botminds.aiRequest a demo
WorkflowQuality & Labeling

CAPA Effectiveness Verification

CAPA effectiveness verification that counts each lot since the fix, shows what the count proves, and searches each plant sharing the part before a board chair signs.

See the demoBook a demo
Follow one check from 06:10 to a reopened CAPA ↓
Watch with transcript →
The problem

A CAPA closes on a quiet spell while the same cause runs at a sister plant

An owner writes monitor for recurrence, a few quiet months pass, and the check is marked effective. The criterion never named the records, the count or the confidence, so a quiet period could only pass. Proof sits in four or five systems, and pulling it lot by lot takes most of a working day per check.

The fix went to the plant that opened the CAPA. A sister plant making the same part on the old program logs the same defect under another code, and a local hand rework there hides it. Nobody searches that plant, because the CAPA was never about it.

Built forCAPA coordinatorsQuality engineersQuality managersCAPA review boardsInternal auditors

The Problem Today

Most CAPAs close on a quiet spell and a signature, with no count of lots and no stated confidence.

6 h
of a quality engineer’s time per check, pulled by hand our estimate

Pulled by hand from each system

  • NCMRs
  • Inspection
  • Yield
  • Audits
  • Criteria that can’t failMonitor for recurrence passes any quiet month
  • One plant at a timeA sister plant on the old process goes unsearched

The Solution

  1. Criterion tested firstVague ones redrafted with the records, the count and the confidence
  2. Each lot on a timelineIn-process, visual and final inspection read for each lot since the fix
  3. Sister sites searchedSame part, program or failure mode, matched by meaning across codes
  4. The chair opens, then signsSign stays locked until each cited record has been opened

Built-in controls

  • Read-only on each quality record
  • Nobody signs their own draft
  • Looser changes wait for a second signature

What You Get

Quality teams close CAPAs on evidence, and catch a repeat cause at the plant the fix missed.

$264K
a year of quality engineer time across one large maker’s checks our estimate
  • Criteria that can failVague criteria redrafted for owners to accept
  • Late checks seen earlyThe last lot’s date projected from each plant’s pace
  • Evidence an auditor can readEach verdict with its lots, sums and signatures
One check, one morning

41 clean lots, then the CAPA effectiveness check finds a sister plant

A burr on the keel of a tibial baseplate, fixed at one plant with a new tool path and a 10× edge check. The effectiveness check is due this quarter; here is the morning of the decision.

demoPlants, people and numbers shown here are invented sample data.

  1. 0106:10

    Forty-one clean lots since the fix, and the 59th lands after the due date

    CAPA coordinator

    The coordinator opens the check as a timeline: one mark per lot made at Plant 2 since the new tool path went live. Forty-one are clean at in-process, 10× visual and final inspection, and 18 are still to come. At 0.81 lots a week the last one lands after the due date. Other sites show as not searched yet.

  2. 0206:10

    Forty-one lots give 87.8% confidence, short of the 95% the criterion asks

    CAPA coordinator

    The sum sits under the timeline with the working shown step by step. Fifty-nine lots with no keel burr give 95% confidence that fewer than 5% of lots carry one. Forty-one give 87.8%, and at 95% they rule out a rate above 7.0%. The board can redo each line, and the record says not yet in numbers.

  3. 0310:41

    Plant 3 makes the same baseplate on the old program, with three burr reports

    CAPA coordinator

    Before any verdict, the coordinator asks the agents to search other sites. A second lane appears: Plant 3 makes the same sizes on the old machining program, because the change order named Plant 2 only. Three nonconformance reports there describe a raised edge at the keel fin root. They were coded edge condition, so nobody had linked them.

  4. 0410:41

    Four parts in one lot with a raised edge, reworked by hand

    CAPA coordinator

    One red mark opens its lot: made on the old program, 194 of 198 parts conforming in process, no 10× keel check in the Plant 3 routing, released after rework. The report names a raised edge at the keel fin root on 4 parts. The panel calls it the same failure as the CAPA, on the old program, with the record cited.

  5. 0510:41

    Six candidates searched, and no record found is never read as none

    CAPA coordinator

    The recurrence search lists each site and supplier sharing the part family, the program, the keel feature or the failure mode. Plant 3 matches on part, program and feature despite its different defect code. Two plants show no record found, which the screen says proves nothing. The supplier that rough-machines the blanks keeps monthly counts only, so it reads can’t tell, with the person to ask.

  6. 0610:44

    Not effective: scope incomplete, with each statement cited

    CAPA coordinator

    The verdict drafter writes not effective, under the reason code for the same cause outside scope, and proposes reopening the CAPA with a new action at Plant 3: bring the revised program and the 10× keel check there. Each statement carries its citation. The coordinator reads the draft, can edit the words, and sends it to the board chair.

  7. 0710:48

    The sign button waits until the chair has opened all three Plant 3 reports

    CAPA review board chair

    The chair gets the draft and, beside it, the three Plant 3 reports he must open first. The screen counts a report only once he has had it open for a proper read. With three of three opened, the sign button wakes, and the meaning it will record is to reopen this CAPA and extend its scope to Plant 3. Send back sits beside it.

  8. 0811:03

    Reopened: Plant 2 keeps its 41 lots, and Plant 3 starts at 0 of 59

    CAPA review board chair

    He signs with his password, and the CAPA reopens. The criterion now asks for 59 clean lots at each plant. Plant 2 keeps its count, and a new Plant 3 lane starts from the first lot on the revised program. Each plant gets its own due date from its own pace, 0.58 lots a week at Plant 3.

  9. 0911:05

    Tighter criterion: it applies at once, and the manager reads it after

    Quality engineering manager

    The criterion check sets the old and new wording side by side, each tested for records, count, confidence and whether a quiet period can pass. Asking for more lots is tighter, so the change took effect at once. A looser one would wait for her approval. Earlier sign-offs no longer cover the CAPA, and its evidence pack shows unsigned until it is verified again.

Plant 3 logged the same burr three times, as edge condition

Each one names a raised edge at the keel fin root on the old machining program. Each lot was reworked by hand and released.

ReportParts in the lotParts affectedDisposition
First report2146Rework by hand deburr
Second report1984Rework
Third report2409Rework
Since the hand deburr joined each lotEvery lot after it, of 29 since the fixNo report raisedThe work-around hides the burr

demoSample data, from the working solution. The search matched them by part, program and feature. A search by defect code finds none of them.

Around the check

The quarter’s CAPA checks, lot by lot and signature by signature

What the coordinator, the board and the auditors work from on the other days.

Due this quarter293 data-based checks across the sites, each with its count, state, owner and days late. Vague criteria and unreadable feeds have their own filters.
Lot ledgerEach lot behind a mark: size, parts, date made, in-process, 10× visual, final and nonconformance records, each read from its own system. The ledger downloads as CSV.
Review boardA batch of 12 effective verdicts with a sample the system picked. The chair opens the sample before one signature that writes a row per CAPA.
Sample sizePick confidence, the rate to rule out and failures allowed, and see the lots needed and what the counted lots prove. A proposed change goes to the manager.
Audit trailEach action with its reason, its session and what each signature means. Agent steps are marked and carry no signature.
SettingsDefault confidence, rate and minimum window, approvers and clocks. A looser default waits for the board chair; a tighter one applies at once.

demoSample data, from the working solution.

What a closed-on-a-quiet-spell CAPA would have passed

Each came from reading the records behind a check, not its status field.

Criteria that cannot fail

Of the quarter’s 293 checks, 46 have a criterion the agents cannot measure as written. Each gets a redraft for its owner to accept.

One cause under two codes

Plant 3 logged the keel burr as edge condition. A code search returns nothing; the match came from part, program and feature.

A stopped feed shown as not read

When one plant’s inspection feed stopped, 9 checks showed those lots dashed as not read, with a named owner, instead of counting quiet days as clean.

A work-around hiding the cause

Plant 3 added a hand deburr to each lot and the reports stopped. The old program still leaves the burr.

demoSample data, from the working solution.

The agents

Six agents count, search and draft, and the board chair signs

Each proposes; a named person decides.

Criterion checker

Tests whether each criterion names the records, the count, the confidence and a baseline, and redrafts a vague one.

  • Never approves its own redraft
  • Never loosens a criterion
Evidence reader

Reads nonconformance reports, in-process and final inspection, lot genealogy, yield and audit findings for each lot since the fix.

  • Never edits a quality record
  • Never reads complaint or field records
Success-run counter

Places each lot, computes the lots needed, today’s confidence, the rate ruled out, the pace and the last-lot date.

  • Never rounds a confidence up
  • Never counts a lot made before the fix
Recurrence searcher

Searches sites, products and suppliers sharing the part family, program, feature or failure mode, by meaning.

  • Never says none when it means no record found
Verdict drafter

Drafts the verdict, reason code, scope change and new action, each statement cited.

  • Never signs, sends or closes
  • Never drafts effective over a failing lot
Due-date keeper

Watches due dates and the board clock, and drafts extension requests with the projected last lot.

  • Never extends a date itself

Six systems behind one lot timeline

Each is read only, and each mark on the timeline cites the record it came from.

Quality management systemNonconformance reports and the CAPA record
Manufacturing execution systemIn-process, 10× visual and final inspection, lot by lot
ERPLot genealogy: which plant, which program, which lot
Product lifecycle managementPart and program revisions, and where a change order applies
Audit moduleInternal and supplier audit findings
Yield and process dataYields and process records by line

demoSample data, from the working solution.

Signing, approving, extending and releasing stay with named people

Agents count and draft. Each of these decisions has a named owner.

Signing a verdictThe board chair signs after opening each cited record. An alternate chair never signs their own work.
Approving a criterionThe owner accepts a redraft and the quality engineering manager approves. A looser criterion waits for the manager.
Extending a due dateThe due-date keeper drafts the request with the projected last lot. The chair decides.
Releasing or shipping a lotThe assistant gives no answer and names the plant’s quality release lead.
Complaint-sourced CAPAsThe complaint handling team verifies those outside this queue, and no agent here reads complaint records.

What quality teams ask about CAPA effectiveness

What is CAPA effectiveness verification?

It is proving that a corrective action worked, beyond showing it was done. Here CAPA effectiveness is checked on the lots made since the fix: agents read each lot’s inspection and nonconformance records, show what the count proves, search other plants for the same cause and draft a verdict a board chair signs.

How many lots does a CAPA effectiveness check need?

That follows from the confidence and the rate you want to rule out, with no failures allowed. The sample size screen shows the formula, a table of common plans, and what the lots already counted prove at the confidence you chose.

What makes a CAPA effectiveness criterion measurable?

It names the records, the count and the confidence, and a quiet period alone cannot pass it. The criterion checker tests each one before any data is counted and redrafts a vague one for the owner to accept and the manager to approve.

Does it search other plants for the same cause?

Yes. The recurrence searcher looks at each site, product and supplier sharing the part family, program, feature or failure mode, and matches by description, so a defect logged under another code still shows up.

What happens when a plant’s records cannot be read?

Those lots show dashed as not read, with the named owner of the feed. They are never counted as clean, and the check cannot be verified on them.

Can the board sign a batch of verdicts at once?

Effective verdicts can go as a batch. The system picks a sample, the chair must open it first, any item can be pulled out, and one signature writes a separate audit row for each CAPA.

Are CAPAs opened from complaints included?

No. The complaint handling team verifies those outside this queue, and complaint records stay unread by the agents.

Quality & Labeling

Other quality system work for device plants

All 4 solutions →
Nonconformance (NCMR) Investigation and MRB Disposition

Every MRB call made on the records, not on memory, with each seat's signature.

1-min demoFDA 483 and Warning Letter Response

Every promise to FDA has an owner, a date and its proof.

1-min demoUDI Data Reconciliation (GUDID and EUDAMED vs Label)

Every database says what the label says, each fix signed before it is sent.

1-min demo

Bring one CAPA you closed
and the lots made since.

We’ll count them against its criterion and search your other plants for the same cause.

Book a demoBrowse solutions