Power BI for container glass: what 13 sites taught us

The boardroom dashboard problem

Ask most plant managers what their Power BI dashboard is for and you'll get the same answer: it's for the monthly steering committee, the slide the VP of Operations pulls up to show the board the plant is under control. That's backwards. If a dashboard only gets opened once a month, it isn't running your plant. It's decorating a meeting.

I've watched this pattern repeat almost identically across a 13-site rollout spanning the Gulf and Southeast Asia. A vendor or an internal BI team builds a sharp landing page, an OEE gauge, a scrap-rate trend, maybe a cullet-ratio tile. It looks good in the review deck. Nobody on the floor touches it between reviews, because none of the numbers tell an operator what to do at 0600 on a Tuesday.

The fix isn't a better chart. It's deciding, station by station, what the hot end needs to see every shift versus what belongs in a quarterly pack.

What the daily screen should actually capture

Start with pack-to-melt: good packed ware divided by total glass melted. Most sites report OEE and scrap rate as separate lines and never reconcile them, which hides exactly where the tonnes are going missing. A pack-to-melt view above 90% is the one number that ties hot-end yield and cold-end reject together, and it belongs on the same screen as changeover time, not buried three tabs deep in a monthly export.

A handful of other metrics earn a permanent place on the shift screen:

• Cavity-to-cavity weight spread against a ±2g target, not just the average fill weight the OEM dashboard reports
• Job-change duration trended against a sub-45-minute target, with 90-plus-minute outliers flagged by section
• Cold-end reject rate under 3%, tracked separately from hot-end job-cull so a conveyor fault doesn't get blamed on the forming machine
• Cullet ratio plotted against furnace specific energy consumption, since every 10-percentage-point rise in cullet typically cuts specific energy by 2-3%
• Blister, seed, cord and stone counts as separate series, never rolled into one generic defects tile

That last one matters more than people think. Blister and seed trace back to dissolved-gas reboil, usually a refiner temperature issue. Cord points at batch inhomogeneity. Stone is refractory or batch particulate. Roll them into one line and you've deleted the root-cause signal before an operator even sees it.

What to leave off the screen

The instinct with any new BI build is to add more tiles. Resist it. A screen with forty metrics gets glanced at once and ignored forever, because nobody can tell which number is the one that matters this shift.

Cut anything that's already averaged past the point of being useful. A blended energy-per-tonne figure that mixes furnace age, cullet quality and pull rate together tells you nothing you can act on. Same with an average fill weight that hides a two-cavity spread wide enough to be rejecting ware nobody's caught yet.

And cut anything that isn't normalised for where the furnace sits in its campaign. A furnace in year one after a cold repair behaves nothing like one in year 12 on pull-rate stability or defect rate. I've seen dashboards trend defect rate across a full 10-15 year campaign as if the refractory never ages. That's not analysis. That's a chart that happens to have data in it.

A screen built for the boardroom teaches the floor to ignore it. A screen built for the floor teaches the boardroom to trust it.

Why the same dashboard template doesn't travel

In 2021 I audited a container glass plant in the GCC running a mixed OEM fleet, an eight-section double-gob Emhart line running 1980s relay logic sitting two bays down from a newer Heye SmartLine with proper on-board data capture. The energy story looked great on paper. Gulf furnaces run on subsidised industrial gas priced well under $4/MMBtu for years, against European TTF spikes into the $30-40/MMBtu range during 2022-23. On the dashboard, specific energy consumption looked like a solved problem.

It wasn't, not fully. GCC glass packaging demand is forecast to grow at roughly 5-6% CAGR through 2030 on beverage, pharma and cosmetics packaging, according to Mordor Intelligence's regional research. That means more furnace capacity going in across the region right now. But domestic post-consumer cullet collection across the Gulf is still thin compared with Europe's closed-loop systems, so plants are blending in imported cullet, and the freight and logistics exposure on that import almost never shows up in an OEM's ROI model (ask your vendor to show you the freight-adjusted cullet cost curve, not just the energy-per-tonne headline, and watch how long it takes them to find it).

Contrast that with Europe, where FEVE reports the bloc running close to 52% average recycled content across roughly 19 million tonnes of annual container glass production, with some Belgian and Swiss sites well above 90%. A European dashboard has a completely different ceiling to defend. With EU ETS allowances trading in the €60-85/tCO2 band through 2025-26, every percentage point of cullet ratio or oxy-fuel conversion carries a carbon cost a Gulf plant doesn't yet face. Build the same dashboard template for both regions and you're optimising the wrong lever in at least one of them.

Building the daily habit, not the monthly report

None of this works without a shift structure that owns it. The hot-end superintendent owns recipe lock, full stop, the operator doesn't touch a set point without sign-off. The Gob Weight Control technician owns the cavity-spread number. The Cold-End Quality technician owns reject rate. If the dashboard doesn't map to who owns what, it's just a screen.

The 0600 handover is where most of this falls apart in practice. On most lines I've reviewed, the night shift's swabbing data doesn't make it into the handover conversation 70% of the time, because it lives in a notebook, not a system. That's exactly the gap our Job Change Tool was built to close, mapped to the 9-stage Job Change Lifecycle so mould-change and recipe-load data feeds the same Power BI model as the shift KPIs, instead of sitting in a separate spreadsheet nobody reconciles.

Twenty-three minutes. That was the real changeover-time delta we measured across three of the 13 sites once job-change data got forced into daily review instead of a monthly report. Not a machine upgrade. A visibility problem.

Start with the shift, not the slide

A daily-review dashboard isn't a technology project. It's an operating discipline with a data layer under it. If your BI build is being led by whichever OEM sold you the last inspection system, you'll end up with a dashboard tuned to sell you the next one. That's the case for bringing in a vendor-neutral container glass consultant to define what the daily screen actually needs before anyone opens Power BI Desktop. We cover this exact ground in our digitalisation and reporting work, built by people who've stood at the hot end during the shift the dashboard is meant to be helping.