Sampling During Production: When Should You Request Mid-Production Checks?

The case for looking at goods while they are still being made is easy to state and harder to apply. Everything said in favour of it rests on one claim: that correcting a fault early costs less than correcting it late. Sometimes that is obviously true. Sometimes it is not true at all, and the visit simply spends money to learn what everybody already knew.

Working out which case applies is a short exercise, and it comes down to what kind of faults your programme actually produces. Programmes that skip this step tend to discover later that they were paying for reassurance rather than for information, and that reassurance is the dearer of the two.

It also comes down to whether anybody is prepared to act. A visit whose findings nobody would change plans for is not an inspection; it is a report written for a file. Ask that question before booking rather than after reading.

What Is a Mid-Production Check?

A visit made while some but not all of the quantity exists, examining work in progress, part-finished units and the operation producing them. It looks upstream as well as at units — settings, technique, material handling — because those are what determine the units not yet made.

Its output is not a verdict. A mid-production visit produces observations and corrections; the release decision belongs to the final inspection, which reads a packed lot against the plan. Confusing those two roles is why some buyers expect a line visit to have stopped a shipment and some suppliers fear it did.

It also differs in what it can honestly claim. Drawing units from part of a run tells you about that part; it says nothing statistical about the lot as a whole, and it should never be presented as if it did. Buyers who understand this get considerably more value from the visit than those who do not.

The other thing worth knowing is what it cannot see at all. There are no cartons yet, no packing quantities, no shipping marks, and no assurance that anything observed during making survives the packing operation. Everything introduced downstream is still unknown at this point.

When Should the Request Actually Be Made?

Before production starts, in the order, alongside everything else about inspection. Requested later it becomes a favour with a price attached, and made too late it cannot take place at all — because the run has moved past the point where a correction would still help anybody.

Then confirm it against progress. The useful arrangement is a trigger rather than a date: once production confirms that the stated share exists, the booking is made. This survives the schedule shifting, which it always does.

What Is the Minimum Useful Completion Level?

Enough that the process is running rather than being set up, and enough that the operations you care about have produced something to examine. In practice teams commonly work to somewhere between 10% and 30% of the quantity existing in finished form, and the number should be written down rather than felt.

Later than that and the correction has less run left to apply it to; earlier and you may be looking at setup output rather than steady-state work. Where the concern is specifically the first units off a machine, say so — a look at first-output units is a different and much earlier check.

Get the trigger wrong at the low end and you examine setup output rather than steady-state work; the first units off any line are atypical in ways that produce findings nobody should act on. Get it wrong at the high end and there is very little of the run left for any correction to apply to.

What Should It Cover?

The faults that would be expensive to find late, and whatever the last report made you worried about. This is a targeted visit rather than a shortened version of the final one, and treating it as the latter is a common way to waste it.

Practically, that means: the operations happening that day, units taken from the line rather than from finished stock, and specific checkpoints named in advance. A general look round the floor produces general observations, and general observations change nothing.

It should also mean looking at what the line will do next rather than only what it has done. Components staged for the next operation, a material lot about to be opened, an operator newly assigned to a station — all three are predictable about tomorrow and invisible to any sample drawn today.

Give the visit one or two questions rather than a remit. "Is the webbing anchor reinforcement present and placed correctly?" produces something usable. "Check quality" produces a summary nobody can act on, which is the honest description of a great many line reports.

How Is the Sample Drawn at Mid-Production?

Deliberately rather than randomly. During a line visit the useful thing is to look at units coming off specific machines, specific shifts and specific stations — because variation between those is precisely what the final random sample cannot see. State which, or accept that the visit will default to whatever happens to be nearest the door.

Record against what the drawing was made. "Twenty units from the closing station across both shifts" can be repeated and reasoned about; "twenty units" cannot. Where a plan from the standard is also being applied, give both so the two numbers are not confused later.

What Happens to the Findings?

They become named points on the final checklist. This is the step that turns the visit from information into control: whatever was seen now gets looked for again at the end, specifically, rather than being trusted to have been fixed.

They should also go back to whoever owns the process, in writing, the same day. A correction mentioned conversationally and confirmed three days later has usually been applied to three days of work nobody examined, and the gap between those two things is where the whole exercise is won or lost.

What Does It Cost Compared With Finding Out Later?

The comparison is not between two invoices. It is between a visit and whatever a late discovery costs: sorting finished units, reworking through the packing, re-inspecting, and a booking that may already be at risk. Against a 35-50 day run, the expensive outcome is almost always the delayed one.

There is also the reverse case, worth stating honestly. Where correcting early saves little — because the fault is cosmetic, because the run is nearly finished, or because the fix applies equally easily at the end — the visit buys reassurance rather than anything else. That is not worthless, but it should be bought knowingly rather than by reflex.

The honest arithmetic includes scheduling. A line visit requires somebody to be somewhere at the moment production reaches a stage it may not reach on schedule, so each booking carries a real chance of being wasted outright. Building in a short confirmation step — written proof the stage has been reached — removes most of that exposure for very little effort.

What Should Be Recorded?

What share of the quantity existed, which machines and shifts were looked at and how many units, what specific observations were made, what correction was agreed, who agreed it and by when. Then the same point re-examined at the final inspection, with the result recorded against the original finding.

How Do You Know It Worked?

Compare the fault picture across lots. A line visit that is doing its job produces fewer structural findings at final inspection, and the ones that remain should be diminishing rather than merely different. Where the final counts look unchanged after two lots of line visits, the visit is not reaching the cause.

Watch for a specific failure mode: the visit keeps finding the same thing and nothing changes. That is no longer an inspection problem. It means the correction is not being applied, cannot be applied, or is being applied to the wrong operation — and continuing to inspect it will not fix any of those.

What Does the Visit Do to the Sample Sizes Later?

Nothing automatic, and that surprises people. Finding nothing mid-run does not entitle anybody to reduce the final sample, and finding something does not entitle anybody to increase it. The plan named in the order governs the final visit unless both sides agree to change it in writing.

There is a legitimate mechanism for a programme on a sustained clean run: a reduced state under the continuing-series rules of ISO 2859-1, which draws roughly 40% of the normal sample. Its trigger is history and output stability rather than the result of one line visit, and it returns to normal the moment a lot is rejected. Line findings inform that judgement; they do not replace it.

Should the Visit Carry Power to Stop Production?

It is worth thinking about before somebody wants it. Turning off a line is a significant commercial act, usually requiring somebody senior on the buyer side to authorise it explicitly — and it should be reserved for findings that would make the remainder worthless rather than merely imperfect.

Most programmes settle on a narrower version: the inspector escalates a specific named fault immediately, and the floor continues while whoever owns the decision responds that day. It achieves nearly all of the benefit without handing any single visitor the authority to halt somebody else's operation.

Who Pays for It?

Whoever required it, following the same convention as the final visit, with re-examination usually following responsibility for a failure. Settle it in the order rather than discovering the answer in an invoice.

Where the visit is being added because the previous lot failed, there is a reasonable argument that the additional cost belongs with whoever caused the failure. Many programmes adopt exactly that, and it has the useful side effect of concentrating attention on getting it right.

How Do Programmes Get This Wrong?

Four mistakes account for most wasted visits. Requesting one as a general safety measure rather than for a named concern. Sending it without named checkpoints so it becomes a tour. Failing to require the findings to reappear on the final checklist. And continuing to run it after several lots have shown that nothing comes of it.

The last one matters most. A visit nobody would cancel if it produced nothing is not quality control; it is a habit with an invoice attached. Review every one after each lot and require a reason to run the next.

The opposite error is rarer and worse: dropping the visit after it found something, because somebody concluded the problem was fixed. The evidence for that conclusion is the next lot's findings, not anybody's confidence.

What Do You Ask the Inspector to Do Differently Here?

Two things, both departing from how final inspections are usually run. Look at the operation rather than only at the output — watch hands, settings and sequence, because those determine tomorrow's units rather than today's. And speak to whoever is supervising while you are there, because a correction mentioned in a report three days later arrives too late to be relevant.

Ask for context photographs rather than only evidence photographs. Pictures of the station, the setting or the material handling are what allow somebody else to reason about cause, whereas a picture of the failing unit only shows the effect.

How Do You Read a Line Report?

By what it says about tomorrow rather than what it says about today. A line report is unusual among inspection documents in being forward-looking: the units it described are not the point, the ones not yet made are.

When Should You Not Request One?

When the answer would not change what anybody does. If the programme would ship the lot regardless of what the visit says, or if the only plausible finding would be something the final inspection catches anyway, the visit is spending authority rather than gathering evidence.

Also consider the run length honestly. A small lot may complete in a few days, and there may be no window left by the time a booking could be confirmed. In that situation, ask for first-output units to be examined and approved at the start instead — which achieves most of the purpose and none of the scheduling difficulty.

Frequently Asked Questions

People Also Ask