How to calculate first-pass yield when rework hides the result

Calculate first-pass yield by dividing the number of distinct units that meet the defined requirements on their first attempt by the number of distinct units entering the same process. Multiply by 100. A unit that passes after rework belongs in final accepted output, but it does not become a first-pass success.

That distinction answers a common shop-floor question: how can almost everything be accepted while the team is still spending hours fixing it? The two measures describe different outcomes. Keep both, and reconcile the units behind them.

This guide is for manufacturing quality and operations managers, with US search demand informing the reader question. Its counting method is applicable beyond the United States. It is a proposed internal reporting method, not a regulatory acceptance rule, certification requirement or instruction to release a product.

The American Society for Quality's glossary excludes units that need reruns, retests or repair from first-pass yield. Minitab likewise distinguishes units that pass without rework or scrap. These definitions concern the first journey through a defined process, not simply the stock eventually available to ship. ASQ quality glossary · Minitab throughput yield guidance

Define the process boundary before extracting data. In an assembly-and-test operation, the boundary might begin when a uniquely identified assembly enters and end at its first completed acceptance decision. Normal planned operations and tests inside that boundary are part of the first attempt. Repeating an operation or test to recover from a failed attempt is different.

Write down the acceptance criteria and the treatment of invalid tests. If a documented equipment fault makes a result invalid, preserve the original event and its authorised disposition. Do not quietly delete failed results or relabel remedial work as normal processing to improve the percentage. Unresolved classifications belong in an exception list until someone responsible decides them.

Count units, not defect observations. One assembly with three defects is one first-pass failure, even though all three defects remain useful for investigation. ASQ's definition of a defective unit allows more than one defect on that unit. ASQ glossary: defective

Every quantity and record in this example is fictional. Assume 800 uniquely identified assemblies enter one defined assembly-and-test process. All have a known first-attempt result and a final disposition. The entry population is fixed; no units are merged, split, substituted or left open. The same acceptance criteria apply throughout.

First-pass yield is 680 ÷ 800 × 100 = 85.0%. The 120 first-pass failures are 90 plus 30, counted once each.

Final accepted yield for this completed cohort is 755 ÷ 800 × 100 = 94.375%, displayed as 94.4%. Final scrap is 45 ÷ 800 × 100 = 5.625%, displayed as 5.6%. The 75 recovered units explain the 9.375 percentage-point gap between first-pass and final accepted yield. That gap is not a 9.375% relative improvement and is not evidence that the underlying process improved.

The reconciliation has two separate views: 680 + 90 + 30 = 800 first-attempt outcomes; and 755 + 45 = 800 final dispositions. Do not add the 90 reworked units to the 45 finally scrapped units and call all 135 different failures. Fifteen units appear in both categories because rework is a journey and scrap is their final outcome.

The rework-entry rate is 90 ÷ 800 = 11.25%. Recovery among units sent to rework is 75 ÷ 90 = 83.33%, rounded. Neither is FPY. Naming the denominator makes each result understandable.

Use one row per unique unit for the yield calculation. Keep the detailed test and rework events linked to that row. If the 800 assemblies generate 910 test events, the extra events do not create 110 new assemblies. Event-level pass rates answer a different question and must not be labelled unit FPY.

Here is a completed fictional control record for the batch above. The record references are invented examples, not evidence of an actual factory test.

• Record and boundary: FPY-017-v1; assembly entry to first acceptance decision
• Population: Batch B-017; 800 unique unit IDs; all outcomes resolved
• Criteria: Work instruction WI-07 rev C; identical criteria across the cohort
• First-attempt source: FIRST-B017; 680 pass, 90 rework, 30 direct scrap
• Final-disposition source: CLOSE-B017; 755 accepted, 45 scrapped
• Counting control: One initial outcome per unit; repeated events linked, never added as units
• Review decision: Report 85.0% FPY alongside 94.4% final accepted yield; investigate the 120 first-pass failures
• Owner and next action: Quality manager checks classifications; process engineer examines failure modes and proposes a controlled trial

For a real extract, retain the unit ID, product and revision, process boundary, first-entry timestamp, first-attempt result, any remedial-work flag, final disposition and source references. Record the extract time and reporting convention. Preserve a correction trail when a duplicate or classification error is found.

Check for missing IDs, repeated initial outcomes and impossible sequences before calculating. If the source only stores the latest passing result, it cannot reconstruct FPY reliably on its own. Recover the attempt history or disclose the limitation; do not assume that a current pass was a first-attempt pass.

Use a fixed entry cohort and follow it to resolution when you need a completed-cohort FPY. That is the reporting convention proposed here. A daily operational dashboard can use a different documented convention, but its results should not be silently substituted for the completed-cohort measure.

Consider a separate fictional cut-off snapshot of 800 entrants: 680 have passed first time, 100 have already failed first time, and 20 have not yet completed their first attempt. Assume the 20 are genuinely unresolved and could still pass first time, with no remedial work already recorded.

The completed-cohort FPY is not yet known. Its eventual value is bounded by 680 ÷ 800 = 85.0% if none of the 20 passes first time, and 700 ÷ 800 = 87.5% if all do. This is a logical range under stated assumptions, not a statistical confidence interval. Display the counts and unresolved balance with it.

Calculating 680 ÷ 780 = 87.18% describes only units whose first-attempt result is known. It is not the final rate for all 800 entrants. Removing the unfinished 20 can make a dashboard look better without any new success. If records are missing rather than merely unfinished, first establish what is known; the range above is valid only with a verified total population and trustworthy classified outcomes.

Likewise, do not divide today's completions by today's starts when they are different units. Link a cohort across shifts or report a clearly defined flow measure. Our guide to KPI denominators explains how to document population changes without presenting them as performance gains.

Start with the first-pass failures, group their observed failure modes, and trace them back to the relevant operation, material, specification and measurement evidence. A frequently recorded symptom is a starting point for investigation, not proof of its cause. Check whether the test itself is trustworthy before blaming the process or operator.

In the completed example, 75 units were recovered. That shows why final accepted yield alone misses extra work. It does not tell you how many labour hours were used, whether a change would save cash, or which defect creates the greatest customer or safety consequence. Keep those decisions separate. Use a reconciled cost-of-poor-quality ledger if you need to assess the cost of the rework.

For a proposed improvement, state the suspected cause, change, responsible person and acceptance conditions before the trial. Compare later cohorts with the same boundary, criteria and product mix, showing both rates and counts. A lower inspection standard or removal of difficult products can raise the rate without improving the original process.

For a multistep line, do not average the step percentages and call that end-to-end FPY. Minitab describes rolled throughput yield as the product of the step yields. Applying that approach to your records requires compatible populations and routing; do not multiply unrelated weekly or product-level percentages and present the result as an observed count of units that passed every step first time. Unit-level route histories provide a direct check. Minitab rolled throughput yield guidance

There is no target percentage established by this example. Judge a local target against requirements, process risk and comparable evidence. A high FPY cannot authorise shipment, replace required inspection or prove that a product is safe. The practical goal is a report that distinguishes first-time success, recovery and unresolved work clearly enough to support the next investigation.

ASQ Quality Glossary, entries “First pass yield (FPY)” and “Defective”; current public text checked 17 October 2026. The page does not state a current revision date for these individual entries.

Minitab: What are throughput yield and rolled throughput yield?, current public support guidance checked 17 October 2026; no publication or revision date stated. These are explanatory resources, not legal requirements.

The cohort record, calculations, cut-off range and reporting controls are QSE Advisory's illustrative analysis. All operational inputs are fictional. Preparation included AI-assisted research, drafting and calculation checks. Reviewer and Assessor: Laraib Naveed Ahmad. Questions or corrections: office@qseadvisory.com.

This article is general information. Verify current requirements and seek qualified advice for your specific circumstances.

Spotted an issue? Contact our editorial desk.