Startup losses are the scrap and slow output a process produces after every start, while it warms up, purges, and settles into stable running. They are the fifth of the six big losses, counted mainly as a Quality loss, and their defining trait is that they repeat: every stop and restart buys another dose of them.
Startup loss is the loss that hides inside "we just got it running." A line comes up after a changeover or a break, the first parts are off-spec or the rate is soft while temperatures climb, and by the time it is stable nobody counts what the ramp cost. On a plant running long, uninterrupted campaigns, that cost is trivial. On a plant that stops and starts all day, it is one of the largest losses on the floor and almost none of it is logged.
What are startup losses?
Startup losses are the reduced yield and reduced rate between the moment a machine starts and the moment it reaches stable, in-spec running. They are the fifth of the six big losses from Nakajima's TPM work, classed primarily as a Quality loss because the visible cost is startup rejects, the warm-up scrap, purge material, and first off-spec pieces made before the process locks in. There is a second, quieter piece: while the process ramps, it often runs slow too, so a slice of startup loss also lands in Performance.
The distinction from ordinary first pass yield loss is timing. Production rejects, the sixth loss, happen during steady-state running and point to a process that drifts out of control. Startup rejects happen predictably at the beginning of every run and point to a ramp that has not been engineered. Same scrap bin, completely different root cause, completely different fix. Telling them apart is the first job: sort your rejects by where in the run they occurred, and the startup share usually surprises people.
Where do startup losses come from?
From physical states that take time to reach, not from carelessness. The machine has to get somewhere, a temperature, a pressure, a clean material path, a settled mechanical position, before it makes good parts at rate. The table breaks down the usual sources and what shortens each one.
| Source of startup loss | What is happening | What shortens it |
|---|---|---|
| Thermal warm-up | Barrels, molds, ovens climbing to temperature | Preheat during external setup; hold heat between runs |
| Material purge | Flushing old color, resin, or product from the path | Sequence runs to minimize purge; standard purge volume |
| First-article settling | Dialing in before the first good piece | Documented settings sheet, not dial-in from memory |
| Mechanical settling | Tension, alignment, and feeds finding steady state | Standard start sequence; verify at fixed cycle counts |
| Slow ramp by policy | Deliberately running soft "until it looks right" | Define stable in cycles, not by feel |
Why do frequent short runs make startup loss balloon?
Because startup loss is a fixed cost per start, so total startup loss is the per-start cost multiplied by the number of starts. Run one long campaign and you pay it once. Split the same volume into ten short runs and you pay it ten times, same warm-up, same purge, same first-article scrap, ten times over. This is why startup loss and batch size are the same conversation, and why a plant chasing flexibility with tiny lot sizes can quietly drown in ramp scrap it never budgeted for.
The trap is that the two costs pull in opposite directions. Big batches crush startup loss but bloat inventory and lengthen lead time; small batches do the reverse. The escape is not to pick a side, it is to make each start cheaper, so small runs stop being punished. That is exactly what changeover work buys you: cut the warm-up and first-article scrap per start with SMED and disciplined setup time reduction and the whole trade-off shifts in your favor.
Put a number on it and the point lands. Suppose each start burns 40 pieces of warm-up scrap and 15 minutes of soft ramp. Run four campaigns a week and that is 160 scrap pieces and an hour of slow output, a rounding error. Chase flexibility with twenty short runs a week and the identical process throws away 800 pieces and five hours, week after week, with nothing in the downtime log to explain where it went. Nothing about the machine changed; only the number of starts did. That is the whole reason startup loss deserves its own line rather than being buried in a general scrap rate, the same physics costs six times as much on a short-run schedule.
How do you reduce startup losses?
Attack the per-start cost first, then the number of starts. In order:
- Separate startup rejects from production rejects. Sort scrap by where in the run it happened. You cannot reduce a loss you have folded into general first-pass-yield numbers, and the startup share is usually bigger than anyone guesses.
- Move warm-up off the clock. Preheat barrels, molds, and ovens during the previous run as external setup, and hold heat between short runs instead of letting equipment cool. The warm-up still happens; it just stops eating production time.
- Replace dial-in with documented settings. A settings sheet per product, temperatures, speeds, pressures, stops, turns the first-article settle from a guessing game into a lookup, so the first piece is right instead of the thirtieth.
- Define "stable" in cycles, not by feel. Specify how many cycles a good startup takes and verify at that count. "Run it soft until it looks right" is a slow ramp with no end condition, and it hides real startup loss behind operator judgment.
- Make each start cheaper with SMED. Every minute and every scrap piece you remove from a changeover is startup loss you stop paying on every future run. This is where setup work and startup work become the same project.
- Then cut the number of starts. Only after starts are cheap, look at batch sizing and scheduling to reduce unnecessary restarts, sequencing similar products together, avoiding needless mid-run stops. Doing this before making starts cheap just trades one loss for inventory.
Where startup loss sits in the standard framework:
- Startup rejects are formally the fifth of the six big losses and are classified as a Quality loss defects produced from startup until stable production is reached, including warm-up and first-article scrap. The taxonomy is documented in the six big losses reference which separates startup rejects from steady-state production rejects.
- In the OEE math, those rejects flow through the Quality factor, good count divided by total count, while the slow-ramp portion lands in Performance. The factor definitions are set out in the OEE calculation reference. Because OEE multiplies its three factors, trimming startup scrap raises the score directly.
How do startup losses show up in OEE?
Mostly through Quality, with a tail in Performance. The warm-up scrap and first-article rejects reduce the good-count ratio; the soft ramp reduces the rate. That split is why startup loss is easy to miss: half of it looks like a quality problem and half looks like a speed problem, and neither looks like a startup problem unless you have tagged the rejects by run position. Pull them together and the fifth loss finally becomes visible on your OEE breakdown instead of being smeared across two factors. The practical test is simple: if your reject records cannot tell you what share of scrap came from the first N cycles of each run, your OEE is quietly crediting startup loss to two other causes, and any countermeasure aimed at steady-state quality will keep missing it.
The size of the prize depends entirely on how often you start. A campaign plant should spend its hours on speed loss or breakdowns and leave startup loss alone. A high-mix plant restarting a dozen times a shift may find startup loss rivaling its biggest logged category once it is measured, and it competes for improvement time with everything else, so it belongs on the same Pareto tracked through your manufacturing KPIs. The measurement is the hard part: startup loss is invisible until counts and reject positions are captured automatically off the line. Plants that stream that data the way Harmony does (see the platform) can finally separate the ramp from steady state, and you can price what closing it is worth with the OEE calculator. Until then, "we just got it running" stays a cost nobody puts a number on, and the crew on the shift after every single changeover keeps paying it in the dark.