Statistical Process Control
Statistical Process Control (SPC), invented by Walter Shewhart at Bell Labs in 1924 and operationalized by W. Edwards Deming, distinguishes COMMON-CAUSE variation (random noise inherent to a stable process) from SPECIAL-CAUSE variation (signals that something has changed).…
The trap
The dominant trap is tampering: a metric is at 3.2% when last week it was 2.8%, the leadership team panics, launches a 'task force,' and changes the process. Both readings were actually common-cause variation in a stable process — the 'task force' added new variation that made things worse. Deming's funnel experiment showed that adjusting a stable process based on each output INCREASES variance. The opposite trap is using SPC as a way to DEFEND bad performance: 'it's just within control limits, we don't need to improve.' Control limits describe what the process IS doing, not what it SHOULD do. A stable process producing 5% defects is stable AND unacceptable.
What to do
Pick one critical metric (cycle time, defect rate, response time). Plot the last 25-30 data points as a run chart. Calculate the mean and standard deviation; draw control limits at ±3σ. Now check: (1) Are any points outside the limits? Investigate THOSE specific incidents. (2) Are there runs of 8+ points on one side of the mean? That's a shift — the process changed. (3) Are there 6+ consecutive increasing or decreasing points? That's a trend. If none of those: the process is stable. To IMPROVE the level of the metric, change the system itself (different equipment, different method) — don't react to individual points.
Formula
In practice
Jack Welch's GE Six Sigma program (1995-2001) embedded SPC at industrial scale. GE's appliance plants tracked dozens of process metrics with control charts; operators were trained to distinguish common from special cause and only escalate the latter. When a sensor reading drifted outside control limits on a refrigerator compressor line in Louisville, the line was halted within minutes — root cause: a worn bearing on a CNC machine, fixed before any defective compressors shipped. SPC moved GE from end-of-line inspection (catch defects after the fact) to in-process control (prevent them in real-time). Welch credited Six Sigma — built on SPC — with $12B in cumulative savings.
Pro tips
- 01
Deming's Red Bead Experiment is the fastest way to internalize SPC: workers draw beads from a box; the proportion of red (defective) beads varies from 4 to 16 per draw — but the system (the bead box) is constant. Punishing/rewarding workers for above/below average is meaningless theater. The only way to reduce red beads is to change the box (the system).
- 02
Western Electric Rules — eight tests for non-random patterns on a control chart: any 1 point beyond ±3σ; any 2 of 3 consecutive points beyond ±2σ; any 4 of 5 beyond ±1σ; 8+ in a row on one side of the mean; etc. Modern SPC software automates these. Most teams only check rule 1 and miss the early-warning signals from the other rules.
- 03
Capability (Cpk) vs. Stability: a process can be STABLE (in control) and INCAPABLE (consistently producing defects within its natural variation). Stability is the prerequisite — you can't improve an unstable process predictably. Capability is the goal — your natural variation must fit inside customer requirements.
Myth vs reality
Myth
“SPC is only useful for high-volume manufacturing”
Reality
SPC works on any time-ordered metric: ER wait times, software incident counts, ad CTR, weekly sales, app crash rates. The math doesn't care if you're making widgets or measuring user behavior. Etsy uses SPC-style anomaly detection on every key metric to avoid tampering on noisy data.
Myth
“If a point is within control limits, the process is fine”
Reality
Control limits show what the process IS doing — they say nothing about what's acceptable to customers. A process can be stable inside control limits at a 5% defect rate that customers will not tolerate. Stability and capability are separate concepts; you need both. SPC tells you when to act, not whether the level is good enough.
Try it
Run the numbers.
Pressure-test the concept against your own knowledge — answer the challenge or try the live scenario.
Knowledge Check
Your weekly customer churn rate has been: 2.1%, 2.4%, 2.0%, 2.3%, 2.2%, 2.5%, 1.9%, 2.4% (mean ~2.2%, σ ~0.2%). This week it's 2.6%. Three months ago it ran 1.8-2.1%. What does SPC suggest?
Industry benchmarks
Is your number good?
Calibrate against real-world tiers. Use these ranges as targets — not absolutes.
Process Capability Index (Cpk)
Manufacturing and service processes meeting customer specificationsWorld-Class (Six Sigma)
≥ 2.0
Strong
1.67-2.0
Capable
1.33-1.67
Marginal
1.0-1.33
Incapable
< 1.0
Source: Walter Shewhart / W. Edwards Deming / GE Six Sigma standards
Real-world cases
Companies that lived this.
Verified narratives with the numbers that prove (or break) the concept.
General Electric (Six Sigma Era)
1995-2001
Under Jack Welch, GE deployed Six Sigma — built on Shewhart/Deming SPC — across every business unit. Every process was charted; every defect tracked. GE Plastics in Mt. Vernon, IN reduced color variation in resin production by tightening SPC limits and re-centering processes; Cpk went from 0.9 (incapable) to 2.1 (world-class) over 18 months without major capex. Operators were trained to distinguish common vs. special cause and stop tampering. Welch publicly credited Six Sigma for $12B in cumulative savings by 2002, with SPC as the core technical method.
Cpk Improvement
0.9 → 2.1 (typical)
Cumulative Savings
$12B by 2002
Trained Black Belts
~5,000
Tampering Reduction
~90% (informal estimate)
SPC is the technical engine of Six Sigma. Without it, 'continuous improvement' devolves into reacting to noise. With it, every team has a shared language for what's signal vs. what's chatter.
Hypothetical: B2B SaaS Customer Success
Recent
A 400-employee SaaS company's CS team was 'in crisis mode' weekly because NPS bounced between 32 and 48 — leadership demanded action every time it dipped. The CS lead applied SPC: NPS mean 40, σ ~5, control limits 25-55. Almost every weekly reading was within control limits — the panic-and-respond cycle was tampering. Once leadership stopped reacting to noise and only investigated readings outside ±2σ or runs of 8+ on one side, NPS actually rose to a stable 47 (the constant interventions had been adding variance). Real special-cause investigations led to two structural improvements that lifted the mean.
NPS Mean Before
40 (panicky volatility)
NPS Mean After
47 (stable)
Weekly Crisis Meetings
1+ → 1/quarter
CS Team Burnout
Down significantly
Most leadership 'crisis response' on noisy weekly metrics is tampering. SPC gives leaders permission to ignore noise and focus only on real signals — the team works on actual problems instead of imagined ones.
Related concepts
Keep connecting.
The concepts that orbit this one — each one sharpens the others.
Beyond the concept
Turn Statistical Process Control into a live operating decision.
Use this concept as the framing layer, then move into the matched diagnostic — or have us scope the build.
Typical response time: 24h · No retainer required
Turn Statistical Process Control into a live operating decision.
Use Statistical Process Control as the framing layer, then move into diagnostics or advisory if this maps directly to a current business bottleneck.