Why Paper Checklists Fail the Container Glass Job Change

It is 2am on a Tuesday and section 4 has just come back up after a colour change. The hot-end inspector pulls a bottle off the conveyor and holds it to the light. Checks. Hairline cracks running around the finish ring at 12%, rising. The shift operator looks at the job card. “Moulds pre-heated: yes.” That is all it says.

No temperature reading. No timestamp. No way to know whether that tick went down at the start of shift or three hours ago. The specified preheat window for that mould set is 400–480°C, measured by IR pyrometer at mould opening. Nobody wrote 400°C. They wrote “yes” and the bottles are now telling us the truth instead.

Not a system. A liability dressed as a procedure.

What tribal knowledge actually costs on the forming line

Most plants that haven't systemised their changeover run on some version of the same arrangement: one experienced operator who knows a particular SKU, a job card that was accurate when it was first written, and a 6am handover where the outgoing shift tells the incoming shift what they did, not what the specification says to do. In 2017 I was running a 4-line plant in Western Australia where one section operator kept a spiral notebook with forming parameters for every SKU on the line. Brilliant operator. Knew his machine better than anyone on site. When he went on annual leave, his section ran 6 OEE points below the plant average for three solid weeks. Not because the job card was wrong. Because the job card didn't exist in any useful form.

That is the real cost of tribal knowledge. Not the experienced operator who eventually leaves. It is the invisible cross-shift variance that runs every day he is present, because no two shifts load the same recipe the same way. Industry data puts this variance at 30–60% on plants that haven't locked their changeover process. A 1% OEE improvement at hot-end scale translates to millions in EBITDA. The recoverable margin is sitting on every line, every shift, and the paper checklist is the reason nobody can see it clearly.

The Glass International 2023 benchmarking survey across 47 container glass plants globally found that plants using paper-based job-change checklists averaged 5.8 hours of changeover downtime per event. Plants using digital or systemised tools averaged 4.0 hours. That is a 45% reduction in changeover duration. Ardagh Group's 2023 sustainability report puts average OEE across their North American container glass fleet at 78%, with job-change downtime averaging 2.1 hours per event, identified as the single largest source of unplanned availability loss. These are not outliers. They are what paper-based changeover produces at scale.

What a paper checklist can and cannot actually verify

The checks defect I described at the opening is a recurring one with a known cause. Blank-mould or finish-mould temperature below 380°C at job-start is the primary driver. The operating window is 400–480°C measured by IR pyrometer at mould opening. A paper checklist records “mould pre-heated: yes or no.” It cannot record 392°C. It cannot capture that the preheat was done 45 minutes before the mould went in because the section wasn't ready, and the mould had cooled below spec before the job started. The tick says yes. The hot-end inspection machine sees checks. The shift blames the glass.

On NNPB jobs, plunger stroke must be set within ±0.3 mm of the job card specification. A 0.5 mm overtravel produces wall-thickness non-uniformity above 0.25 mm on the shoulder, failing the 0.15 mm upper spec limit on inline thickness gauges. The paper job card has a box for plunger stroke. It does not enforce that the number entered is within spec. It does not timestamp the entry. It does not link the entry to the operator who wrote it. It cannot flag when section 6 is 0.6 mm out while sections 1 through 5 look fine.

On older Hartford 28 and AIS systems, section timing is re-entered manually at every job change (and yes, I know the argument for running it from memory on a known SKU, but the data does not support it). Counter blow and reheat timing carry a combined interaction effect: a 0.1-second error on counter blow combined with a 0.15-second error on reheat produces detectable cord defects in more than 30% of containers within 20 machine cycles. The paper checklist records the target values. It does not record what was actually entered. It cannot tell you whether sections 3 and 7 drifted before the operator made it back around the machine.

The paper checklist tells you what should have happened. A systemised job change tool tells you what did happen, when, and by whom.

The sequencing problem no tick-box can solve

A job change on a 10-section IS machine involves 14 to 22 individual parameter changes across hot-end, annealing lehr, and inspection. These changes have documented inter-dependencies. The hot-end superintendent owns recipe lock, and the first of those dependencies matters immediately: blank-mould change must precede gob-weight re-zero by at least 8 machine cycles, approximately 3.2 minutes at 150 containers per minute. Get that sequence wrong and the gob-weight baseline is set against a machine state that no longer exists. The first 60 minutes after restart account for 40–60% of total job-change scrap in an unmanaged changeover. Paper cannot enforce any of this.

A checklist is a list. It is not a sequencing engine. An operator can tick step 8 before step 4. An operator on a double shift at hour 11 can tick steps 9 through 14 in confidence because he knows this job. There is no mechanism to catch it, no timestamp to audit it, and no way to build a meaningful post-mortem from what actually happened versus what the card shows.

But the sequencing gap goes beyond the hot end. Hot-end coating application with MBTC requires the container surface to be at 530–600°C. If lehr conveyor speed is not recalculated for the new container geometry and weight during the changeover, the thermal profile shifts and coating adhesion drops. The result is cold-end lubricity failures below 0.35 COF, showing up as line jams 2 to 4 hours post-changeover. By then the job-change record is closed and the paper checklist is filed. That failure mode is invisible to the paper record that was supposed to prevent it.

There is also a regulatory dimension that most paper-checklist users haven't yet confronted. Under the EU Packaging and Packaging Waste Regulation agreed in April 2024, container glass producers face mandatory quality-conformance documentation that effectively requires job-change records, including mould identity, gob-weight logs, and inspection calibration data, to be part of a traceable quality management record. A paper checklist with a single operator tick and no timestamp does not constitute a traceability record. Plants preparing for PPWR compliance audits are going to find that out at an inconvenient moment if they haven't addressed it already.

What a properly systemised approach changes

In 2016, the plant I was running at O-I won the Most Improved Job Change Plant award globally. The improvement did not come from a new IS machine or a capital programme. It came from making the invisible visible: every parameter, every sequence step, every timestamp, every section. When you can see what actually happened at each changeover, you can close the gap between the best shift and the worst shift. In most plants I audit today, that gap runs 30 to 60 minutes per event. At four events per week on a single line, that is two to four hours of recoverable production time. Per line. Per week.

It's not glamorous work. It's the work.

The systemised Job Change Tool that Lean Glass built operates at three levels. The SKU Library locks every recipe, mould set, and forming spec with version control and change history. No more spiral notebooks. No more “the day shift runs it slightly hotter” as an unofficial standard that nobody has documented. The digital execution layer maps to the nine-stage Job Change Lifecycle and enforces section-by-section completion sequencing: step N+1 cannot be confirmed until step N is timestamped and signed off by the responsible operator or changeover coordinator. KPI tracking then trends changeover time, first-ware quality, and time-to-stable-pack across shifts, lines, and SKUs over weeks and months.

This is not a rebadged SMED programme. Generic Lean boutiques applying SMED to container glass consistently misclassify mould pre-heating as external setup, ignoring that preheat timing depends on real-time glass temperature and furnace pull-rate variability. The changeover-time estimates those approaches produce are typically 20–40% optimistic and unachievable without equipment interlocks the methodology never specifies. This is also not OEM software tied to a single IS-machine platform. It is an operator-built methodology with a digital execution layer, vendor-neutral by design, because the sequencing dependencies in a job change do not care whose machine you are running.

The 5.8-hour average is not a fixed cost of doing business. Four hours is achievable with a structured process, and well-run plants with a locked system consistently perform below that. If your plant is still running job changes on paper, that gap is measurable and it is costing you. Talk to a vendor-neutral container glass consultant about what a structured changeover audit looks like for your operation. Book a 30-minute call from our contact page.