Manufacturing KPIs are the small set of numbers a plant reviews on a fixed rhythm to run the business: safety, quality, delivery, and cost, in that order. A good plant scorecard fits on one page, roughly twelve KPIs, and pairs lagging results with the leading indicators that predict them.
Most plants do not suffer from too few metrics. They suffer from three hundred of them, spread across reports nobody reconciles, with no agreement on which twelve matter. The fix is old and unglamorous: one page, four groups, reviewed daily at the line and weekly at the plant. This post gives a 12-KPI scorecard you can adapt, the leading-versus-lagging logic that makes it predictive, and a short field guide to the vanity metrics worth deleting.
How should a plant scorecard be organized?
Organize it as SQDC, Safety, Quality, Delivery, Cost, in that priority order. The sequence is deliberate: it encodes that you will not buy delivery with injuries or cost with defects, and it gives every meeting the same spine, from the shift huddle to the monthly review. Some plants add People/Morale as a fifth column (SQDCP); the logic is unchanged.
Here is a working 12-KPI scorecard, three per group:
- Safety, TRIR (total recordable incident rate): the lagging outcome. For context, U.S. manufacturing recorded 2.8 recordable cases per 100 full-time workers in 2023 per the Bureau of Labor Statistics.
- Safety, near-misses reported: the leading counterpart. You want this number healthy and high; silence is not safety.
- Safety, safety actions closed on time: proves the reports change something.
- Quality, first pass yield: right the first time, no rework laundering.
- Quality, customer complaints / returns: the lagging voice of the market.
- Quality, cost of quality: scrap, rework, and failure cost in dollars. The American Society for Quality notes quality-related costs commonly run 15–20% of sales revenue, reaching 40% at poor performers reason enough for a seat on page one.
- Delivery, on-time-in-full (OTIF): did the customer get what was promised, complete, on the date.
- Delivery, schedule attainment: did each line produce to plan; the leading indicator OTIF follows.
- Delivery, OEE at the constraint: capacity health where it decides throughput; run it through the OEE calculator to keep the method honest.
- Cost, unplanned downtime hours: the most controllable cost driver on most floors, and a leading indicator for three other lines on this card.
- Cost, scrap dollars: waste in the units finance believes.
- Cost, labor productivity (units or value per labor hour): the trend line that funds everything else.
Twelve is a ceiling, not a target. If a KPI has no named owner, no target, and no action when it turns red, it is decoration, cut it.
What is the difference between leading and lagging KPIs?
Lagging KPIs report outcomes after they happen; leading KPIs measure the behaviors and conditions that produce those outcomes while there is still time to act. TRIR is lagging; near-miss reporting and safety actions closed are leading. OTIF is lagging; schedule attainment and constraint OEE lead it. Scrap dollars lag; first-pass yield and process checks lead.
A scorecard of pure lagging metrics is a rear-view mirror, accurate and useless for steering. A scorecard of pure leading metrics is a dashboard of effort with no proof it works. The pairing is the point: every lagging result on the page should have at least one leading driver next to it, so when the result turns red the conversation moves in one step from "what happened" to "which driver failed." This is the same pairing logic as maintenance KPIs where PM completion (leading) sits beside unplanned downtime and MTBF (lagging).
How do KPIs cascade from plant to line to shift?
Each level keeps the same four SQDC groups but swaps in the version of the number it can actually move within its time horizon. The plant manager reviews OTIF monthly; the line reviews schedule attainment daily; the shift reviews "units by hour 2 vs. plan" right now. Same family, three altitudes.
The cascade only works if the numbers reconcile, the shift's units must sum to the line's attainment, and the line's attainment must explain the plant's OTIF. That is a data plumbing problem as much as a management one: when each level pulls from a different spreadsheet, every meeting starts with an argument about whose number is right. One shared source, feeding every altitude the same figure, is the quiet prerequisite, it is exactly what CLS got by replacing paper production logging with real-time data and automated daily reporting (the CLS case study).
How do you roll a scorecard out without drowning in it?
Start manual, start at one line, and let the board earn its automation. A four-week pattern that works: week one, agree the twelve KPIs and owners with the leadership team, expect this to be the hardest week. Week two, stand up one line's SQDC board with marker-on-whiteboard data and run the daily ten-minute huddle. Weeks three and four, fix the definitions that turn out to be ambiguous (they will), then extend to the next line.
Only automate what the huddle already uses. Automating a board nobody reviews produces a very current board nobody reviews. But once the rhythm holds, automation matters in the other direction: hand-collating twelve numbers across shifts burns an hour of supervisor time per day, and hand-collated numbers are the ones that quietly diverge between meetings. That collation step, floor logs to live scorecard, same figure at every altitude, is exactly the layer worth digitizing first.
Which metrics are vanity metrics?
Any number that always looks good, has no owner, or moves no decision. The usual suspects:
- Plant-wide average OEE. Averaging unlike lines produces a number nobody can act on; keep OEE per line, at the constraint.
- Cumulative anything. "Units produced since January" only goes up. Rates and ratios against targets, always.
- Percent metrics with soft denominators. Efficiency against a padded standard runs at 105% forever. If a metric hasn't been red in a year, the target, not the performance, is the anomaly.
- Activity counts posing as outcomes. Audits performed, meetings held, reports issued. Effort matters only where it pairs with the result it is supposed to drive.
- Days since last incident. It punishes reporting and rewards silence; near-miss reporting rates carry the same intent without the perverse incentive.
The one-page test catches most of these: if the scorecard must fit on one page, every vanity metric is displacing a real one, and the fight over the twelve slots is precisely the strategy conversation most plants never have.