Manufacturing

First pass yield: why one batch can produce three different percentages

Use five inspections to separate first-time quality from successful rework and decide where to investigate.

5 min readBy Mayur Kale

A quality engineer wants to know how many units clear final inspection on the first attempt. The manufacturing CSV contains five inspection events for four serial numbers in one batch. P2 fails and then passes a second inspection. P4 fails its only recorded inspection. P1 and P3 pass immediately. The business grain is one serial at the final station, while the source grain is one inspection attempt.

Count the four first inspections and the two that passed. First-pass yield is 50%. If the question is how many serials have passed by the end of the observation period, P2 joins P1 and P3, giving three of four, or 75%. Dividing successful inspection events by all attempts gives three of five, or 60%. Each calculation uses the same source table. Only one supports a claim about first-pass yield.

On the Manufacturing worksheet change the second attempt for P2 from pass to fail. First-pass yield stays at 50% while eventual passes fall to two. Change P4's first attempt to pass and first-pass yield rises to 75%. Before applying the method to a plant, check whether serials can return from another line or be reused. Enforce uniqueness for serial, station and attempt within the agreed scope. A missing first attempt should enter an exception queue. Counting the earliest observed attempt as the first would overstate yield whenever a failed inspection went missing.

Download the example workbook. The formulas cover the supplied rows. Extend the ranges before adding new records.

Start with the part that came back

Imagine watching P2 reach final inspection. The operator finds a defect, records the failure and moves the part to a rework bench. A colleague corrects the defect and sends the same part through final inspection again. The second inspector signs it off. Shipping has one acceptable unit. Final inspection has processed two attempts. Production has used extra labour to recover a part that missed the intended standard on its first pass. Each team has a legitimate interest in the part, and each needs a different count.

Put a mark on a piece of paper for each physical unit entering the station. Under each mark, write the sequence of inspection outcomes from left to right. Cover everything except the first outcome and count the passes. Then uncover the rest and count how many units eventually passed. Explaining the difference to a supervisor this way usually reveals the local meaning of a retry. In some plants an operator rescans the label to reprint a certificate. Find out whether that creates another inspection record before treating it as another test.

Choose the station and the population

First pass yield needs a defined starting population. In the sample every serial reaches the same final station, so it is straightforward to count entries there. In a plant with parallel lines, some products may bypass final inspection under an approved sampling plan. Others may be inspected twice at different stations as part of their normal route. Trace one product of each type with the quality engineer. Mark which visit counts as its first eligible attempt and why. Leave room for products whose route is still unknown.

A station measure answers how well units entering that station perform. It tells you little about units scrapped upstream. If ten assemblies begin production, two are scrapped before final inspection and seven of the remaining eight pass final inspection on their first attempt, the station result is seven out of eight. A production manager asking how many starts became first-time acceptable finished assemblies is asking about a different starting population. Carry the two scrapped units into that discussion instead of quietly losing them at the station boundary.

Choose how to assign a unit to a reporting period. Suppose a part starts at the end of the night shift, fails and is reworked after the day shift arrives. For a report on first attempts, the first inspection time can assign the unit to a shift. For labour planning the rework time belongs to the shift that did the work. If you attach the entire history to whichever shift happened to finish the part, the night team appears to have produced fewer units and the day team inherits failures it never inspected.

Distinguish a correction from rework

An inspector can record the wrong outcome and correct it a minute later. That is a change to the inspection record. An operator can physically repair the unit and submit it again. That is another inspection attempt. Ask the quality lead to show one example of each in the operational system. Look for an attempt identifier, a revision identifier and an explanation for the amendment. If the source cannot tell those situations apart, keep the ambiguity visible in the report and agree how staff will resolve it.

Test the difference using P2. Make a copy of its failed first attempt as though a scanner sent the same message twice. A report that now counts five inspected serials has confused deliveries with units. Now replace the first result with an authorised correction to pass. First pass yield should change because the recorded judgement on the original inspection changed. Finally, restore the failure and add a third successful attempt. Shipping may care about the latest acceptance, but the first inspection still failed.

Work out how long to wait for missing events. An offline inspection terminal might upload the second attempt before the first. Publishing the earliest received pass immediately could flatter the shift that just closed. One option is to mark the affected serial pending until the source sequence is complete. Another is to publish a provisional number together with the count awaiting investigation. Decide this with the person who acts on the shift report. They can tell you when a quick provisional answer is worth more than waiting for a final one.

Turn the percentage into an investigation

Two of the four example serials failed on their first attempt. Ask which defect codes were recorded and whether the parts shared a component lot, tool or setup. The supplied records have no defect or tool fields. They can identify P2 and P4 for follow-up, but they cannot establish why either failed. A second extract should include the inspection reference and defect observation so an engineer can join it to the correct attempt. Joining every defect ever seen on P2 to its first inspection could implicate an unrelated later problem.

If one failed attempt has three recorded defects, preserve the distinction between counting defects and counting affected units. A quality manager ordering corrective work may want the frequency of each defect. A production supervisor preparing the yield report needs to count that serial once. Display both figures with plain labels. When the sum of the defect categories exceeds the number of failed parts, the labels should already explain how that can happen.

Before signing off, give the next shift supervisor four serial cards and ask them to predict the result when one card changes. Have them find the corresponding record in the reporting system, check the shift assignment and identify who can approve an amendment. Keep a sample report with the four initial inspections, the two first-time passes and the rework outcome written beside it. The practical test is whether somebody operating the line can follow a reported fall in yield back to the parts that need attention.

Review a change in product mix

Suppose a line produces two product families. One has a long, difficult inspection and the other has a short routine check. A week with more of the difficult assemblies may produce a lower overall yield even if both families performed exactly as they did the week before. Break the weekly change into the first passes and inspected units for each family. Ask the planner whether the production schedule changed before assigning the movement to an operator or a new machine setting.

A small batch also calls for care in the wording of a shift note. In the four-unit example a single changed first inspection moves the result by twenty-five percentage points. Record the serial and the reason for the change beside the percentage. Over several shifts, combine the pass counts and the inspected counts for a more stable view of that product. Preserve the individual shift observations as well. The quality engineer investigating a tool change may need precisely the local sequence that a longer reporting window smooths away.

Turn the trend into shipped software

I contract as a senior fullstack and data engineer, building the production systems your customers and your reporting depend on.

Explore services↗