation of any root-cause analysis.

How do you move from measurement to cause?
By tying each measured loss to its production context: material batch, time window, team, combination of settings. This cross-referencing traces back from the symptom to the real cause and points to the right lever for action.

When should you turn to machine learning?
For complex multi-parameter causes that simple analysis hasn’t solved, once reliable measurement is in place. ML extends measurement, it doesn’t replace it, and only makes sense on complete data.

Can you start directly with predictive?
No. The order never reverses: measurement precedes explanation, which precedes predictive. Skipping the measurement step is building on sand. You start with measurement, which creates immediate value, and ramp up in sophistication afterwards.