Choosing plant KPIs means cutting a long wish list down to the six to eight metrics that actually change what people do on the floor. The test is not whether a number is available, but whether someone owns it, can move it, and will act on it.
Every plant can generate forty metrics without trying. Dashboards make it worse: if a number can be plotted, someone plots it, and soon the morning meeting is a wall of gauges nobody acts on. The problem is never a shortage of data. It is that attention is finite, and a scorecard with forty numbers has the same effect as a scorecard with none, because no single number carries any weight. This post is about the filter, the selection method that turns a wish list into a short list. For the layout of the short list once you have it, the SQDC scorecard in manufacturing KPIs is the companion piece; this post is how you decide what earns a spot on it.
Why do most plant scorecards have too many metrics?
Scorecards bloat because adding a metric feels safe and removing one feels risky. Every number has a sponsor who fought to get it tracked, so cutting it feels like telling that person their work does not matter. Meanwhile new metrics arrive every time a problem does: a bad quarter on returns adds three quality metrics, a safety incident adds two, and nothing ever leaves. The result is accretion, not design. A scorecard that grew this way is a museum of past fires, not a live control panel.
The cost is dilution. Guidance on performance measurement consistently lands on a small working set, on the order of three to seven KPIs per goal, because beyond that people cannot hold the picture in their heads or act on all of it at once. Even the international standard for manufacturing KPIs, ISO 22400-2, defines only thirty-four indicators for the entire domain of manufacturing operations (ISO 22400-2:2014), and no plant is meant to run all thirty-four on one board. With U.S. manufacturing capacity utilization sitting in the mid-70s percent range, most recently about 75.7 percent (Federal Reserve, G.17), the few metrics that expose idle capacity are the ones worth the board space, not the forty that merely fill it.
What makes a KPI worth keeping?
A KPI earns a place on the board when it passes four tests: it ties to a plant goal, someone owns it, the floor can actually move it, and a real decision changes when it moves. Miss any one and the metric is not a KPI, it is a statistic. The distinction matters because the daily board's only job is to drive behavior between now and the next shift. A number that fails these tests can still be useful, but it belongs in a reference report you pull when investigating, not on the wall competing for attention every morning.
The "floor can move it" test is the one most often skipped, and it is the one that kills morale fastest. If you put a metric on the board that the crew cannot influence with the levers they actually hold, you have built a weather report, not a control. People stop looking at numbers that go up and down for reasons outside their reach. Controllability is what turns a metric from a scoreboard into a steering wheel. Balance leading and lagging as well: OEE and scrap tell you what already happened, while short-interval checks and staged-material readiness tell you what is about to. A board of only lagging numbers is a rear-view mirror.
How do you run the selection filter?
Do it as a deliberate cut, with the whole list on the table, not one metric at a time in a meeting. Run this sequence:
- Dump every candidate metric onto one list. Pull from dashboards, reports, and the morning meeting. Do not judge yet. You want the full forty visible so the cut is a design choice, not an accident.
- Tag each to a plant goal or drop it. Safety, quality, delivery, cost. If a metric does not clearly serve one of the plant's stated goals, it fails the first test and comes off the board.
- Name an owner for each survivor. A single person accountable for the number, not a department. If no one will put their name on it, it will not get acted on, so it fails.
- Apply the controllability test. Ask whether the crew can move the number with levers they actually hold this shift. If it moves only for reasons outside their reach, move it to a reference report.
- Apply the decision test. For each remaining metric, name the decision that changes when it crosses a threshold. No decision, no place on the board.
- Balance leading and lagging. Check that the survivors are not all outcome metrics. Pair each key lagging number with at least one leading indicator that predicts it.
- Cut to six to eight and set targets. Force the list down to what fits on one page. Give each a target and a threshold color so the board reads at a glance, not a study.
How many KPIs should a plant track?
On the daily board that drives the shift, six to eight. That is not arbitrary; it is the number a supervisor can read in the ten seconds before a stand-up meeting and a crew can hold in their heads through the shift. You can track far more in the background for investigation and reporting, but the working set that people act on every day has to be small enough to remember. A bowling chart keeps each of those few metrics visible against target month over month without adding new ones.
Layers help here. The daily floor board carries the six to eight leading and near-real-time numbers the crew steers with. A weekly review carries a slightly broader set, including the lagging outcomes that take a week to settle. A monthly business review carries the roll-ups. The mistake is putting the monthly roll-ups on the daily board, where they cannot be acted on fast enough to matter, or the daily leading indicators on the monthly review, where their noise drowns the signal.
How do you retire a metric without a fight?
Move it, do not delete it. The reason people resist cutting a metric is fear that the information disappears. It does not have to. A number that fails the board tests goes to a reference report that anyone can pull when they need it; it just stops occupying daily attention. Framing the cut as "moving to the report shelf" rather than "killing your metric" removes most of the resistance, because the data is still there for the investigation it was actually good for.
Name vanity metrics for what they are. A vanity metric looks good and drives nothing: total units produced this year, cumulative hours run, a raw count that only ever goes up. It cannot go down, so it cannot signal a problem, so it cannot change a decision. Those are the easiest cuts once the decision test is on the table, because their sponsors cannot answer the question "what would you do differently if this number moved." Distinguish them from real outcome metrics like OEE and first-pass yield which do move both ways and do drive action.
Which metrics almost always earn a spot?
On most discrete and process lines, a small core survives the filter nearly every time because each ties to a goal, has a natural owner, is controllable, and drives a clear decision. OEE or its components sit at the center, because they connect availability, performance, and quality in one number the crew can move; the underlying six big losses tell them how. Reason-coded machine downtime earns a spot because it points directly at the next fix. First-pass yield or scrap covers quality; a safety leading indicator covers safety; on-time or schedule attainment covers delivery. That is already five or six, which is why the discipline is mostly about what you leave off, not what you add.
How do you cascade the survivors so they line up?
Cascade so that each level's board rolls up cleanly into the one above it, with no orphan metrics. The plant board sets the direction, the line board carries the same goals expressed in numbers a line can move, and the shift board carries the leading indicators a crew steers with hour by hour. When they line up, a good shift visibly adds to a good line day, which adds to a good plant month. When they do not, the floor is chasing numbers that never sum to anything leadership recognizes, and trust in the whole system erodes. The OEE calculation is a useful spine for this cascade because it decomposes the same way at every level, and you can pressure-test targets with the OEE calculator before you commit them.
This is also where getting the numbers to arrive fast enough matters. A cascaded scorecard only works if each level sees its numbers in time to act, which paper reporting rarely allows. Real-time capture on the floor, feeding the same figures up the cascade, is what lets the shift board, the line board, and the plant board tell one consistent story instead of three reconciled-after-the-fact versions. That shift from paper to live capture is what CLS built across its shops (see the CLS case study), and it is what makes a lean scorecard more than a wall decoration. See how the platform ties the floor to the numbers on the features overview. No rip-and-replace, just a short list that everyone can see and act on.