Why the Same Container Glass SKU Runs Differently on Every Shift

It's 06:15 on a Tuesday and the day-shift operator is staring at his IS settings with a look you recognise immediately if you've ever stood on a hot-end floor. The night crew ran the same 375ml wine bottle for eight hours. Same declared recipe. Same mould set. Same furnace pull rate. Different everything else.

He'll spend the first 45 minutes chasing it back. That's not unusual. That's most mornings on most forming lines running without systemised standard work.

The variance is real, and it's bigger than most plants measure

The 30-60% efficiency spread on identical SKUs across shifts is not a staffing problem. It's a systems absence. The night crew aren't sabotaging the line. The day crew aren't better operators. What's different is what each shift inherits when they sit down, and what they're permitted to do about it.

In 2017 I was reviewing forming data on a three-line plant in South Australia running a mixed wine and spirits portfolio. We pulled the job change records for one 750ml Bordeaux mould set across six weeks. Same SKU, same IS machine, same glass composition. The efficiency delta between best shift and worst shift was 41 OEE points. Not 4.1. Forty-one. The plant manager thought it was a furnace problem. It wasn't a furnace problem.

What it actually was: three operators maintaining three separate mental models of what the recipe should be, none of them matching the formal spec sheet, all convinced their version was correct. One had the gob weight running at a 0.6% CV because he'd never seen the target was ≤0.4%. Another was trimming the forehearth profile in zone 3 by hand during steady-state, chasing a cord fault that had been fixed two months earlier at source but never communicated downstream. Standard work didn't exist. The SKU Library didn't exist. The variance existed.

What the 0600 handover actually transfers

The 0600 handover on most container glass plants transfers three things: the defect log, the scrap rate, and whatever the outgoing operator chooses to mention. What it almost never transfers is the live set-point state, which parameters were adjusted during the shift and why, or whether a section was temporarily disabled and brought back before end of shift.

I've sat in on handovers where the outgoing operator mentioned "a bit of a leaner problem on section 4, sorted it out" with no record of what was changed to sort it. The incoming operator starts blind. If the leaners come back, he adjusts something else. Something his predecessor would never have touched. Now you have two corrective overlays sitting on top of each other, neither of them traceable, and nobody quite sure which one is making the problem better or worse.

The forehearth profile makes this concrete. A well-run line targets ±2°C across five zones during steady-state pack. When operators manually compensate without documentation, you can find swings of ±8°C that nobody formally sanctioned and nobody formally reversed. The incoming shift inherits the profile in whatever state the outgoing operator left it, and the only way to know the intended state is to go back to a recipe document that may not have been updated since the last job change. That's not a people failure. That's an absent system.

Standard work isn't a laminated sheet above the IS controls

The usual response to cross-shift variance is a laminated sheet taped above the IS machine. Recipe. Target gob weight. Pack spec. It gets ignored within two weeks because nothing enforces it and nothing connects it to what the operator actually sees on shift.

The gap between the declared recipe and the live set-point state is where cross-shift variance lives. Close that gap and the variance closes with it.

Standard work, in the container glass sense, is a living system. The hot-end superintendent owns recipe lock. No operator changes a set point without formal sign-off during steady-state pack. Section timing targets are posted and checked. Within 10ms on a well-set IS machine is achievable. Most plants running without active monitoring drift to 25ms or more. Swabbing intervals are logged and handed over, not improvised per shift.

And it's a SKU Library that makes the whole thing machine-readable. Not a folder of PDFs. A locked, versioned recipe state that the operator loads at the start of the job and cannot silently override. That's the core function of the Job Change Tool: replacing three separate mental models with one verified ground truth. When plants make that shift, cross-shift variance on identical SKUs typically falls 40-60% within 90 days. That's not a theory. That's what happens when you replace guesswork with a system.

The management layer that makes standard work stick

Standard work degrades under production pressure if the management layer isn't reinforcing it. Any system without active oversight drifts toward the path of least resistance, and the path of least resistance is an operator making a call and not documenting it.

Leader Standard Work is what most plants I've audited are missing. The hot-end superintendent needs a structured check with specific targets and a record. Did the recipe load match the SKU Library spec? Was the forehearth profile within ±2°C at T+45 min after job change? Did the handover capture the live set-point state?

In 2023 I audited a four-furnace operation in the GCC region running an older Emhart I.S. eight-section fleet on legacy Siemens S7 PLC controls. The variance problem there was compounded by unstructured management rounds. Shift managers walked the floor, asked how things were going, and got the answer operators always give: "Fine, all good." The data told a different story. Without a structured check tied to specific KPIs, the walk adds nothing. You feel busy. The variance stays.

Once you connect the Leader Standard Work cadence to the same KPI set the SKU Library tracks, the conversation changes. The superintendent stops asking "how are things going" and starts asking "show me the section timing log from the last four hours." On that same plant, baffle marks had been attributed to glass composition for two quarters. The real cause was baffle alignment drift, visible in the section timing data every single time the problem recurred (and yes, the centring tool had been logged as 'checked' on the maintenance record without anyone physically verifying the reading). The variance closed. Not because the operators changed. Because the system gave them one ground truth to work from.

What a five-day audit actually finds

Reducing cross-shift variance in container glass is not complicated. It's just not comfortable. It means locking recipes that operators have been adjusting at will for years. It means a management standard that asks specific questions instead of open-ended ones. It means a handover record that captures set-point state, not just defect counts.

A vendor-neutral container glass consultant looking at your forming line for five days will almost certainly find the same things: undocumented set-point changes during steady-state pack, a handover that doesn't transfer recipe state, and management rounds that don't check the numbers that matter. The findings aren't surprising. The size of the recoverable OEE usually is.

So what's stopping your plant from running that audit this quarter?