How to reduce rework

This guide helps an SME recover capacity it has already paid for. Rework is repeated effort caused by missing information, unclear requirements, weak decision criteria, late corrections or divergent definitions of “complete”. Reducing it starts by locating the loop, tracing the cause upstream and placing controls where they are cheapest.

What this guide is for

The guide is designed to measure how much work is being repeated in one flow, identify the few causes that generate most of it, and install practical barriers at the source.

In many SMEs, this work releases a meaningful fraction of a full-time role without hiring, software or major investment. It is often the guide to use before granting an overloaded department an additional person.

When to use it

Use it when a team asks for more staff because it is constantly redoing work.

Use it when margins erode while volume remains stable.

Use it when lead time is unpredictable without a clear reason.

Use it when the company language is full of “final version 3”, “I sent it back corrected”, “we had to redo it”, or “the customer changed something again”.

It is also valuable in a value-creation context. Rework reduction is one of the few EBITDA improvements that may require almost no capital expenditure, which makes it a typical early post-acquisition quick win.

Before starting

Choose one flow. Quotes, orders, invoicing, design and technical validation are common candidates.

Make one rule explicit: count returns, not blame. Rework usually has structural causes: missing inputs, ambiguous requirements, conflicting definitions of completeness, unclear ownership. If people fear the measurement will become a personal scorecard, correction loops will disappear from email and continue verbally. Hidden rework cannot be reduced.

Operational sequence

Step 1 — Measure first pass yield.

Take 20 to 30 recent completed instances of the flow. For each one, count how many times the work returned to a step it had already passed.

Use traces that already exist: email threads, file versions, changes after confirmation, CRM notes, ERP edits, approval comments.

First pass yield is the percentage of instances that pass through the flow without returning. It is the starting measure. Be prepared for an uncomfortable result: in unmeasured administrative flows, 20% to 40% rework is common.

Step 2 — Classify upstream causes, not correction locations.

For every return, record where the problem was created, not where it was fixed.

A quote may be corrected in technical office, but the cause may be incomplete requirements collected by sales. An order may be re-entered in administration, but the ambiguity may have been created by the order form.

Group causes into concrete categories: missing measurement, requirement interpreted differently, customer change after start, incomplete master data, discount condition changed after approval. Usually three or four causes explain most returns.

Step 3 — Separate internal rework from customer changes.

Customer change requests are a separate category. They cannot always be eliminated, but they can be governed: declared revision rounds, clearer requirement collection and explicit change rules.

Mixing customer changes with internal rework contaminates the diagnosis. It also gives the organisation an easy excuse: “the customer keeps changing their mind”. Count them separately and treat them last.

Step 4 — Move control to the source.

For each main cause, design an entry barrier rather than a final inspection.

Examples: an intake checklist, a required field in a form, a definition of done agreed between the role passing the work and the role receiving it, a quality gate that refuses incomplete input.

The economic principle is simple: the cost of an error increases each time it crosses another step before being discovered. The cheapest control is upstream.

Step 5 — Give the barrier the right to stop.

Most quality gates fail at the exception: “this is urgent, take it now and the data will arrive later”. After three weeks the checklist becomes decorative.

A barrier works only if the receiving role has the explicit right to reject incomplete input, including urgent work. True urgency can have a separate path, but that path must carry a visible cost. Without the right to stop, the previous steps are cosmetic.

Step 6 — Measure again after six to eight weeks.

Repeat the first pass yield measure using the same method and a comparable sample.

A realistic first-cycle target is not perfection. It is cutting the gap meaningfully: for example, moving from 44% first pass yield to 70% with two or three well-placed barriers.

Report the result in hours recovered per week, not only in percentages. “We released 14 hours per week” is easier to understand and defend than “first pass yield improved by 26 points”.

Expected output

The expected output has three parts.

First, first pass yield before and after, measured with the same method.

Second, a cause classification showing what share of returns each cause produced.

Third, source barriers installed, each with an owner and the right to reject incomplete input.

A valuable by-product is the written definition of done between departments. It often survives the exercise and prevents future rework.

Mistakes to avoid

Do not turn rework measurement into an individual scorecard.

Do not add a final inspection when the real need is an upstream barrier.

Do not accept “the customer changed their mind” as a universal explanation before separating customer changes from internal rework.

Do not install a checklist without giving the receiving role the right to reject.

Do not declare victory after one cycle. Rework grows back where nobody watches.

Example

A 28-person window and door company is considering hiring in the quoting office for 38,000 euros per year. The office is overloaded.

A measurement on 25 recent quotes shows a first pass yield of 44%. Fourteen quotes returned at least once, with an average of 2.1 loops, equal to about 16 hours per week of rework.

The causes are classified: eight returns from incomplete or unverified site measurements, four from commercial conditions changed after the quote had started, two from internal calculation errors.

The intervention targets the first two categories. A mandatory site survey form with photos is introduced; the quote does not start without it. Commercial conditions are frozen at the start; later changes reopen the quote as a new version with a recalculated delivery date.

Eight weeks later, first pass yield reaches 76%. Residual rework falls to five hours per week. The hire is postponed. Interestingly, calculation errors also fall, even though they were not directly targeted. With fewer corrections and less rush, the office makes fewer mistakes. Rework generates rework; reducing part of it removes more than that part.

MARTRO connection

In MARTRO’s reading, rework is a thermometer of ambiguity between roles. Classifying causes exposes poorly designed interfaces: who must give what to whom, in what form, and with which completeness criteria.

This is why the exercise is worth more than the immediate hours recovered. Definitions of done between departments are small acts of operational governance. Their absence is often the reason rework existed.

Next step

If the causes point to ambiguous responsibility rather than missing information, continue with RACI.

If the flow with the highest rework is also the company bottleneck, every recovered hour increases system throughput. Continue with bottleneck analysis.

For the conceptual background, read Rework in the Knowledge library.

Frequently asked questions

Is all correction rework? No. Some correction is legitimate learning or customer change. Rework is repeated effort that could have been avoided with clearer input, criteria, ownership or earlier validation.

What is first pass yield? It is the percentage of cases that pass through the flow without returning to a previous step. It is a practical measure of how often the flow gets things right the first time.

Should rework be assigned to individuals? Not at first. Assigning blame makes rework invisible. Start by classifying structural causes. Individual training may be needed later, but only after the system causes are visible.

Where should controls be placed? As close as possible to the source of the error. A final inspection catches problems when they are already expensive.

How soon should we see improvement? Six to eight weeks is usually enough for a first cycle if the barriers are real and have the authority to stop incomplete input.