Building trustworthy data and translating VOC to measurable CTQs.

When SPC Is Not Enough: Combining Statistical Process Control with Predictive AI in 2026

Short answer first. Predictive AI should complement your control charts, never replace them, especially when you need earlier warning from multivariate or time-dependent patterns that a single chart cannot see. Control-chart limits remain the reference for evaluating statistical stability, while models can extend the warning horizon. This article compares control-chart signals with predictive alerts, [...]

Synthetic Data in Six Sigma: What It Can Test and What It Cannot Prove

Synthetic data has a narrow, useful job in process improvement. It can exercise your data pipeline, stress analytical logic, and create rare scenarios that your historical records never captured. It cannot prove how often a defect actually occurs, what truly causes it, or how a process performs under real operating conditions. That distinction gets [...]

Acceptance Sampling Plans: Why AQL Is Not a Guarantee for Every Lot

Acceptance sampling plans are often misread. A buyer may see “AQL 1.0” on a purchase order and assume every accepted shipment contains no more than 1% nonconforming units. That assumption is incorrect. Under ISO’s current AQL standard, the acceptance quality limit indexes a sampling scheme for a continuing series of lots; it does not [...]

Gage Bias and Linearity: When Good Repeatability Still Gives the Wrong Answer

A gage can produce nearly identical readings and still be wrong. Repeatability shows how tightly readings cluster under the same conditions; it does not show whether those readings agree with a suitable reference value. This article explains how to evaluate gage bias and linearity across the working range. It covers reference standards, study design, [...]

Cpk Confidence Intervals: How to Report Capability When the Estimate Is Uncertain

A Cpk value printed to two decimals looks certain. It is not. Every Cpk you calculate comes from a sample, and a different sample from the same process would hand you a different number. That is why a capability estimate should usually be reported with an uncertainty range. Sample size, distribution shape, process stability, [...]

Rational Subgrouping: Why Mixing Machines Can Hide a Control-Chart Signal

Rational subgrouping keeps short-term, common-cause variation inside each subgroup so the chart can expose meaningful change between subgroups. When you blend two machines, two spindles, or two shifts into one subgroup, that difference becomes part of the within subgroup variation. The control limits widen, the signal gets absorbed, and the process looks calm while [...]

Test Automation Coverage Gaps: A Six Sigma Lens on Modern QA Metrics

Test automation coverage should be judged against product risk and failure modes, not only the percentage of automated cases. A suite can report high coverage while still missing a critical payment path or failure condition. The number may look reassuring while the underlying risk remains. This article looks at coverage gaps through a Six [...]

AI Document Extraction Quality: Measure Field Errors Before Automating Data Entry

Document extraction quality should be measured field by field, weighted by business risk, and judged by exception outcomes before automation is scaled. One overall accuracy percentage can hide the errors that matter most. The NIST text-recognition evaluation guide distinguishes character- and word-based OCR metrics, so high text-recognition accuracy does not by itself prove that [...]

CAPA Effectiveness Checks: Define the Evidence Before Closing the Action

A corrective and preventive action should close only when predefined evidence shows the cause was addressed and performance stayed acceptable across meaningful exposure. That may include production volume, transactions, service encounters, or other opportunities relevant to the process. Completing implementation proves the planned work occurred. It does not by itself prove the problem stopped. [...]

EWMA vs. CUSUM: Choosing a Chart for Small Process Shifts

EWMA and CUSUM charts both use information from successive observations to detect process-mean changes that may be difficult to see on a conventional Shewhart chart. EWMA applies exponentially decreasing weights to older observations, while CUSUM accumulates deviations from a reference value until the statistic resets or crosses its decision limit. This comparison explains their [...]

Go to Top