E-commerce
July 1, 2026
"Are you reading my messages?" "Why was my conversation reviewed?" "Is it anonymous or is my name visible?" Three tickets where the conversation quality monitoring lacks a clear response.
The e-commerce quality review support explains sampling, service improvement purpose, internal access, and confidentiality, distinct from recording (#899), training data (#907), and internal QA review (#277).
This guide #911 deploys policy CONVQMON-SUP, flow QM-1 to QM-8, and matrix CONVQMON-MAP. Pair of target customer service quality review bot (#912).
Summary
Why does the quality review generate tickets?
Customer discovers the mention "conversation reviewed" or fears permanent listening. Agent replies "we do not read anything" when a QA sample exists. Without CONVQMON-MAP, confusion with convrec_ #899 or traindata_ #907.
Five typical quality review frictions
Perceived surveillance: customer believes real-time listening is occurring
Opaque sampling: who is selected and why
Vague internal access: who reads transcripts
Doubtful anonymization: order name visible
Training confusion: QA review ≠ AI learning
DTC retail example
DTC Fashion, 5 convqmon_ tickets/month. After CONVQMON-MAP: convqmon_trust_resolution_rate 89 %, privacy escalations -44 %.
CONVQMON #911 vs CONVREC #899, TRAINDATA #907, QA #277 and bot #912
Seven quality privacy contents, seven distinct angles.
Quick Matrix
#911 CONVQMON: explain quality review sampling privacy client-side
CONVREC #899: separate retention recording QA review
TRAINDATA #907: separate training data quality sample
CONVOPT #909: separate training opt-out internal review
Audit #259: monthly Voice of Customer audit separate monitoring issue
Bot #912: explain quality review widget-side
#277 = internal process. #911 = replying to the customer who asks why we read.
Promise #911
Policy CONVQMON-SUP, CONVQMON-GATE tree, 8 macros, quality review register, KPI convqmon_trust_resolution_rate.
Which typologies should we classify?
Action-oriented classifier: surveillance ≠ sampling ≠ access ≠ training overlap.
Eight CONVQMON-MAP typologies
convqmon_surveillance_fear: fear of permanent real-time listening
convqmon_why_reviewed: why my conversation was reviewed
convqmon_sampling_question: how sampling works
convqmon_who_accesses: who reads transcripts internally
convqmon_anonymization_ask: identifying data visible or not
convqmon_vs_training: QA review distinct from training #907
convqmon_opt_out_overlap: opt-out link #909 if applicable
convqmon_escalate_dpo: formal GDPR request
Policy CONVQMON-SUP: agent rules and review registry and review
The CONVQMON-SUP policy sets answers from the quality review registry, neither denial nor over-promising.
Six CONVQMON-SUP rules
REGISTRY-FIRST: CONVQMON macro from the review registry
Sample honesty: do not deny QA if the process is documented
Clear purpose: service improvement, not marketing
Limited access: documented QA lead support roles
Distinct training: QA review ≠ traindata handoff #907
Formal GDPR → DPO: convqmon_escalate_dpo
Minimum quality review registry
Purpose: response quality friction detection
Sampling: % or selection criteria
Access: authorized roles mandatory training
Anonymization: identifier masking rules
Exclusions: opt-out #909 sensitive conversations
Flow QM-1 to QM-8: Quality review questions processing
Eight sequential steps, SLA P3 convqmon < 72 h, escalate to DPO if escalate_dpo.
Flow QM-1 to QM-8
QM-1 Triage: quality review vs recording #899 vs training #907?
QM-2 Classify: convqmon_* via CONVQMON-MAP
QM-3 Registry: purpose lookup sampling access anonymization
QM-4 Explain: macro CONVQMON honest scope
QM-5 Distinguish: vs training vs recording vs opt-out
QM-6 Specific case: why_reviewed if ticket identified
QM-7 Escalate: DPO or opt-out #909 if overlap
QM-8 Close: KPI convqmon_trust_resolution_rate + brief #912
Eight ready-to-paste CONVQMON-* macros
Macros aligned with purpose register, sampling, and access.
CONVQMON-* Library
CONVQMON-ACKNOWLEDGE: "We understand your question regarding the quality review."
CONVQMON-PURPOSE: "Purpose: {{finalité}}. No resale or marketing."
CONVQMON-SAMPLING: "Sampling: {{méthode}}. No real-time listening."
CONVQMON-ACCESS: "Access limited to: {{rôles}}. Mandatory training."
CONVQMON-ANON: "Anonymization: {{règles}}. Identifiers masked if {{cas}}."
CONVQMON-VS-TRAINING: "Quality review is separate from AI training. Training details: #907."
CONVQMON-WHY-REVIEWED: "Your conversation: {{raison}}. Context: {{finalité}}."
CONVQMON-DONE: "Recap: {{question}}. Response: {{résolution}}. Reference: {{id}}."
CONVQMON-GATE tree and why_reviewed cases
Decision tree before denying review or over-explaining individual cases.
CONVQMON-GATE
Recording retention question ? → handoff CONVREC #899
AI training question ? → handoff TRAINDATA #907
Contribution refusal ? → handoff CONVOPT #909
General surveillance question ? → PURPOSE SAMPLING ACCESS ANON
Specific case identified ? → WHY-REVIEWED from QA registry
Formal GDPR ? → ESCALATE-DPO
Documented why_reviewed case
Typical reasons: random sample, friction tag, escalation, low CSAT, bot audit. Do not make up a reason without looking up the internal QA ticket.
KPI, QA and handoff to bot #912
Measuring CONVQMON detects review denial and training confusion.
Four CONVQMON KPIs
convqmon_trust_resolution_rate: review questions resolved without privacy escalation
convqmon_registry_compliance: % of responses aligned with the review registry
convqmon_deny_review_rate: documented QA process denial target low
convqmon_training_misroute_rate: training confusion target low
Handoff #912
Export CONVQMON-MAP to bot: convqmon_surveillance_fear convqmon_sampling_question priority. Guardrail CONVQMON-REGISTRY-GATE brief #912 copy registry widget.
Edge cases: customer notification, bot only, sensitive data
Three cases outside the standard flow.
Client notified « conversation reviewed »
WHY-REVIEWED mandatory. Verify legitimate notification before explaining modern reason.
Bot-only conversation
Specify bot review + human handoff if applicable. Link bot audit #143 if accuracy is questioned.
Sensitive topic: health claim
Restricted access documented. Escalate to DPO if sensitive data and access is contested.
Agent training: 20 minutes CONVQMON
Module: PURPOSE SAMPLING ACCESS, distinguishing #899 #907 #277 #912.
Exercises
Ticket A: "do you read everything?" → SAMPLING not real-time
Ticket B: "are you training on me?" → handoff TRAINDATA #907
Ticket C: "why review?" → WHY-REVIEWED lookup QA
How Qstomy structures CONVQMON in your stack
Qstomy route convqmon_*, sync quality review registry, PURPOSE SAMPLING and handoff macros #912 registry gate.
Three building blocks
Routing: intent quality_review vs conv_recording vs traindata
Review registry: target sampling access anonymization
Bot #912: tier 1 monitoring sampling on the widget side
Scenario: DTC, 5 tickets/month convqmon. REGISTRY-FIRST agents, bot #912 tier 1. convqmon_trust_resolution_rate goes from 71% to 90% in 5 weeks.
FAQ and deployment checklist for CONVQMON
FAQ
Deny the quality review?
No if process is documented. REGISTRY-FIRST honesty.
Difference from #277?
#277 = internal scoring ritual. #911 = responding to the client.
Difference from #907?
#907 = training data. #911 = service quality review.
Difference from #912?
#911 = agents. #912 = bot explaining the widget review.
7-day Checklist
D1: CONVQMON-SUP + CONVQMON-MAP + review registry
D2: 8 helpdesk macros
D3: routing matrix #899 #907 #909 #277
D4: 20 min agent training PURPOSE SAMPLING
D5: convqmon_* tags + KPI
D6: monitoring vs training vs why_reviewed test
D7: bot brief #912 REGISTRY-GATE
Interlinking
Enzo
July 1, 2026
