What operational due diligence finds that the data room hides

It's April 2021 and I'm walking a two-furnace flint plant outside Toledo with a private equity operating partner who's already read the CIM twice. He's got a spreadsheet open showing 92% line utilisation and a five-year EBITDA bridge that assumes the furnace runs exactly like it did last year. I've got a clipboard, and I'm asking the batch house operator when he last saw a delivery ticket matching the stated cullet percentage. He can't remember one.

That's the gap. The data room tells you what the plant reported. It doesn't tell you what the plant is actually doing, section by section, shift by shift. I've sat through enough of these walks now to know the two documents rarely agree, and the difference is usually worth seven figures.

The utilisation number in the CIM is almost always the wrong number

A job change on a typical IS (Individual Section) line, colour or mould-set, runs 4-8 hours of downtime depending on how disciplined the crew is. That matters because a plant running frequent short campaigns can post high nameplate utilisation while burning far more mould life and refractory cycling than a plant running long campaigns. The utilisation percentage in a data room is meaningless on its own. It only means something cross-checked against job-change frequency logs, and most sellers don't hand those over unless you ask twice. It's precisely the blind spot our own Job Change Tool is built to close, timestamping every changeover against the 9-stage lifecycle so a buyer isn't relying on a foreman's memory of last quarter.

I saw this play out on the Toledo walk. The plant had historically run on brewery contracts, and the model assumed that mix held flat for five years. Brewers have been shifting from bottle to can for over a decade, and that demand-mix risk doesn't show up in an industry-average utilisation curve. It shows up in customer-specific contract renewal data that OEM-affiliated valuations tend to smooth over. Four hours lost, times three changes a week, times a shrinking beer book. That's the real number, not the one on slide fourteen.

Ardagh's North American glass unit went through a balance-sheet restructuring in 2024 after years of softening beer-bottle demand, and O-I's "Fit to Win" programme has targeted roughly $150-180M USD in annualised structural savings through furnace idling and network closures, per O-I Glass's own investor disclosures. Neither of those is a secret. What's usually missing from the CIM is whether the target asset's specific customer book carries that same exposure, or whether a beverage or spirits mix insulates it.

Campaign life is the capex line most models bury

A typical end-port regenerative container furnace runs a 10-15 year campaign before it needs a full cold repair. That rebuild, on a mid-size 250-400 tonne/day furnace, costs tens of millions of dollars and takes 8-16 weeks of lost production. It's routinely the single largest capex event in a plant's life, and I still see it buried in a footnote or left out of the five-year model entirely.

But the campaign-life question is the one nobody in the room wants to answer first. Ask for the refractory inspection history, not the OEM's original campaign-life estimate. Ask for specific melting energy in GJ per tonne, because that number tells you where the furnace actually sits in its life. Efficient regenerative or oxy-fuel furnaces run roughly 3.5-4.5 GJ/tonne. Furnaces well past mid-campaign commonly drift to 5.5-7+ GJ/tonne, and that drift shows up as gas-bill variance long before anyone labels it "furnace age" in the accounts.

Cullet ratio ties straight into that number, and it's one of the more common aspirational figures I find in a data room. Each 10% increase in cullet reduces specific energy by roughly 2-3%, but the "target cullet %" a plant states is often just that, a target. Check it against batch-house delivery tickets, not the sustainability slide.

Mould inventory tells you more than the maintenance log

OEMs quote mould life in cycles, 400,000 to 1,000,000+ depending on cast-iron grade and coating. Actual plant life depends on swab interval discipline, commonly every 25-40 minutes, and cavity temperature uniformity (and if the mould shop can't tell you the cavity count on the current set without asking maintenance to check, that's your answer). A plant with inconsistent swabbing shows elevated check-defect rates, checks being the finish or neck micro-cracks that come from thermal shock at the blank mould, and higher mould spend per tonne than the OEM catalogue figure ever implies.

Regional duty cycle matters here too. Middle East demand skews toward beverage container glass, juice, water, spirits, rather than the long beer campaigns European lines are built around, so job-change frequency on GCC IS lines tends to run higher. That shortens effective mould life against OEM-quoted cycle counts even when the coating and cast-iron grade are identical. I've walked GCC plants running mixed-OEM fleets off a single mould shop and watched cavity condition vary section to section in ways a cycle-count spreadsheet never flags.

Equipment generation matters for how you read that inventory too. An old Emhart 8-section still running 1990s relay logic tells a very different mould-spend story than a recent Heye SmartH1 line with servo-controlled cooling. Same cycle count on paper, different real-world wear.

Coating drift and redox are QA risks, not machine specs

Pack-to-melt, good ware packed divided by glass pulled from the furnace, is the cleanest hot-end health number there is. Below roughly 90% it signals chronic defect loss, and each defect mode points somewhere different:

• Checks, finish or neck micro-cracks from thermal shock at the blank mould, point at mould temperature control.
• Stones, unmelted batch or refractory inclusions, point at the batch plant.
• Cord points at incomplete batch homogenisation in the furnace.
• Seeds and blisters point at the refining zone or redox chemistry.

A data room's blended reject-rate figure hides all four behind one number.

Redox state, tracked through the iron-to-sulphate ratio and visible in colour consistency, is the variable that separates a plant that can hold flint spec from one drifting amber. Not a machine problem. A batch-chemistry problem, and it's exactly why OEM-affiliated consultancies rarely flag it. The hot-end superintendent owns recipe lock on a well-run line; the operator doesn't touch a set point without sign-off. Where that discipline is loose, colour drift shows up weeks before anyone books it as a QA cost.

A data room can tell you what a furnace pulled last quarter. It can't tell you what month nine of a cold repair is going to cost you, because nobody's counted it yet.

The regulatory number in the deck is usually stale

In Europe, EU ETS Phase IV cuts free allocation via a Linear Reduction Factor of 4.3% a year from 2024 to 2027, rising to 4.4% from 2028, tightening the glass sector's benchmark every allocation period regardless of a plant's historic emissions (European Commission, DG CLIMA). CBAM's definitive regime from January 2026 covers cement, steel, aluminium, fertilisers, hydrogen and electricity. Container glass isn't on that list, under EU CBAM Regulation 2023/956, and I still see decks pricing in a carbon-import cost the regulation doesn't apply.

Currency and standards risk gets missed just as often outside Europe. Egypt's EGP devaluations in 2022, 2023 and again in 2024 repeatedly repriced imported cullet, refractory and spares contracts for USD-denominated plants, and I've seen data rooms present that exposure only in EGP nominal terms, which flatters the trend. GCC buyers face their own version of this: UAE and Jordan importers answer to GSO food-contact migration limits that track EU 84/500/EEC more closely than US FDA rules, so a plant built to a US spec can fail GCC customs inspection without a secondary QA investment nobody budgeted for.

Look, the spreadsheet says one thing and the floor says another, and that's true on every one of these walks, not just the ones with obvious problems. The plants that get priced correctly are the ones where somebody actually reconciled the two.

Our founder Zaid Hassoneh ran the floor at O-I Brisbane from 2005, made plant manager by 2019, and led the $220M USD Arglass Yamamura greenfield build in the USA. That background is why Lean Glass runs due diligence the way an operator would, section by section, not the way an equipment vendor reads a spec sheet. It's also why we stay vendor-neutral: a container glass consultant tied to an OEM has no incentive to tell you the furnace has four years of campaign life left, not ten.

If you're heading into a buy-side or sell-side process on a container glass asset, get the operational read before the number gets locked into the model. Our hot end audit is built for exactly this, and for plants weighing a bigger network decision, our strategic advisory work picks up where the audit stops.