Implementation Guide
Manufacturing KPIs and Metrics.
Manufacturing KPIs are the measured numbers a plant runs itself by, usually grouped as safety, quality, delivery, inventory and productivity (SQDIP). The KPIs that matter share three properties: they are measured at the process, one named person owns each number, and each number informs a decision in a regular review. Measure a few numbers that someone owns, not many numbers that nobody owns. A KPI without an owner and a decision is reporting, not management.
What manufacturing KPIs are for
A manufacturing KPI exists to inform a decision. That is the whole job. OTIF at 87 percent is not information until someone owns the number, knows why it is 87 and not 94, and uses it in a review to decide where the next countermeasure goes. Everything else that gets measured, and plants measure a great deal, is a metric: potentially useful, occasionally consulted, but not the basis on which the plant is run.
The useful mental model is a small hierarchy. At the top sit five to eight plant-level KPIs, conventionally grouped as safety, quality, delivery, inventory and productivity, the SQDIP structure that also organizes daily boards. Below them, each area or line carries its own three to five numbers that its shift can actually influence. Below that live all the metrics the systems record anyway. The failure mode is flattening this hierarchy into one 40-line scorecard where everything is equally important, which means nothing is.
Scope note: this page defines the core KPIs, gives their formulas, and shows how to select a starting set and baseline it. The SQDIP guide covers the daily board built from these numbers, the OEE guide goes deep on equipment effectiveness, and the KPI tree guide covers cascading plant KPIs down to shift level.
Where KPIs sit in the transformation roadmap
On the TeamGuru deployment roadmap this topic is the Baseline KPIs practice in the Diagnose stage. It follows the current-state assessment, which names the business problem the KPIs must serve, and runs in parallel with value stream mapping, which supplies measured numbers like lead time, WIP and changeover. The baseline then feeds strategy deployment, which sets targets on it, and daily management, which reviews it every day.
- Before Current-State Assessment
- You are here Baseline KPIs
- In parallel Value Stream Mapping
- After Strategy Deployment
- After Daily Management
When more KPIs make things worse
Measurement feels like progress, which is exactly why it gets overdone. Every number added to a scorecard costs attention, and attention is the scarcest resource in a plant. Before adding a KPI, check the list below; most struggling measurement systems fail in one of these four ways, and none of them is fixed by measuring more.
- The 40-metric scorecard. Nobody can name the owner of line 23, reviews skim instead of decide, and the three numbers that matter drown in the 37 that do not. If the monthly review cannot discuss every KPI, the list is too long, not the meeting too short.
- Averages that hide variation. A monthly OTIF of 91 percent can contain one perfect week and one disastrous one, and the disaster is where the learning is. Review trends and distributions at the working level; reserve single averaged numbers for summaries.
- Vanity metrics. Units produced looks great on a wall and rewards overproduction, the worst of the seven wastes. Any output number without a demand, quality or time denominator flatters effort instead of measuring performance.
- KPIs nobody can influence on shift level. Inventory turns on a production cell's board is noise: the team cannot move it from their station. Cascade instead: the cell tracks schedule adherence and FPY, and those roll up into the turns the plant manager owns.
The definitions table: core KPIs by SQDIP category
These 17 KPIs cover what most plants genuinely need to choose from. Formulas are written in plain language on purpose: if a number cannot be explained at the board in one line, it will not survive on the shop floor. No plant should track all 17. Pick a starting set with the rules in the next section.
| KPI | Definition | Formula | Typical use |
|---|---|---|---|
| Safety | |||
| TRIR | Total recordable incident rate: recordable injuries and illnesses normalized to 100 full-time workers per year. The standard lagging safety measure. | TRIR = (recordable incidents x 200,000) / total hours worked | Comparing safety performance across sites and over time. The 200,000 constant is 100 workers times 2,000 hours. |
| Near-miss reporting rate | How many near misses and unsafe conditions people report per period. A leading indicator: a rising count usually means growing trust, not growing danger. | Near misses reported per month (often per 100 employees) | Judging whether the safety system finds problems before they become injuries. |
| Quality | |||
| First pass yield (FPY) | Share of units that pass one process step correctly the first time, with no rework, repair or scrap. | FPY = units passing first time / units started | The most honest single quality number for one process step, because it counts rework that scrap rate hides. |
| Rolled throughput yield (RTY) | Probability that a unit passes every step of the whole stream first time: the product of all step FPYs. | RTY = FPY step 1 x FPY step 2 x ... x FPY step n | Exposing the hidden factory. Five steps at 97 percent each give an RTY of only 86 percent. |
| Scrap rate | Share of started units (or production cost) lost as scrap. | Scrap rate = scrapped units / units started | Sizing the material loss. Pair it with FPY, which also captures the rework scrap rate misses. |
| PPM defect rate | Defective parts per million units delivered to the customer. | PPM = (defective units / units delivered) x 1,000,000 | Customer-facing quality, the number that appears on supplier scorecards. |
| Cost of poor quality (COPQ) | All money spent because things are not right first time: scrap, rework, warranty, returns and excess inspection. | COPQ % = (scrap + rework + warranty + inspection cost) / sales | Translating quality into money for prioritization and the monthly management review. |
| Delivery | |||
| On-time delivery (OTD / OTIF) | Share of customer orders delivered on the promised date. OTIF adds: in full and correct. | OTIF = orders delivered on time and in full / total orders | The customer's view of delivery. Measure against the first promise date, not the latest reschedule. |
| Schedule adherence | How closely production followed the plan for a given interval, in quantity and sequence. | Schedule adherence = units produced to plan / units planned | Separating delivery misses caused by planning from misses caused by execution. |
| Lead time | Elapsed time from order to delivery, or from material entering the plant to leaving the dock. Queues included. | Lead time = delivery date minus order date (or dock-to-dock days) | The stream-level number a value stream map measures. Almost always dominated by waiting, not processing. |
| Inventory | |||
| Inventory turns | How many times per year the inventory is consumed and replaced. | Turns = cost of goods sold / average inventory value | The finance-level view of how much cash the operation ties up. |
| Days on hand (DOH) | How many days current inventory would last at current consumption. | DOH = 365 / inventory turns (or on-hand quantity / daily usage) | The same fact as turns in operational units. Works per material, per stage and per warehouse. |
| Work in process (WIP) | Units sitting between raw material and finished goods, best expressed in days of demand. | WIP days = WIP units / daily demand | The direct driver of lead time. Capping WIP is the fastest lead time lever a plant has. |
| Productivity | |||
| OEE | Overall equipment effectiveness: how much of scheduled equipment time produces good parts at the ideal rate. | OEE = availability x performance x quality | Finding the biggest of the six losses on constraint equipment. See the OEE guide for the full calculation. |
| Labor productivity | Good output per hour of direct labor. | Labor productivity = good units produced / direct labor hours | Trending the effect of improvements. Risky as a shift-level target, because it rewards overproduction. |
| Changeover time | Time from the last good unit of product A to the first good unit of product B at full rate. | Changeover = first good B unit time minus last good A unit time | The lever behind batch sizes and flexibility, and the subject of SMED. |
| Unplanned downtime | Scheduled production time lost to unplanned stops. | Downtime % = unplanned stop time / scheduled production time | Feeding maintenance priorities. Record stop reasons at the machine, not in a monthly report. |
Two of these deserve their own pages and have them: OEE, where the three factors and the six losses need a full walkthrough, and changeover time, where the reduction method is SMED. Lead time and WIP are measured, not estimated, during a value stream mapping week, which is why the two practices run in parallel.
How to choose your KPIs: six selection rules
Selection is where measurement systems are won or lost, and it is a design activity, not a collection activity. Apply these rules in order to the table above and a plant-level set of five to eight numbers falls out almost mechanically.
Start from the constraint and the business problem
The KPI list is derived, not collected. If delivery is the business problem and machining is the constraint, then OTIF, machining OEE and changeover time belong on the list, and a dozen plausible others do not. A KPI earns its place by connecting to what limits the plant today.
Pair every lagging KPI with a leading one
Lagging numbers (OTIF, TRIR, COPQ) tell you how the month went. Leading numbers (schedule adherence, near-miss reports, first pass yield at the constraint) tell you how the month is going while you can still act. A scorecard of only lagging KPIs is a rearview mirror.
One named owner per KPI
An owner is the single person who explains the number, knows why it moved and brings the countermeasure. A KPI owned by a department is owned by nobody. One person can own several KPIs; no KPI can have two owners.
Baseline before target
Measure for two to four weeks before setting any target. Targets set before the baseline exists are negotiated fiction, and people learn to game the measurement instead of improving the process.
Measure at the process, not from reports
The number on the board must be countable where the work happens: units, stops, minutes, defects. ERP extracts and month-end reconciliations come too late and get argued with. A number the operators helped count is a number they accept.
Every KPI needs a review where someone reads it
Name the meeting where each KPI is read and the decision it informs: the daily tier board, the weekly performance review or the monthly management review. If you cannot name the meeting, the KPI is decoration and should be cut.
Building a baseline people accept
A baseline is two to four weeks of honest measurement before any target exists. The sequence matters. First, define each KPI in writing: the formula, the data source, the counting rules and the edge cases, because the first argument at any board is always about the definition, not the performance. Second, measure at the process, with the people who run it counting or at least seeing the counts. Third, have each owner accept the number: not like it, accept it as true. Only then set targets.
Expect the baseline to be worse than everyone believed. Reported numbers drift upward over years of quiet definitional generosity, and measuring at the process removes the generosity. An OTIF that drops from a reported 95 to a measured 87 has not gotten worse; it has become real, and real numbers are the only ones that can improve.
Worked example: picking six starting KPIs
The numbers below are illustrative but internally consistent, for the same 450-person components manufacturer used across the transformation roadmap: 456 units per day of demand across two shifts, delivery named as the business problem, machining identified as the constraint, and a mapping week that found 8,400 pieces of WIP and a 47-minute changeover. The site chose six starting KPIs, and every choice traces back to a selection rule:
| KPI | Measured baseline | Owner | Why this one |
|---|---|---|---|
| TRIR | 2.4 (trailing 12 months) | Plant manager | Safety leads every board and every review. The plant manager owns it personally so the message is unambiguous. |
| First pass yield, assembly | 94.2% | Assembly area leader | Assembly rework was destabilizing the schedule daily. FPY at the final step catches the whole stream's quality. |
| OTIF | 87% (first promise date) | Planning manager | Delivery is the business problem the current-state assessment named. Measured against first promise, so replanning cannot improve it. |
| WIP, in days of demand | 18.4 days (8,400 pcs at 456/day) | Value stream manager | The mapping team showed WIP is the lead time. One number stands in for both. |
| OEE, machining line | 61% | Machining area leader | Machining is the constraint. OEE anywhere else would be a distraction; OEE here is capacity. |
| Changeover time, machining | 47 min average | Machining area leader | The leading partner to WIP and OTIF: the mapping week showed the 47-minute changeover forces the batches that create the 18.4 days. |
Notice what is absent. No inventory turns, because WIP days carries the same signal at a level the value stream manager can act on. No labor productivity, because with delivery as the problem it would pull attention toward running machines flat out, which is how the 8,400 pieces of WIP happened in the first place. And no targets yet: those arrive through strategy deployment after four weeks of baseline data, so the first targets are set against reality instead of aspiration.
From numbers to management
A baselined KPI set is an instrument panel, not a management system. Three practices turn it into one. First, cascade: a KPI tree connects each plant number to the area and shift numbers that drive it, so a shift improving its schedule adherence knows it is moving the plant's OTIF. Second, rhythm: the shift-level numbers go onto SQDIP boards reviewed in the daily management cycle, where a red number becomes an action with a name and a date within 24 hours. Third, escalation upward: weekly and monthly reviews work the trends the daily cycle cannot fix alone.
This is also the point where spreadsheets give out: definitions fork into local copies, the baseline gets retyped, and the review runs on numbers nobody fully trusts again. In TeamGuru, each KPI lives once, with its definition, owner, baseline and history, through KPI management, and the boards and reviews read from that single record through reporting, so the daily board, the monthly review and the plant manager's phone all show the same number.
On the roadmap, the next moves are strategy deployment, which sets targets on the baseline and turns the biggest gaps into the plant's few owned priorities, and daily management, which puts the numbers to work every morning.
