Where quality is lost between the check sheet and the buyer’s report
Each problem sits in a record QA already keeps: the end-line check sheet, the rework register, the inspector’s report, the NCR log. The cost shows up as rework hours, a lot held back from shipment or an open finding at the next buyer audit.
End-line results on paper that nobody can add up
Checkers mark defects on paper at the end of each line. The totals reach the QA manager a day late, and DHU by line and the weekly defect Pareto are typed into Excel by hand, or skipped. When a buyer asks for last month’s DHU on line 7, the answer is a guess. A 2019 MIT thesis by Hsu says most factories in one large sourcing network still record quality data by hand, and the records are incomplete and inaccurate.
8% recorded, 13% found
One Bangladeshi factory in a 2021–22 ILO pilot recorded 8% defects. Once it captured defects more strictly, the next measurement found 13%. The report puts the rise down to the stricter capture.
Source: International Labour Organization (ILO), Implementation of the Factory Improvement Toolset in the Asian garment sector: Final report (opens in a new tab), 2023.
Rework that never shows in the defect rate
Garments are mended at the end of the line before anyone counts them. Overtime, re-pressing, re-washing and re-packing absorb the failures, and the finishing defect rate still looks low. The cost never appears as its own line in the accounts. The owner sees a thinner margin, and the QA head is asked to improve quality with no baseline to start from.
3% vs almost 25%
At one exporter, the finishing unit’s defect rate was 3%, yet almost 25% of garments left the sewing line with a defect and were repaired in two hours of daily overtime. A QA professional’s example, 2008.
Source: Apparel Resources (StitchWorld), Quality Costs (opens in a new tab), 2008.
NCRs that stay open, and a defect that comes back next month
The NCR is raised in a mail or on paper. Containment is late, the root cause reads “operator negligence”, the CAPA has no owner or due date, and nobody runs the effectiveness check. The same open seam returns on another line. Studying four Bangladeshi supplier factories, Hoque and Maalouf (2022) named the lack of a systematic approach to quality control as a key reason for poor quality.
50%
In one Bangladeshi knit export factory, across five styles, open seams alone were half of all observed defects, skip stitch 29% and label mistakes 7%. Three defect types made up most of the total.
Source: Textile & Leather Review, Investigating Root Causes of Sewing Defects Using TQM Tools for Quality Improvement in the Knit Garment Industry (opens in a new tab), 2023.
A lot rejected at final inspection, and the dispute lost on paper
The buyer’s third-party inspector fails a lot at AQL 2.5. Your team believes the sample came from two cartons, or that a minor was called major. To ask for a second opinion, the QA head needs the inspector’s report, the buyer’s AQL sheet and defect list, and your own pre-final results, spread across mail and folders, while the vessel cut-off comes closer. The merchandiser and the owner wait on the answer.
From check sheet to NCR: five jobs for the quality agent
It works from the inspection reports, AQL sheets, lab reports and NCR log you upload or keep in Google Sheets, and cites the report or rows it used. An admin approves or declines anything sent to a buyer, an inspector or the line; it goes only if approved.
The AQL check, worked through from the report
Answers with sources
Ask whether a lot passes and the agent reads the inspection report and the buyer’s AQL sheet. It states the lot size, the sample size and the accept and reject numbers for majors and minors, then sets the recorded counts beside them. If the buyer’s sheet is silent, it uses the playbook’s ISO 2859-1 Level II table for normal inspection and says so. One critical defect fails the lot, however large the lot is.
Ask“Lot L-2209 is 4,000 pieces, AQL 2.5 major and 4.0 minor. Does the inspection pass? Show the sample size and the accept and reject numbers.”
Rework counted beside DHU, from the check-sheet rows
Answers with sources
Keep the inline or end-line check sheet in Excel or CSV: one row per line and hour, with the week, shift, style, pieces checked, defects found and pieces reworked. Upload it and name a line and a week, a shift or a style. The agent adds up the matching rows, divides for DHU and sets the reworked pieces beside it, so mending at the end of the line no longer hides. Anyone can check the figure against the rows it lists.
Ask“What was DHU on line 6 in week 45, and how many pieces did line 6 send back for rework that week?”
Every test on the buyer’s protocol, checked for a gap
Reads your documents
Upload the buyer’s test protocol and the lab reports for a lot. The agent goes down the protocol test by test: wash, light, rubbing and perspiration fastness, shrinkage, pilling, tensile, GSM and fibre composition. For each it gives the result, the grade required and the method used. It names any test no report covers, a wash grade with no scale (change in shade or staining), and shrinkage with no sign.
Ask“Lot L-2209 ships on the 28th. Which tests on the buyer’s protocol are missing or incomplete in our lab reports?”
The NCR written up the way your SOP asks
Creates files
Give the agent the finding, the lot and the NCR number. It writes the NCR as a document from your facts and the reports it cites: what was found, the containment, the root cause your team reached, the CAPA and the run where the effectiveness check will happen. It leaves the CAPA owner and due date blank until you give them. Ask for a change and it edits that section; every revision is kept.
Ask“Write NCR-0412 for the open seams on line 6, lot L-2240, from today’s end-line report. Containment: three cartons segregated.”
Containment sent to the line, once an admin approves
Acts in your tools
The playbook gives you 24 hours to segregate a suspect lot. The agent prepares a WhatsApp Business or Slack message telling the finishing supervisor which cartons to hold, and the new row for your NCR log in Google Sheets. Both wait in Mavick for a department admin, who approves or declines each one. Each runs only if approved.
Ask“Lot L-2240 failed our pre-final on open seams. Draft the hold message to the finishing supervisor and add the NCR to our log.”
A fabric delivery, a rejected lot and a defect that keeps coming back
A fabric delivery with too many points
Fabric inspection in-charge
The 4-point inspection of 10% of the rolls on a shell-fabric delivery is done, two days before cutting is due to start.
Asks
“From today’s roll inspection sheet for GRN-5531, how many rolls failed the buyer’s 40 points per 100 square yards, and what does the buyer’s fabric standard say to do next?”
The agent
Reads the uploaded roll inspection sheet and counts, from its rows, the rolls your inspector marked as failing, listing each roll with its length, its recorded points per 100 square yards and its main defect. It quotes the action the buyer’s fabric standard sets, such as inspecting every roll when too many fail, and cites the clause. It writes a deviation note as a document.
Result
The QA head decides to accept, accept with deviation or reject before cutting plans the lay, with the failing rolls counted from the sheet and cited.
The buyer’s inspector rejects a lot
Final inspection in-charge
The buyer’s third-party inspector fails PO 8816 at AQL 2.5 on majors, two days before the vessel cut-off.
Asks
“The inspector failed PO 8816 on majors. Set their counts beside our 3× pull, class by class as each report records them, and draft an email asking for a second opinion.”
The agent
Reads the inspector’s report, the buyer’s AQL sheet and defect classification list, and your in-house pull at three times the sample size, uploaded as an Excel file. Sets the two sets of counts side by side, class by class, with the carton numbers recorded, and names any document missing from the pack. Builds the pack as a spreadsheet and prepares the email to the buyer’s QA contact, copied to the inspection company, to go through Outlook.
Approval
The email waits as an action until the QA head, as department admin, approves or declines it. It goes to the buyer’s QA and the inspection company only if approved, and the audit trail shows who decided.
Result
The QA head decides on a dispute built from counts, carton numbers and the buyer’s own defect list. The playbook’s rule: never argue with an inspector, document.
The same open seam, back again this quarter
QA head
QA raises another NCR for open seams on line 6, the same defect as an NCR earlier this quarter.
Asks
“How many NCRs for open seams have we raised in Q4? Show each one’s root cause, CAPA owner and whether its effectiveness check passed.”
The agent
Counts the rows in the uploaded NCR log with open seam as the defect and Q4 in its quarter column (a log with only dates needs one), and lists each with its recorded root cause, CAPA owner, due date and effectiveness check. It quotes the playbook: a repeat within a quarter means the last CAPA did not work, so re-open it for a cross-functional review. On a third repeat, the QA head presents it to the COO or CEO. It writes the review note as a document.
Result
The QA head goes into the review with every NCR cited. Questions about stitch settings go to Garmenting & Stitching, who own them.
The cost of poor quality, sized by published figures
Each figure is from a published study or trade-press report on garment factories in Asia, source linked. They size poor quality in other factories, not anything Mavick did. Before a pilot, take your FRI pass rate, lab pass rate and NCR ageing as the baseline.
Rework and rejection counted, not absorbed
14.05% of sales
Average cost of quality in a study of 61 Indian apparel manufacturers, up to 30% in some. Internal failure was the largest part at 9.86% of sales; prevention got 0.26%. Published 2009.
How the agent helps
The agent does not price quality for you. Upload the alteration and rejection registers as Excel or CSV files, and it counts reworked or rejected pieces from the rows for the lines, stages, styles or defects you name, a few in one answer. If a register carries a cost on each row, it sums it. It never makes up a cost.
Track: Alteration and rejection % by line and stage
Source: GTZ (now GIZ), report by Rajesh Bheda, Profiting through Quality: Improving “Right First Time”. Learning from systematic and collaborative implementation of Quality Improvement Program in 10 apparel firms (opens in a new tab), 2009.
DHU for each line, counted from the rows
13.70%
Average sewing-room DHU at the start of a 2021–22 ILO pilot, in the four Bangladeshi factories that worked on their sewing rooms. Three such factories in Cambodia started at 13%, three in Pakistan at 21%. Not a national average.
How the agent helps
A baseline needs one number per line, week after week. Name the lines and the week, and the agent counts each line’s DHU from the uploaded check sheet, a few lines in one answer, citing the rows. For the whole floor, it works through the lines a few at a time and puts the figures into a spreadsheet the QA head can send on to each line chief.
Track: DHU by line, week by week
Source: International Labour Organization (ILO), Implementation of the Factory Improvement Toolset in the Asian garment sector: Final report (opens in a new tab), 2023.
Fewer lots failed at final random inspection
36 person-hours
Time one exporter spent rechecking, altering and repacking 419 skirts after the buyer’s QA failed the shipment at final audit for run-off stitches. It still shipped on time. A QA professional’s example, 2008.
How the agent helps
The agent does not inspect. Before the buyer’s inspector arrives, ask it to set your pre-final counts for each lot against the buyer’s accept and reject numbers, class by class, citing the report and the AQL sheet. It quotes the playbook’s sampling rules, such as drawing across cartons and having at least 80% packed.
Track: Buyer third-party FRI pass rate
Source: Apparel Resources (StitchWorld), Quality Costs (opens in a new tab), 2008.
What to baseline before a pilot
Measure each one before the pilot starts, then again at its end.
| KPI to baseline | Why it matters |
|---|---|
| In-house FRI pass rate | The playbook targets 98% or better. Below 95%, it calls for a review of the upstream processes. |
| Buyer third-party FRI pass rate | The number the buyer sees. The playbook’s target is 100%, and it makes any fail an item for the quarterly review. |
| Lab-test pass rate | The playbook targets 99% or better, varying by buyer. A fail on a critical parameter, on a lot with a confirmed PO, goes to the QA head, plant head and Merchandising within two hours. |
| NCR open ageing: NCRs open more than 30 days | The playbook’s target is none. Open NCRs and their status are on its pre-audit checklist, so expect a buyer’s auditor to ask for them. |
| CAPA effectiveness rate | The share of CAPAs that pass their check at the next equivalent run. The playbook targets 90% or more, and doubts any figure above 95%. |
| Repeat-defect rate within a quarter | The playbook targets 5% or less of defects repeating within a quarter. A repeat NCR on the same defect means the last CAPA did not work. |
What the agent works from
Records it reads
- Inspection report
- AQL sheet
- Buyer defect classification list
- Inline and end-line check sheet
- Incoming fabric inspection report (4-point)
- Lab test report
- NCR (non-conformance report) and CAPA plan
- CAPA effectiveness check
In your ERP
Reads Odoo production orders (order, product, quantity, status) and Odoo stock transfers (status, scheduled date, source document). Drafts a transfer note, or an order’s planned start or assignee; an admin approves or declines, and it runs only if approved. No ERP gives Quality its inspection, lab or NCR records: they come from uploads and Google Sheets.
One ERP connection serves the whole factory: SAP Business One, SAP S/4HANA, Odoo, Oracle NetSuite, Microsoft Dynamics 365, WFX Cloud ERP or Datatex NOW. This agent reads only the records named above; we check yours on the call.
In your tools
Reads the inspection company’s emails in Gmail or Outlook, or your NCR log in Google Sheets, when you ask. Prepares the email asking for a second opinion, via Gmail or Outlook, a new row for the NCR log, or a WhatsApp Business or Slack message telling a supervisor which cartons to hold. An admin approves or declines each one; it runs only if approved.
Tools named today: Gmail, Outlook, Google Sheets, Slack and WhatsApp Business.
Never made up
- Inspection results, defect counts or AQL verdicts
- Lab test values or test-method references
- NCR numbers, CAPA status or closure dates
- Buyer standards, tolerances or acceptance criteria
If it is not in your connected data, the agent says so.
Belongs to another department
- Stitch and machine settings behind a sewing defectGarmenting & Stitching
- Metamerism: a shade that matches under D65 but fails under TL84Dyeing & Processing
- Buyer standard negotiation and approvalsMerchandising
- Certification scope and audit calendarsCompliance & Sustainability
- Machine capability and repairMaintenance & Engineering
What the Quality Assurance & QC playbook covers
SOP: from NCR to a closed CAPA
An NCR is raised on any out-of-spec finding. The suspect lot is segregated within 24 hours, and the root cause found within 72 hours with 5 Whys or a fishbone. The CAPA gets an owner and a due date. The NCR closes only when the effectiveness check passes at the next equivalent run.
Decision rule: the buyer’s inspector rejects a lot
Read the inspector’s report first: which defect class, at which AQL. Pull three times the sample size in-house and classify each defect the same way, against the buyer’s list. If you agree, rework or part-ship. If not, ask for a second opinion. The playbook says disputes are won with better sampling and clear photographs.
KPI: CAPA effectiveness, read both ways
The target is 90% or more of CAPAs passing their effectiveness check. Above 95%, the playbook says closure is being overstated. Below 70%, the root-cause work is stopping too early.
Edge case: your lab passes, the buyer’s lab fails
Check the differences in order: method (ISO or AATCC), specimen, conditioning at 20 ± 2 °C and 65 ± 4% RH, calibration, then the visual rating, where two graders can differ by half a step. In the playbook’s words, test method matters more than test result.
Written as general textile practice and reviewed by four independent AI models. Not yet signed off by a human subject-matter expert. Upload your own SOPs and standards, and every answer shows which source it used.
Ten questions it ships with
The template’s starter questions. Your team can edit them, or ask in its own words.
Inspection result for a lot
What did the inspection record for this lot? Show defect types, quantities, and the verdict.
Does this lot pass AQL
Does this lot pass the buyer’s AQL level? Show the sample size, defect counts by class, and the accept/reject limits used.
Lab results against the standard
What are the lab results for this lot — colour fastness, shrinkage, pilling, tensile, GSM, fibre composition — and how does each compare with the required standard?
Open NCRs and CAPA status
List the open non-conformance reports, what each covers, and where the corrective action stands.
Recurring defects across lots
Which defects recur most often across recent lots, and what do the inspection reports attribute them to?
Assemble an inspection pack
Assemble the inspection pack for this shipment from the reports on record, and tell me which documents are missing.
Rejection rate by stage
Show the rejection rate by process stage for this month, and tell me where it moved compared with last month.
Incoming inspection on a delivery
What did the incoming inspection record for this delivery, and was any material rejected or accepted with deviation?
Buyer standard for a test
What standard does this buyer require for this test, and what test method does it specify? Cite the document.
Lots pending final inspection
Which lots are pending final random inspection, when are they due to ship, and which are at risk?
Quality Assurance & QC: questions before a demo
Can it read the reports our third-party inspection company sends?
Yes, when the PDF has text in it. Upload it to the quality department, and the agent reads the sample size, the AQL, the defects by class and the result, and cites the report. It reads the text and tables, not the photographs, so it cannot judge a defect from a picture. A scanned report, or a photo of a page, has no text to read: ask the inspection company for the original PDF.
Our end-line check sheets are on paper. Can it still give us DHU?
Not from the paper. The agent cannot read a sheet that stays on the line, and a photo of one has no text to read. Once results are typed into an Excel or CSV file you upload, it gives DHU for the lines or shifts you name, a few in one answer, counting defects and pieces checked from the rows and citing them. For the whole floor, it works through the lines a few at a time and puts the figures into a spreadsheet. In Excel, keep the results on the first sheet: that sheet is the table, up to 5,000 rows. Mavick does not capture defects on the floor.
Does it inspect fabric or garments with a camera?
No. Much of what is written about AI in quality control is camera inspection of fabric. Mavick does none of that. It has no cameras, it does not connect to inspection machines or lab instruments, and it does not grade a defect or a fastness result from a picture. It works from the records your inspectors and lab keep, once you upload them or connect the sheet, and when a reading is not there, it says so.
Can the agent read inspection results from our ERP’s quality module?
No. Even where your ERP keeps inspections, lab results or NCRs, no ERP connection gives them to the quality agent: it works from the files you upload to quality and the Google Sheets you connect. From the ERP it reads Odoo production orders and stock transfers. No inspection result, lab value or NCR is ever written to an ERP. In Odoo the agent can draft a note on a stock transfer, such as a lot on hold for re-inspection; the note informs and does not block stock. An admin approves or declines it, and it runs only if approved. On any other ERP, the hold goes as a new row in your NCR log in Google Sheets, with the same approval.
Can the agent pass or fail a lot against AQL?
It does not inspect, and it never invents a count or a verdict. It reads the counts your inspector recorded and sets them against the sample size and the accept and reject numbers in the buyer’s AQL sheet, class by class. When the buyer’s sheet is silent, it uses the playbook’s ISO 2859-1 General Inspection Level II table for normal inspection and says so; if the buyer has moved you to tightened inspection, the buyer’s numbers apply. Critical defects are never on the plan: one critical fails the lot. The decision stays with your QA team.
Can Merchandising or the sewing floor see our NCR log?
Not if you upload it to quality. A file uploaded to one department stays in that department, and Mavick builds that boundary into how it stores and retrieves everything. A file is seen beyond quality only if someone shares it across the factory. The boundary holds both ways: the quality agent cannot read the files the sewing floor uploads or merchandising’s mail. Asked about the stitch settings behind a defect, it says Garmenting & Stitching owns them.
Bring Quality Assurance & QC. We’ll build it on the call.
Book a 30-minute demo (opens in a new tab)- support@mavick.si
- Demo calendar
- calendly.com/safock/30min (opens in a new tab)