What Benchmarking Against O-I Taught Me About Plant Discipline

In container glass, benchmarking against O-I gets treated like a cheat code. Get the numbers, copy the standard, watch your OEE climb. I get asked some version of that every few months, usually right after a board has seen a competitor's efficiency numbers in an investor deck. It's the wrong instinct. Some of what travels from a Fortune 500 operator into a mid-size independent plant is genuinely portable. Most of it isn't, and treating a benchmark like a shopping list wastes a year and a consulting budget most plants can't spare twice.

Most of what O-I does well is discipline, not equipment

I spent 15 years at O-I Glass, starting on the floor in Brisbane in 2005 and finishing as plant manager before I left to build Lean Glass. The equipment wasn't the edge. Plenty of independent operators run newer IS machines, newer forehearths, better batch house automation than some of the older O-I sites I worked. What O-I had, and what most plants I audit now don't, was a recipe lock nobody touched without sign-off, a mould change process that ran the same way on a Tuesday day shift as a Sunday night shift, and a reporting culture where a bad number got escalated in hours, not buried until month-end.

That's the part that benchmarks. Not the machine spec sheet.

The number that doesn't transfer is the target, not the method

Plants get this wrong constantly. They see O-I run gob weight CV at 0.3% on a mature SKU with 15 years of process history behind it, and they set that as this quarter's target on a line that changed job three weeks ago. That's not a stretch goal. It's a setup for a bad first-ware call and a QA team that stops trusting the numbers.

A GCC-region operator I audited in 2022 had exactly this problem. Head office had mandated an O-I-benchmarked forehearth profile of ±2°C across five zones, copied straight out of a vendor case study, onto a plant running a mixed OEM fleet with an older Sorg fish-tail forehearth on one line and a newer Falorni unit two lines over. The spec wasn't wrong. The assumption that every zone could hit it on day one of the programme was. We rebuilt the target as a curve, tightening zone by zone as the section timing data came in, and got there in 11 weeks instead of failing a two-week deadline three times running.

And this is the bit head office never wants to hear: the benchmark number is a lagging indicator of years of discipline, not a lever you can pull on its own.

What actually transfers: the habit, not the headline

Three things travelled cleanly from my time at O-I into every plant I've audited since. Job change accountability sits with a named role, not a shift. Escalation timing gets measured in minutes, not shifts. And every mould set has a written standard the night-shift operator can follow without phoning the day-shift superintendent at home.

• Recipe lock owned by the hot-end superintendent, with operator sign-off required for any set-point change
• Mould preheat curve documented to a target, not left to "the way we've always done it"
• First-ware inspection criteria written down, not carried in one QA lead's head

The 0600 handover is where this usually falls apart. On most lines I audit, the night-shift swabbing data doesn't make it into the morning meeting close to 70% of the time, which leaves the day-shift hot-end superintendent running blind on a variable that drives stones and blisters on the next job change (and I've had superintendents insist their handover process was solid right up until we pulled the log and proved otherwise). That's not a training problem. It's a systems problem, and it's exactly what a management audit is built to surface, because nobody inside the plant flags their own handover gap unfiltered.

Benchmarking tells you where the fence is. It doesn't tell you why your plant can't get over it yet.

Why an independent read matters more than a vendor one

On a two-furnace, five-line plant I audited in 2018, the day shift was certain their job changes ran to target. Nobody had pulled the changeover data in six months. When we did, the average overrun was 23 minutes a change, across four changes a week. 23 minutes. That was the real number, not the guess everyone had been reporting up the chain.

This is also where the Job Change Tool came out of, not as a rebadge of generic SMED thinking, but as the systemised version of what I watched work for real on real shifts at O-I and since. Nine stages, named owners, KPI tracking that doesn't disappear into a shift log nobody reopens. Plants running it properly have cut cross-shift job change variance by -42% within two quarters, and the gain holds because the discipline is built into the process, not into one good superintendent's memory.

An OEM will benchmark you against their own installed base, which is useful and also self-interested. That's not a criticism, it's just what OEMs do. A vendor-neutral container glass consultant benchmarks against the discipline pattern instead, because that pattern is what actually moved the needle at O-I, at the Arglass Yamamura greenfield we built in Valdosta from 2019 to 2022, and at every independent US and GCC plant I've audited since. It's not glamorous work. It's the work that turns a one-off benchmarking exercise into a plant that holds its numbers six months later, when the consultant isn't in the building anymore.

If your board is asking why your numbers don't match a Fortune 500 competitor's investor deck, that's a fair question. Just make sure whoever's answering it is telling you what to build, not what to buy.