AOI Holds: When First-Pass Yield Recovers Then the Same Pad Fails on the Next Panel

September 22, 2026Emma Johnson

AOI camera over a PCB on an SMT line in a dark factory

AOI Holds

A electronics manufacturing hold starts when two records disagree. AOI Holds is the version of that disagreement that shows up on the floor first. The passing number is already printed. The failing number arrives a few minutes later, or on the next shift, or from a second instrument. The wrong move is to average them, to retest until one matches, or to write a note that the first result was a glitch.

Write the time of the first mismatch in the batch or lot record before anyone touches the equipment again. Name the two instruments or methods. Name the person who saw each result. If the line is still running, stop it. If product already left the cell, identify the last good unit by serial, VIN, bag, vial, or heat number. Do not wait for a full investigation charter to do that identification.

The hold tag belongs on the physical material and on the electronic status in the same minute. A tag on a tote with an open status in the system is how mixed lots happen. A closed status with no tag is how someone ships from the dock. Both failures are inspection findings because they are visible in a walk-through.

Pull the raw data, not the summary. Balances, checkweighers, HPLC sequences, historians, CMM programs, torque controllers, NDT technique sheets, DCS alarm logs, gas-detector bump logs, pipette gravimetric sheets, AOI recipes, reflow profiles, sealers, and mill certificates all keep more than the pass/fail bit. The extra fields are what let you decide whether the passing result was real.

Identity is the first reconstruction. Confirm the lot, drum, cassette, film roll, VIN, blade serial, tank, pipette ID, PCB panel, catheter lot, pouch lot, gauge-block set, or heat number against the document in hand. A correct number on the wrong object is a different failure than a drifting instrument. Mixing those two stories in one CAPA wastes weeks.

Calibration and suitability come next. Print the last calibration or SST or bump or intermediate check with as-found data. If as-found was already out, every result since that date is in the hold, not only the unit in front of you. If as-found was in, look at the next layer: tare, blank, coupon, redundant probe, second method.

People and sequence matter. Second-person verification that stamps after the fact does not reconstruct a weigh. A supervisor who 'knew the line was fine' is not a data source. Interview the operator with the record open, not in a conference room. Ask what they saw, what they did, and what they were told to do. Write their words. Do not translate them into quality language until the record is complete.

Time alignment is where most of these holds stall. Clocks on balances, historians, printers, and badge readers drift. Convert everything to a single time base before you draw a timeline. If you cannot, say so in the record. A pretty timeline with mixed clocks is worse than an ugly one with a note.

Decide the material disposition with written rules, not mood. Typical options are continue under hold pending test, rework under a protocol, reject, or return to vendor. Each option needs a named owner, a sample plan if testing is involved, and a stop date. Open-ended holds become unofficial inventory.

If you retest, the protocol has to state why the original result is still in the file. Retest that silently replaces the first number is the finding. Keep both. Explain the scientific reason the second method or second sample is informative. If there is no scientific reason, you are shopping for a passing number.

Change control sits beside the hold when the likely cause is a recipe, a fixture, a film lot, a cassette lot, a sealer temperature, or a technique sheet. Do not 'tweak and retry' on the same lot. Document the proposed change, the risk, and whether the current lot can be treated under the old or new condition. Mixing conditions inside one lot is how you fail the next audit of that lot.

Supplier and incoming identity belong in the same hold when the second record is a certificate, a heat number, or a film COA. Notify the supplier with the actual mismatch, not a vague quality complaint. Ask for the mill or converting record that should have prevented the mix. If they send a corrected CoA with no explanation, that certificate is not evidence.

Training records are supporting evidence, not the root cause by default. If the operator followed the SOP and the instruments still disagreed, rewriting the SOP might still be needed, but blaming the person first hides the instrument or identity problem. If the operator skipped a step that the SOP required, say which step and show the record gap.

Electronic audit trails must show who changed status from hold to released. A release with no hold history is a configuration problem. A hold that disappears when someone re-prints the ticket is a printer or template problem. Test both before you close.

Write the investigation so a future inspector can walk the same path. Attach the two disagreeing printouts, the identity photos, the calibration, the timeline, the disposition, and the CAPA number if one opened. Do not point to a shared drive folder as the record. The batch file has to carry the story.

Metrics that help: time from mismatch to physical hold, time from hold to identity confirmation, percent of holds closed with both original results retained, and repeat holds on the same instrument or supplier in 90 days. Vanity metrics such as 'holds closed same shift' push people to release early.

What this looks like in practice on a busy shift: the passing result is already in the packet. Someone wants to keep the line moving. The quality person on call has ten minutes. Those ten minutes are for tagging, freezing data, and naming the two records. The rest of the science can wait until morning. The ten-minute failure is the one that ships.

Regulators and customers read holds as a test of whether the quality system can stop work. A clean hold with both results, a frozen electronic status, and a dated tag is stronger than a long CAPA that starts after the product left. Build the stop first.

How to freeze the electronic status

Use the same status code the warehouse and the lab already know. Inventing a new 'soft hold' in email is how product moves. If the system only has released and rejected, add a true hold state in change control, and until that lands, use rejected-with-comment plus a physical tag and a written list of locations.

For aoi-hold-electronics-2026-09-21, keep a running list of every person who touched the hold: operator, verifier, quality on call, maintenance, and the person who changed electronic status. If a name is missing, the record is incomplete.