Two Yield Numbers and One Scorecard
You manage quality data for a contract electronics assembler. The weekly customer scorecard is generated from the manufacturing execution system and reports first-pass yield of 96.8% for March. The customer's own incoming inspection data for the same shipments implies 91.2% - a difference of roughly 380 units. Investigating, you find the definitions differ. The MES counts a unit as first-pass if it leaves final test without a rework routing. The customer counts any unit that needed touch-up at any station, including rework done at sub-assembly before final test, and their figure also includes units your line re-tested after a fixture recalibration on 12 March. Your operations director has already told the customer that 96.8% is the system of record and asked you to publish the monthly report on that basis. The customer's supplier scorecard determines whether the account is renewed in September, and the gap is large enough to put you in the lowest tier.
1. Is either figure wrong? Set out the reconciliation before deciding what to publish.
Read the worked answer
Neither number is wrong; they answer different questions, and the danger is publishing either one as if it answered both.
96.8% measures final-test first-pass on units that reached final test. 91.2% measures units shipped without any rework at any station, plus the 12 March retest population. The sub-assembly rework is real work on real units and the customer is entitled to count it; the MES simply never saw it because the rework routing was raised one level down and never rolled up.
So the reconciliation has two parts. First, a definitional bridge: 380 units is the sum of sub-assembly touch-ups plus the recalibration retest cohort - verify that the arithmetic actually closes, because if it does not, there is a third cause nobody has found yet. Second, a defect in your own reporting: the MES roll-up does not capture lower-level rework, which is a system gap worth fixing regardless of this customer.
Publish the bridged view - 96.8% on your definition, 91.2% on theirs, and the units that explain the difference. A reconciled pair is credible; a defended single number is not.
2. The operations director wants the 96.8% figure published alone and says the definitional argument can be handled later. How do you handle it?
Read the worked answer
The director is not asking you to lie, but they are asking you to omit the one piece of context that makes the number meaningful to the reader - and the reader already has the other figure. Publishing 96.8% alone in September, after the customer has circulated 91.2%, will read as concealment even though nothing was concealed at the time.
Response: agree with the substance where you can. The director is right that the MES figure is the system of record for final-test yield; nobody is disputing that. Then propose the version you can both sign - headline 96.8% with a one-paragraph reconciliation note and the definitional difference stated. That protects the account better than a bare number does, because it demonstrates you understand your own data.
If the director insists on the bare figure, ask for the instruction in writing and copy the account quality contact. Do not publish a figure you believe will mislead, and do not reclassify the sub-assembly rework retroactively in the MES to make the gap disappear. The data should be fixed going forward, not rewritten backward.
3. What is the corrective action plan, and what would you fix first?
Read the worked answer
Fix the measurement before fixing the yield. The 3.2-point gap exists because one system counts work and another does not; every improvement project launched on the 96.8% baseline is chasing a number that excludes part of the process.
First: extend MES rework capture down to sub-assembly so any touch-up rolls into the unit's history, and agree a written definition with the customer covering station scope, retest after equipment recalibration, and how scrap is treated. Both sides sign it - that is what prevents a repeat next quarter.
Second: treat the 12 March fixture recalibration as a separate investigation. Units retested after a fixture fault may have been passed by a fixture that was out of adjustment, so the question is not only how many were retested but whether any escaped. That is a containment question, not a reporting one.
Third: only then run yield improvement against a stable, agreed definition. Chasing the delta before the definition is settled produces activity that looks good on a chart and changes nothing on the line.