Gary is made up, but most plant managers have stood exactly where he’s standing. He runs a plastics injection moulding plant outside Dayton, Ohio, with fourteen presses on three shifts. For years his OEE number came from a clipboard. Each operator wrote down downtime at the end of the shift, a supervisor typed it into a spreadsheet, and on Monday morning Gary looked at last week’s figure and tried to remember what went wrong on Tuesday night.
The number usually came out around 70%. He suspected the real figure was lower. Nobody writes down a four-minute jam when a shift ends in ten minutes.
That suspicion is why so many plants have moved to automated oee tracking. Machine data comes straight from controllers and sensors instead of memory, and AI tools are now starting to read that data for patterns people miss.
What OEE actually measures
Overall Equipment Effectiveness tells you how much of your planned production time was truly productive. A score of 100% would mean making only good parts, at full speed, with zero downtime. No real plant gets there.
It breaks into three pieces:
Availability compares how long the machine actually ran with how long it was scheduled to run. Breakdowns, changeovers, and waiting for material all eat into it.
Performance compares the actual cycle speed with the ideal one. A press that runs, but slowly, loses points here.
Quality is the share of good parts out of everything you started. Scrap and rework count against you.
Multiply them together, and you get OEE. Say Gary’s press 7 was available 90% of planned time, ran at 95% of ideal speed, and produced 98% good parts. Its OEE is 0.90 × 0.95 × 0.98, or about 83.8%. Around 85% is often quoted as world-class for discrete manufacturing, but the trend on your own machines matters far more than any benchmark.
Why the clipboard number was wrong
Manual tracking has a familiar set of problems. Operators round numbers, short stops go unrecorded, and by the time anyone looks at the data, the shift that caused the trouble has gone home.
When Gary’s plant connected its presses to an automated system pulling signals from the PLCs, the first week’s OEE came in at 61%. Nothing had got worse overnight. The software was simply counting the dozens of two- and three-minute micro-stoppages that had never made it onto paper.
That number stung, but it was finally honest.
Seeing downtime while it’s still happening
A live dashboard changes the conversation on the floor. When press 11 stops, everyone can see it straight away, along with how long it’s been down. Waiting until Monday is no longer the default.
Gary’s team found that one press was losing almost forty minutes a shift waiting for resin. The dryer feeding it was undersized. On the clipboard, that had shown up as “material” with no detail. With real-time data, the pattern was obvious within a few days.
The Six Big Losses, with numbers attached
Lean manufacturing groups production losses into six familiar buckets: equipment breakdowns, setup and changeover time, small stops, reduced speed, startup rejects, and production rejects. Everyone in manufacturing knows the list. The hard part is knowing how much each one costs on your floor.
Automated tracking puts a figure on each bucket. At Gary’s plant, small stops turned out to be the biggest loss on four presses, which surprised his maintenance lead, who’d been focused on breakdowns. Mould changeovers were second. So the next quarter’s improvement projects went there instead of where everyone had assumed.
Quality losses can be harder to catch in real time. Some plants now add camera inspection at the end of the line, much like AI vision for quality control on filling lines, so rejects feed straight into the quality figure.
Where AI comes in
Most modern OEE systems collect far more data than a person can read. That’s where machine learning is starting to earn its place.
The most practical use is predictive maintenance. Models watch for patterns in cycle times, temperatures or fault codes that tend to come before a breakdown, and they flag the machine before it fails. Nexelem’s platform, for example, mentions predictive maintenance alerts alongside its OEE monitoring and AI-assisted production scheduling.
Other tools spot unusual behaviour. They’ll notice a press that has slowly lost two seconds per cycle over three weeks, a drift nobody would see on a daily chart. Some systems also suggest a likely downtime reason from the machine signals, so operators only have to confirm it instead of picking from a long list.
Gary’s rule is that AI flags and people decide. A predicted bearing failure gets a maintenance check, not an automatic shutdown. A suggested reason code still gets a tap from the operator, who knows that the “fault” was actually a planned colour change.
Getting it running without a revolt on the floor
Gary learned a few things the hard way.
Start with the machines, not the dashboards
Connecting presses directly to the system meant operators didn’t have extra data entry to do. If tracking becomes another chore, people resent it and work around it.
Agree on reason codes before going live
His first rollout had “Other” picked for 40% of downtime. He sat down with supervisors from all three shifts and agreed on twelve clear downtime codes and eight reject codes, with plain-English definitions. “Other” dropped to under 5% within a month.
Put the screen where operators can see it
A big display at each cell showed the press’s live status and that shift’s OEE. Operators started fixing small stops on their own, mostly out of pride, because nobody wanted their press to be the red one.
Use the data for maintenance, not blame
He made a point of never using OEE numbers to discipline individual operators. Once people think the system is there to catch them out, the data quietly gets worse.
Six months in, the plant’s OEE sat at 74%, up from that brutal first reading of 61%. Gary still has the old clipboard in his desk drawer. Last week, the new night supervisor asked what it was for.
Related: 7 Best LLM Observability Tools in 2027: Which One Should You Choose?
