Skip to content

Dan Hill

Quick Guide · Non-conformance and CAPA

Spotting repeat non-conformances across shifts using Copilot Chat

Containment actions often disguise repeat non-conformances across different production shifts. This quick guide shows how technical teams can use Copilot Chat to cluster annual register data and identify recurring packaging defects under BRCGS clause 3.7.

22 Oct 2026 · 39 sec · Copilot Chat + Excel (Copilot Chat, no licence needed)

The scenario

Site: Cooked and sliced meats facility (dummy example: Redford Meats Ltd)
Who: Quality Assurance Manager
Trigger: A recurring seal integrity defect appears on slicing Line 2 across alternating shifts, despite multiple closed CAPAs over the past twelve months.
Why it matters: Persistent packaging faults create operational waste and compromise shelf-life. In effective BRCGS audits, repeating the same non-conformance without verified preventive action risks a major non-conformance against fundamental clause 3.7.

BRCGS context

Clause	What the Standard expects	Sensible practice	Where technology helps
3.7 Corrective and preventive actions	Failures are investigated, root cause identified where needed, and actions implemented and verified.	Sensible sites separate containment corrections from long-term preventive fixes, setting mandatory 30-day or 60-day effectiveness checks before marking CAPAs fully closed.	Copilot Chat can rapidly scan free-text logs to surface matching defect patterns and identify entries where preventive action is missing or identical to immediate containment. It cannot verify whether the physical engineering fix was competent or effective.
3.8 Control of non-conforming product	Non-conforming product is identified, segregated and managed.	Sites maintain consistent batch-quarantine references in defect registers to trace disposition and rework outcomes cleanly.	Copilot can reconcile defect descriptions against quarantine disposition columns in an exported sheet to flag incomplete hold notes. It cannot check the physical status of quarantined stock.

Clause intent in plain English, not the Standard's wording. Check requirements against your own licensed copy of BRCGS Food Safety Issue 9.

The approach, step by step

01
Upload spreadsheet and cluster recurring defects
```
Analyse the attached non-conformance register. Group recurring packaging defects by line and issue description, providing a count for each group.
```
You check: Verify that distinct engineering issues have not been inappropriately merged into a single generic cluster.
02
Identify deficient preventive actions and missed reviews
```
Review the Line 2 seal defect records. Flag rows where Preventive_Action repeats Immediate_Correction or where Effectiveness_Review_Date is blank.
```
You check: Cross-check flagged rows against the raw register to ensure data was not misread or blank cells hallucinated.
03
Analyse shift and timeline trends
```
For the identified Line 2 seal records, summarise occurrences by date and shift. Note whether the recurrence links to specific shift patterns.
```
You check: Review production shift logs for those dates to confirm whether tooling changes or maintenance handover logs correlate with the findings.
04
Draft weekly quality meeting brief
```
Draft 3 concise bullet points for the quality meeting summarising Line 2 recurring seal issues, showing why previous CAPAs failed and proposing next review steps.
```
You check: Check tone and accuracy against actual site engineering terms and ensure no unverified assumptions are presented as fact.

Takeaways

Treating immediate containment as preventive action allows non-conformances to resurface over time across shifts.
Copilot Chat can cluster unstructured defect descriptions across exported registers to highlight duplicate corrective actions.
Clause 3.7 requires verified preventive action; register analysis must always be confirmed by physical line and tooling inspections.

Limitations and risks

Copilot Chat cannot read live SharePoint lists, production SCADA systems, or maintenance databases directly; data must be exported and uploaded manually.
Unstructured text descriptions written in colloquial factory phrasing may lead to inconsistent grouping if phrasing differs significantly.
The model cannot inspect physical sealing jaws, heat sensors, or packaging film tension to determine the actual physical root cause.
Copilot cannot evaluate whether an effectiveness check was technically robust, only whether the register indicates a date or entry.
Over-reliance on automated summaries leading the QA team to accept AI groupings without checking the physical line condition or batch records.
Uploading confidential operational spreadsheets without confirming enterprise data protection under an authenticated work tenant.
Hallucinating causal links between shifts or machine settings if the underlying register lacks granular environmental or setup data.

Sources

BRCGS Food Safety Issue 9 (official)

TFCD content is general guidance, not professional advice for your site. Always check requirements against your own licensed copy of the BRCGS Standard and your customers' codes of practice. Narrated with an AI voice clone of the presenter; Copilot screens are illustrative recreations with dummy data.