Test a Work-Method Change Without Trading Quality for Speed

A faster trial is only the start of the decision

A material tray is moved closer to a sewing operator. The next bundle finishes sooner. The supervisor sees a useful idea; the operator sees less reaching. Both may be right. Yet the result could also reflect an easier bundle, a fresh needle or extra help from the person feeding the station.

Work method improvement in a garment factory needs more than a convincing demonstration. The team needs to know what changed, how the comparison was made and whether the gain survives normal work. Quality and worker health belong in that decision from the start.

This guide follows one invented tray-layout trial. All counts, times and outcomes are teaching figures. They show how to build and question a comparison, rather than promise a particular productivity gain.

The proposed change affects material presentation. It does not change the seam, machine speed, stitch setting or quality requirement. A different scope would need a different trial plan.

Put the question into words before taking times

“Improve productivity” leaves too much open. A team could achieve a higher count by adding a helper, passing defects forward or moving work outside the measured period. Those outcomes answer different questions.

For this trial, the question is: does layout B increase first-pass accepted output at the named operation, compared with layout A, under the stated conditions? The decision must also consider support labour, quality and worker feedback.

Layout A is the current approved arrangement. Layout B places the incoming pieces in a reviewed tray position. The exact tray location is chosen for the workstation and people involved. This article does not prescribe a universal height or reach distance.

The NIST guidance on experimental objectives begins with a team discussion of the goals. It distinguishes a narrow initial comparison from broader confirmation. That distinction matters here: an early trial may justify more study without justifying rollout.

Before observations begin, the team writes down what would warrant the next stage. A small output gain might still be useful if the method reduces strain. A larger gain may be unacceptable if it creates a new hazard or defeats the quality check.

There is no universal minimum gain for all factories. The decision should reflect the problem, the effort of changing the method and the evidence needed for the risk involved.

Review the change with the people who will use it

A tray can shorten a reach while creating a different problem. It may block knee space, hide a control or make the feeder twist while loading pieces. Looking only at the sewing operator misses part of the work.

OSHA's sewing-station guidance discusses reaching, bending and material placement. It offers possible ways to reduce awkward movements. These are options for assessment, not proof that one tray suits every worker.

A competent person checks the proposed arrangement before anyone uses it in a timed run. Machine guards, safe access and required controls remain in place. Any equipment change follows the site's approved process and the manufacturer's instructions.

Workers should help shape the trial. Ask about the point where a bundle becomes hard to hold, how the last pieces are reached and what happens when the tray is refilled. A demonstration with a full, neatly arranged tray may miss the difficult part.

The HSE guide to worker consultation describes a two-way process in which workers can influence health and safety decisions. Its legal statements concern its own jurisdiction. The practical approach supports a better trial wherever the factory operates, alongside local duties.

Explain the purpose and records to participants. This is a test of a method, not a contest to identify the fastest person. A worker must be able to raise discomfort or a concern without pressure to finish the run for the sake of the data.

Choose comparable runs instead of a convenient before and after

Suppose four operators take part. Each already knows the operation and has practised both layouts safely. The trial covers two selected styles: a light style and a heavier style. The team observes both methods on two days.

That produces:

4 operators × 2 styles × 2 days × 2 methods = 32 timed runs.

Each run lasts ten minutes. The study therefore contains 320 operator-minutes of timed observation, with 160 minutes under each method. Practice, setup and trial administration are recorded separately. They are not claimed to take no time merely because they fall outside the output comparison.

Within one operator-style-day combination, A and B form a pair. This keeps that operator, style and day together when comparing methods. There are sixteen such pairs.

NIST's explanation of blocking describes grouping runs around factors that could otherwise obscure the comparison. In this example, comparing within pairs is more useful than giving A to the most experienced operators and B to new staff.

The fabric lot, machine condition and quality criteria are kept alike within each pair. The record also notes size, bundle identity and any relevant handling difference. “Same style” is not enough if one batch has a different finish or behaves differently under the machine.

The two styles are deliberately selected. They are not a random sample of every product the factory makes. The result can inform work on these conditions; it cannot automatically describe denim, delicate fabric or another operation.

Change the order without pretending that learning disappears

If everyone uses A first and B second, B may benefit from practice. If B always runs late in the shift, fatigue or material delays may work against it. Method and order become entangled.

The plan assigns the order within each pair in advance. It uses a random schedule that balances A-first and B-first pairs across the trial. The actual sequence is retained with the records, including any change made on the day.

NIST's randomisation guidance explains random assignment of run order. Here it reduces the chance that the preferred method always receives the more favourable slot. It does not guarantee that every small trial will be perfectly balanced on every hidden influence.

Familiarisation still matters. People need enough practice to use each layout correctly before comparison begins. A fixed two-minute allowance would not establish that for every task or person.

Some learning carries from one method to the other. A worker may retain a better movement even after returning to A. A short break will not necessarily remove that effect. If the team sees such carryover, it needs a revised design or a more cautious interpretation.

The observer should record whether the method was actually followed. A label saying “B” is weak evidence if the operator quietly moved the tray back to A. That response may reveal a usability problem worth understanding.

Define a good unit and the time boundary

The counted unit is one piece that passes the named operation's check on its first inspection. It is not a whole saleable garment. The trial does not add several operations on the same garment and call them several finished products.

Each run starts with ready inputs but no partly sewn pieces from an earlier run. At ten minutes, completed pieces are identified. Any unfinished piece is tagged and kept outside that run's completed count. Its later completion does not become an extra observation in the next run.

All completed pieces receive the same quality check. The inspection can occur after the timed sewing window, but its labour is recorded separately. A piece that fails first inspection is excluded from first-pass accepted output even if it is later repaired successfully.

The run clock includes ordinary handling and small interruptions within the defined operation. It is not stopped whenever a piece becomes awkward. Otherwise, the study could remove the very losses that the method is meant to change.

The quality reviewer needs clear acceptance criteria. Where practical, coded batches can hide the layout from the reviewer. That can reduce preference in judgement, although the operator and observer cannot be blinded to a visible tray change.

Counting should be rehearsed before the main comparison. The observer and quality reviewer can trace a small batch from input through completion and inspection. Missing identifiers discovered afterward are much harder to fix without guesses.

Make the record easy to audit

One row per run should link the timing sheet to the inspected pieces. A separate note can hold the exact layout description and approved photograph or sketch used on site. Keep enough detail for another team to repeat the method.

The production manager does not need a public ranking of named workers. Access should suit the task and local privacy rules. The people reviewing the trial do need stable operator codes, so they can see when several observations came from the same person.

Read the fictional results in more than one way

The table combines the four operators and two days for each style. Each method has eighty operator-minutes for each style. Every completed piece was inspected; there are no missing quality outcomes in this example.

Layout B produces 28 more first-pass accepted pieces during equal operator time. Its local accepted-output rate is higher:

A rate = 468 ÷ 160 = 2.925 accepted pieces per operator-minute.

B rate = 496 ÷ 160 = 3.100 accepted pieces per operator-minute.

Relative increase = (3.100 ÷ 2.925 − 1) × 100 = 6.0%, rounded.

The combined figure hides a difference between styles. Accepted output rises from 259 to 282 for the light style, about 8.9%. For the heavier style, it rises from 209 to 214, about 2.4%.

The paired records reveal another limit. For one operator on the heavier style, A yields 50 accepted pieces over the two days while B yields 49. A positive total does not mean every person and style benefited.

That result deserves a discussion and another observation. It does not prove that the operator resisted change, nor that B can never work for them. The count alone cannot identify the cause.

Quality has its own denominator

A has eight pieces failing first inspection out of 476 completed. B has nine out of 505. Their first-pass acceptance rates are:

A = 468 ÷ 476 × 100 = 98.3%.

B = 496 ÷ 505 × 100 = 98.2%.

These rounded rates are close. They do not establish that quality is equivalent, and neither rate is a contractual acceptance decision for a shipment. A small trial may miss an infrequent but serious defect.

Review defect types as well as totals. One extra minor fault and one safety-related product defect have different implications. A predefined serious defect should trigger the agreed stop or containment response, even if the overall percentage looks good.

Keep the rejected pieces and findings traceable to their run. Repairs, disposal or release follow the factory's normal authorised controls. The trial is not permission to ship uncertain work.

Include the labour that the workstation count leaves out

The feeder finds B more demanding to replenish. In this invented record, feeding takes sixteen minutes for A and twenty-four for B across their respective runs. Inspection takes twenty-four minutes for A and twenty-six for B.

These are recorded teaching values, not estimates derived from the output percentage. They include the normal support work within the study boundary. Extra work done only to administer the experiment is logged separately.

Using this broader boundary gives a smaller gain:

A = 468 ÷ 200 = 2.340 accepted pieces per counted labour-minute.

B = 496 ÷ 210 = 2.362 accepted pieces per counted labour-minute, rounded.

The relative increase is about 0.9%, calculated from the unrounded rates. This measure includes operator, feeding and inspection labour only. It does not represent every labour cost in the factory.

The difference between 6.0% and 0.9% is useful. It shows where part of the apparent gain went. It does not make the first calculation wrong; it makes its narrower boundary visible.

The team might improve replenishment before another trial. It should label that arrangement B2 and record what changed. Combining B and B2 observations as though they were one unchanged method would blur the answer.

Ask whether the product mix changes the result

Equal trial time for both styles is useful for comparison. It may not match the factory's next order. The overall six-percent figure depends partly on how much light and heavy work the trial contains.

Suppose a planning question concerns twenty percent light-style time and eighty percent heavier-style time. For this illustration only, assume the measured operator rates hold and switching creates no extra loss. These are strong assumptions that still need testing.

For A, the light-style rate is 259 ÷ 80 = 3.2375 accepted pieces per operator-minute. The heavier-style rate is 209 ÷ 80 = 2.6125.

The weighted A rate is:

0.20 × 3.2375 + 0.80 × 2.6125 = 2.7375.

For B, the two rates are 282 ÷ 80 = 3.525 and 214 ÷ 80 = 2.675. The weighted B rate is:

0.20 × 3.525 + 0.80 × 2.675 = 2.8450.

That gives a conditional local gain of about 3.9%, rather than 6.0%. It still excludes support labour. It is a sensitivity check on the arithmetic, not a forecast supported by a full production study.

The weights represent shares of operator time. They are not shares of units, sales value or order count. Substituting those other weights would answer a different question and could distort the result.

A planner should therefore retain style-level evidence. Once the product mix changes, a single pooled improvement percentage may no longer describe the proposed work. A method that helps light pieces more than heavy ones needs a scope that reflects that difference.

Treat repeated pieces as observations, not independent workers

The trial contains 981 completed pieces: 476 under A and 505 under B. It still involves only four operators. Pieces made by the same person during the same run share conditions and can be related.

The NIST glossary of experimental design distinguishes the unit receiving a treatment from the measurements made on it. Here, the method is assigned to a timed operator run. A large piece count does not create 981 separately assigned trials.

Nor are the thirty-two runs automatically independent. The same people appear across styles and days. Any formal analysis needs to respect the pairing and repeated use of operators.

For an early review, retain each run's counts and show the paired differences. Separate the styles and examine the run order. An average with no underlying record can hide unstable results or a difference driven by one run.

A statistician can help design a larger study when the decision requires an estimate with stated uncertainty. The needed sample depends on variation, the smallest worthwhile difference and the desired strength of evidence. There is no sound rule that four operators or thirty observations will suit every method change.

These teaching figures support arithmetic and questions. They are not accompanied by a claim of statistical significance, proven causation or a factory-wide six-percent gain.

Agree what happens when a run goes wrong

Some interruptions are part of the method. If B repeatedly catches a piece during loading, deleting those runs would conceal its weakness. Record the problem and include it in the practical assessment.

Other events may defeat the planned comparison. A power outage, a wrong material lot or a failed timer can leave a run unusable for its intended purpose. The rule for handling such events should be decided before results are known.

Retain the original record, explain the reason and identify any replacement run. Do not quietly replace a slow observation with a better one. If a whole pair is needed for analysis, show what happened to its unmatched partner.

Safety takes priority over completing a data sheet. The plan should name who can stop the trial, how people report concerns and who decides whether a restart is suitable. A missing measurement after a stop is not a failure by the worker.

If quality inspection is delayed or a batch is lost, do not fill its accepted count with the overall average. Mark the outcome as missing and investigate the record gap. Missing evidence should remain visible in the conclusion.

Confirm the method under the work it must survive

Ten-minute runs do not reproduce a whole shift. They may miss tray depletion, refill peaks, end-of-shift fatigue or a rush of difficult sizes. An observer's presence can also change behaviour.

The next stage should address the uncertainties found in the first. For this example, that includes the heavier style, the one operator with a lower accepted count and the extra feeding work. The follow-up should include normal interruptions and the workers who will use the arrangement.

NIOSH's guidance on evaluating ergonomic changes calls for follow-up to check that controls reduce risk without creating new problems. Worker feedback and quality results form part of that wider review. A short period without a complaint does not prove that a repeated task is harmless.

Ask about the whole task, including loading and clearing. A tray that works well at the start may become awkward near empty. A person may also adapt by leaning or raising a shoulder, preserving speed at a cost the output sheet misses.

ApparelDiary's hazard identification and risk assessment guide explains the separate risk review. The method trial supplies observations to that work; it does not replace it.

The line also needs to use the gain. If the next operation is full, faster sewing may only create a larger queue. ApparelDiary's line-balancing guide provides the wider capacity context.

Make the decision traceable

The example gives a reason to investigate B further. It does not yet justify replacing A everywhere. The local output gain is uneven, support labour absorbs much of it, and longer exposure remains untested.

A useful decision record states the exact method version, people and styles covered. It gives accepted output, quality findings, labour boundaries and unresolved concerns. It also states whether the result supports rejection, redesign, a wider trial or controlled adoption.

Adoption then requires the factory's normal approval, training and document controls. This trial record should feed those processes rather than quietly changing an instruction at one station.

Finally, do not turn the best observed run into a new standard minute value. The trial did not establish a normal sustained pace, allowances or the full basis of a time standard. Its purpose was to compare methods under stated conditions.

The useful result is a decision that another person can examine. They should be able to see where the extra pieces came from, what they required and which questions still need an answer.

References

• NIST: setting the objectives of an experiment.

• OSHA: sewing-station material handling and stitching risks.

• HSE: consulting workers on health and safety.

• NIST: randomized block designs.

• NIST: random assignment and run order.

• NIST: experimental units, replication and related terms.

• NIOSH: evaluating an ergonomic intervention.