Most small plants can tell you what they produced last month and what it cost. Very few can tell you why it was not more.
That gap is not an accounting failure. It is that the numbers explaining production — where time went, why lines stopped, what was made twice — are generated continuously on the floor and captured almost nowhere. They exist in supervisors' heads, in the shape of "it was a bad week", and they leave with the shift.
The result is a plant that knows its outcome precisely and its causes vaguely, which makes improvement a matter of intuition.
The five numbers usually missing
Why the line stopped
Downtime is often recorded as total hours. What matters is the reason, and reasons are what make it actionable.
Material not available, no operator, tool change, breakdown, no order, power — these have entirely different remedies, and only the third and fourth are maintenance problems. A plant recording eight hours of downtime knows it lost eight hours. A plant recording the reasons frequently discovers that the largest category is not the one everyone assumed.
The requirement is a short fixed list of reasons — six to eight — selected at the moment of the stoppage. Free text produces something unanalysable.
How long changeover actually takes
In plants running mixed products, changeover is frequently the single largest recoverable loss, and it is almost never measured.
It is invisible because it is normal. Everyone knows it takes "about an hour", nobody has timed it across a month, and nobody knows the spread between the fastest and slowest instance — which is where the opportunity is, because the fastest instance proves what is achievable.
What was made twice
Rework is the most systematically hidden number in small manufacturing.
Rejection at final inspection gets recorded. A part corrected at the station, or a batch quietly rerun, usually does not — it appears as slightly lower output and slightly higher material consumption, distributed invisibly.
Recording rework separately from rejection is uncomfortable precisely because it exposes a cost that has been absorbed for years.
Where the material went
The gap between material consumed and material in finished goods is scrap, spillage, off-cuts, error and occasionally loss.
Most plants know this figure in aggregate at month end and cannot attribute it to a line, a product or a shift. Aggregate variance is unimprovable; attributed variance is a list of specific things to fix — the same logic as what actually prevents stock variance applied to production rather than stores.
How long an order actually took
Not the promised lead time — the actual elapsed time from order to dispatch, including the waiting.
Plants consistently underestimate this, because they measure processing time and forget queuing. The difference between the two is the number that determines whether delivery promises are met, and it is knowable only if the timestamps exist.
Why these are missing
Not because they are difficult to record. Because recording them costs the operator time, and no operator has been given a reason to spend it.
Three consequences follow:
Anything not near-automatic will be approximated. A stoppage reason selected from six buttons on a terminal at the line gets recorded. A form filled in at shift end gets remembered incorrectly.
Anything used against people stops being true. If downtime data becomes the basis of blame, the reasons recorded will shift toward whichever category is safest. This is the fastest way to destroy a dataset, and it is usually done unintentionally.
Anything never fed back stops being taken seriously. Operators asked to record data that never returns as anything visible conclude, correctly, that it does not matter.
The last point is the one most often overlooked. Data collection is sustained by visible use, and a plant that shows the floor last week's top stoppage reason gets better data than one that sends it to head office.
Starting with one number
The correct scope for a first attempt is one number, on one line, for one month.
Stoppage reasons is usually the right choice: it is the easiest to capture, it has the shortest path to an obvious action, and it produces a result within weeks rather than quarters.
What tends to happen is that the largest category surprises everyone — most often material availability or waiting for a decision rather than machine failure. That surprise is the entire value, and it is unavailable to a plant that records only total downtime.
Sequencing this against the rest of a digitisation programme is covered in what to digitise in the first year, and the systems layer underneath it in production monitoring for small manufacturers.
Making it visible
Data that lives in a report nobody opens has no effect on the plant.
A single screen on the floor showing today against target, and this week's largest loss category, does more than a monthly report — because it reaches the people who can act within the period where action still matters.
The design discipline is severe: one screen, few numbers, legible from a distance, current. The general principle is in operational dashboards that people actually use.
Frequently asked questions
Do we need sensors on machines?
Not to start. Operator-selected stoppage reasons on a terminal capture the most valuable data, and reasons are something a sensor cannot supply anyway.
How do we stop operators recording the easiest option?
Keep the list short, make every option neutral rather than blame-carrying, and show the results back to the floor. Data used against people becomes fiction quickly.
What about very small plants?
The numbers matter more, not less, because there is less slack to absorb losses. The capture can be simpler — a shared terminal rather than one per line.
Is this the same as OEE?
OEE combines availability, performance and quality into one figure. The numbers here are the inputs. Starting with the components is more useful than starting with the composite, because a single index tells you the score without telling you what to do.
How long before it is worth anything?
A month of stoppage reasons on one line is usually enough to identify something worth fixing. This is among the fastest-returning measurements available in a plant.
Where to start
Put a list of six stoppage reasons at one line and record every stop for a month.
Whatever comes out on top is your first improvement project, and it will almost certainly not be the one you would have named in advance.
If you want a plant's data capture designed around what operators can realistically sustain, get in touch.