Number of inspection points / characteristics per unit
Defects across all units inspected
DPMO
—
Defects per Million Opportunities
SIGMA LEVEL
—
With 1.5σ shift
Yield (RTY)
—
Process yield %
DPU
—
Defects per unit
—
—
DPMO = (Defects / (Units × Opps)) × 1,000,000
Yield = (1 − Defects/(Units×Opps)) × 100%
σ = NORMSINV(1 − DPMO/1,000,000) + 1.5

Sigma level reference (with 1.5σ shift)

Sigma LevelDPMOYieldCpkQuality classification
6σ3.499.99966%2.0World class — aerospace, medical implants
5σ23399.977%1.67Excellent — top automotive, surgical devices
4σ6,21099.38%1.33Good — typical IATF 16949 supplier
3.5σ22,75097.7%1.17Acceptable
3σ66,81093.32%1.0Industry average — most Indian SMEs
2.5σ158,65584.13%0.83Below average
2σ308,53869.15%0.67Poor — cannot meet customer expectations
1σ691,46230.85%0.33Critical — process not in control

DPMO and Sigma Level

DPMO (Defects Per Million Opportunities) and Sigma Level are the two most common quality metrics in Six Sigma programs. They normalize defect rates across products and processes, making direct comparison possible.

DPMO Formula

DPMO = (Defects × 1,000,000) / (Units × Opportunities per unit)

"Opportunities" is the number of distinct ways a unit can be defective. A simple bracket might have 5 opportunities (length, width, thickness, hole position, surface finish). A complex assembly might have 100+.

Sigma Level Conversion

Sigma level is calculated from DPMO assuming a long-term 1.5σ shift (standard Motorola convention):

Worked Example

You inspect 500 brackets. Each has 5 opportunities for defect. You find 12 defects total.

Industry Sigma Benchmarks

Improving Sigma Level

Each sigma level requires roughly 10× fewer defects than the previous. Going from 3σ to 4σ is ~10× improvement; 4σ to 5σ another 10×. The work shifts from "fix obvious problems" (3-4σ) to "advanced process control" (5-6σ).

Related Tools

For root-cause analysis on top defect modes, see FMEA RPN. Verify capability with Cp/Cpk. Quantify the financial impact in Cost of Poor Quality. Set sampling plans with AQL Sampling.