Did introducing Statistical Process Control (SPC) charts reduce defects per batch — or is the evidence still too weak to say?

The audit question
Defect numbers look lower since the charts were introduced. But every batch is a little different anyway. Is the drop big and consistent enough to credit the charts?

Scenario

A manufacturer introduces SPC (Statistical Process Control) charts on the assembly line, giving operators real-time visibility into process drift. The question for the audit team: did the charts measurably reduce defects? They pull defect counts from 20 production batches before the introduction and 20 after, and run both sets through the tool.

Data

Counts (data0)
4
11
2
17
6
13
3
21
8
5
14
4
9
2
19
7
3
12
5
16
Counts (data1)
3
10
2
16
6
12
3
20
8
5
14
4
9
2
18
7
3
11
5
15

What the tool returned

The auditor pasted both datasets into the tool and ran the analysis. The following result came back:

Analyser
🖊️
Before the change (data0), the metric averaged 9.1, typically ranging between 2.0 and 19.0. Since the change (data1), it now averages 7.1. With 94% confidence the true figure sits between 5.0 and 9.5. 89.1% of samples tested confirmed a reduction of approximately 2.0 — good assurance that the decrease is real, though short of conclusive.
Size of the change
−2.0
Verdict
Moderate decrease
Model: LogNormal (auto-detected) n₀ = 20 · n₁ = 20 · ESS = 465

Audit conclusion

Across the 20 batches measured before and after, defects fell by a moderate but steady amount once SPC charts were in place. The drop is unlikely to be coincidence — the charts seem to be doing their job — but the evidence isn't yet overwhelming.

So the verdict is "on track," not "proven." The sensible next step: measure again once more batches have run, and double-check that nothing else changed in the meantime — same products, same line speed, same inspection rules — before giving the charts full credit.

Tool usage benefits

A drop of two defects per batch is easy to dismiss as noise — and just as easy to oversell as success. ChangeVerifier does neither: it quantifies how strong the evidence actually is, telling the auditor this is a real improvement worth crediting, but not yet a closed case. That distinction — between proven, promising, and unproven — is exactly what eyeballing two columns of batch counts cannot provide. From pasting 40 numbers to reading the verdict took under a minute, with no statistics background needed.