How we cut customer PPM by 60% in 12 months — and what we learned
Twelve months ago our 12-month rolling PPM (parts per million defective) sat at 1,030. Today it's 405 — and falling. Here's a candid account of what worked, what didn't, and what we'd do differently.
PPM is a blunt instrument. Customers love it because it's one number; quality engineers hate it for the same reason. But it does measure something real: the rate at which defects escape your process and reach the customer. In early 2025 ours was uncomfortably high, and the root cause was hiding in plain sight.
What we changed first (and why it didn't work)
The instinct was to add inspection. We doubled FQC sampling, added a second visual gate, and put a CMM operator on dedicated evening shift. PPM went down — but only briefly. After three months it crept back up because inspection doesn't fix process drift; it just delays the discovery of it.
The actual root cause: process variation, not inspection gaps
We took a fresh look at the data. Of the last 100 customer escapes, 78 were dimensional non-conformances on features that were on the print but not on our SPC plan. Our control was reactive: we were inspecting features that customers complained about, not features that mattered.
The fix was two changes that took six months to roll out:
- CTQ-driven SPC. Every critical-to-quality feature in the customer drawing now gets its own SPC chart in MES, refreshed every 2 hours of production. Operators see drift in real time.
- Pre-emptive die maintenance. Instead of waiting for a shot weight excursion, we use a 30-shot rolling Cpk on every cavity. When Cpk approaches 1.33 from above, we pull the die for maintenance — before a part ships out of spec.
What we learned about people, not process
The biggest unlock wasn't the SPC dashboards — it was getting first-line supervisors to trust them. For the first three months operators kept "fixing" excursions by adjusting parameters in micro-steps; the chart looked OK but the underlying issue (a worn ejector pin, in one case) accumulated until catastrophic.
We had to retrain supervisors that a drift alert is a stop-the-line event, not an adjust-the-process event. That cultural change took 90 days and several uncomfortable shift meetings.
"Quality is a discipline of stopping, not adjusting. Adjusting hides the problem. Stopping fixes it." — Quote we ended up putting on the line
The numbers, twelve months later
- Rolling PPM: 1,030 → 405 (−61%)
- On-time delivery: 96.2% → 98.6%
- Customer complaints (CR): 47/year → 12/year
- FQC inspection cost as % of revenue: actually rose 0.4 pts (we kept the extra sampling — it's now a verification, not a backstop)
What we'd do differently
We'd start with CTQ alignment, not inspection density. The first 90 days of effort were largely wasted. Customers who measure suppliers on PPM alone will get inspection-heavy suppliers; customers who measure on first-pass yield drive better long-term behaviour.
Our next target is sub-200 PPM by end of 2026, driven primarily by deeper PFMEA integration with our two largest accounts. We'll write another post when we get there — or fail trying.
