E-commerce
July 1, 2026
"Why is my order blocked without explanation?" "Was my refund rejected by a robot?" "I want a human to review this decision." Three tickets where an automated decision frustrates the customer due to a lack of clear support response.
The e-commerce automated decision support covers anti-fraud filters, automatic refund rejections, dynamic pricing, account restrictions, and the GDPR right to explanation. It complements AI recommendation consent (#861) and personalization errors (#863): here, the customer challenges a decision produced by a system without any visible human involvement.
This guide #865 deploys policy AUTODEC-SUP, flow AD-1 to AD-8, and matrix AUTODEC-MAP. Future CS paired with explainability bot (#866).
Summary
Why do automated decisions generate tickets?
Anti-fraud filters, refund engines, loyalty scoring, and logistics rules make decisions in milliseconds. The customer sees the result (order pending, refund refused, limited account) without understanding the mechanism. The agent promises a callback or redirects to a generic FAQ with no right to an explanation or human review procedure.
Five typical frictions in automated decisions
Blocked order: fraud filter or payment risk without a clear message
Refund refused: auto rule (delay, product condition) perceived as arbitrary
Different price or promo: personalized price or targeted offer misunderstood
Restricted account: purchase limit or access restricted without prior notification
Human review requested: customer invokes the right not to be subject solely to an automated decision
GDPR Article 22 regulates decisions based solely on automated processing that produce legal or significant effects (CNIL, Article 22).
Marketplace example
D2C Decoration, 14 autodec_ tickets/month. After AUTODEC-MAP: autodec_explain_resolution_rate 88%, unnecessary DPO escalations -35%.
AUTODEC #865 vs AIRECOCON #861, AUTOPERS #863, AIGENCON #859 and bot #866
Five personalized AI contents, five distinct client-side angles.
Quick matrix
#865 AUTODEC: explain and review system decisions (fraud, refund, account limit)
AIRECOCON #861: consent AI recommendations opt-in opt-out
AUTOPERS #863: personalization error incorrect reco
AIGENCON #859: AI-generated content description sheets
AUTO-CANCEL #769: auto-cancellation order stock payment
#865 = why did the system refuse or block? #861 = did I consent to AI suggestions.
Promise #865
Policy AUTODEC-SUP, tree AUTODEC-GATE, 8 macros, auto decision vs personal error vs consent matrix, KPI autodec_explain_resolution_rate.
Which autodec_* typologies should be classified?
Action-oriented classifier: fraud dispute ≠ refund refusal ≠ Article 22 request.
Eight AUTODEC-MAP typologies
autodec_fraud_hold: order or payment blocked by risk filter
autodec_refund_denied_auto: refund refused by automatic rule
autodec_pricing_auto: promotional rate or dynamic pricing disputed
autodec_account_restriction: account limited or suspended by scoring
autodec_shipping_auto: carrier or method imposed by logistical rule
autodec_human_review: explicit request for human review
autodec_why_decision: wants to understand the factors behind the decision
autodec_gdpr_article22: right to object to an exclusively automated decision
AUTODEC-SUP Policy: Agent Rules and Escalation
The AUTODEC-SUP policy sets out a readable explanation without disclosing anti-fraud secrets or promising systematic review outside of the legal framework.
Six AUTODEC-SUP rules
EXPLAIN-FIRST: AUTODEC-EXPLAIN macro decision category and next step
Generic factors: AUTODEC-FACTORS without proprietary algorithm
Human review if eligible: AUTODEC-HUMAN-REVIEW according to internal registry
Not solely automated: AUTODEC-NOT-SOLELY if mixed human + system decision
Agent limits: AUTODEC-LIMITS when ops or legal escalation is required
Article 22 → DPO: documented formal autodec_gdpr_article22 handoff
Decision matrix (agent)
Fraud / risk: EXPLAIN + HUMAN-REVIEW if policy allows
Automated refund: FACTORS + reference to return policy, no standalone agent override
Dynamic pricing: EXPLAIN promo scope, commercial handoff if display error occurs
Personal recommendation error: AUTOPERS handoff #863 if bad product suggestion occurs
Flow AD-1 to AD-8: standard resolution
Eight sequential steps, SLA P2 order blocking < 12 h, escalate DPO if article22.
Flow AD-1 to AD-8
AD-1 Triage: read contestation, autodec_* tag, blocking or refusal?
AD-2 Lookup: auto decision log, order status, triggered rule
AD-3 Educate: EXPLAIN category without technical jargon
AD-4 Classify: autodec_* via AUTODEC-MAP
AD-5 Execute: HUMAN-REVIEW, FACTORS, handoff ops #769 or AUTOPERS #863
AD-6 Confirm: macro AUTODEC-DONE delay and expected outcome
AD-7 Test: client understands why and what follows
AD-8 Close: KPI autodec_explain_resolution_rate
Eight AUTODEC-* macros ready to paste
Educational macros aligned with the internal automated decisions registry.
AUTODEC-* Library
AUTODEC-EXPLAIN: "Your {{élément}} has been processed by our automatic verification system for {{motif_catégorie}}."
AUTODEC-FACTORS: "Factors taken into account include: order history, payment method, return times. Full details upon written request."
AUTODEC-HUMAN-REVIEW: "We are forwarding your file to an advisor for review within {{délai}}."
AUTODEC-NOT-SOLELY: "This decision combines automatic verification and human control according to our procedures."
AUTODEC-LIMITS: "I cannot modify this rule from support. Escalation to {{équipe}} in progress."
AUTODEC-GDPR: "To exercise your rights regarding an automated decision, contact privacy@{{domaine}} with ticket reference {{id}}."
AUTODEC-ESCALATE: "File forwarded to {{ops}}. You will receive a response within {{SLA}}."
AUTODEC-DONE: "Recap: {{action}}. Next step {{suite}}. Reply if the status does not update within {{délai}}."
AUTODEC-GATE tree and agent-ready decision registry
Decision tree before override promise or unjustified AUTOPERS handoff.
AUTODEC-GATE
Order blocked for fraud? → EXPLAIN + HUMAN-REVIEW if eligible
Refund automatically refused? → FACTORS + returns policy, no agent-only refund
Promo price misunderstood? → EXPLAIN scope, do not AIRECOCON #861
Product recommendation incorrect? → AUTOPERS handoff #863
Formal Article 22 request? → GDPR + DPO
Minimum Internal Register
Document helpdesk: automated decision types, human review SLA, authorized lookup fields, DPO procedure. Train agents: automated decision ≠ personalization error #863 ≠ stock cancellation #769.
KPI, QA and handoff to bot #866
Measuring AUTODEC detects under-explanation and illegitimate promises of immediate unblocking.
Four AUTODEC KPIs
autodec_explain_resolution_rate: customer understands decision / total
autodec_human_review_rate: % autodec_human_review with HUMAN-REVIEW tracked
autodec_wrong_silo_rate: % wrongly routed to #863 or #861 alone
autodec_repeat_7d: same auto decision dispute within 7 days
Bot #866 handoff
Export AUTODEC-MAP to intents bot_autodec_why, bot_autodec_human_review. AUTODEC-NO-OVERRIDE-BOT guardrail: explain without promising an unblocking that the agent cannot uphold.
Edge cases: guest, mixed decision, dynamic pricing
Three cases outside the standard flow.
Guest without an account
Lookup by order email. EXPLAIN + GDPR if rights requested. No account scoring visible.
Mixed human + system decision
NOT-SOLELY before HUMAN-REVIEW. Do not lead to believe that only a robot decided if ops validated.
Dynamic pricing or targeted promo
Generic FACTORS. Distinction between commercial decision price vs display error. Commercial handoff if pricing bug.
Agent training: 20 minutes AUTODEC
Module: auto decision ≠ AI reco, EXPLAIN before excuse, LIMITS when no override, AUTOPERS if bad suggestion.
Exercises
Ticket A: order blocked → EXPLAIN + HUMAN-REVIEW, no promised unblocking
Ticket B: refund refused → FACTORS + return policy
Ticket C: bad product reco → handoff #863 not AUTODEC alone
How Qstomy structures AUTODEC in your stack
Qstomy route autodec_*, displays readable agent decision log and handoff #866 for tier 1 explanation without jargon.
Three building blocks
Routing: intent automated_decision vs autopers_error vs airecocon_consent
Decision log: sync macros AUTODEC-* status and category reason
Bot #866: explain factors without illegitimate override
Scenario: electronics retailer, 12 tickets/month auto dispute. Bot #866 resolves autodec_why_decision, agents handle fraud_hold and article22. autodec_explain_resolution_rate increases from 62% to 86% in 6 weeks.
FAQ and AUTODEC deployment checklist
FAQ
Can the agent unblock any fraud order?
No. HUMAN-REVIEW according to policy. LIMITS if out of agent's scope.
Difference with #863?
#865 = system decision (blocking, refusal). #863 = personalization recommendation or profile error.
Difference with #861?
#865 = contestation of auto decision. #861 = AI recommendation engine consent.
7-Day Checklist
D1: AUTODEC-SUP + AUTODEC-MAP + auto decisions registry
D2: 8 helpdesk macros
D3: routing matrix #863 #861 #769
D4: 20-min agent training
D5: autodec_* tags + KPIs
D6: HUMAN-REVIEW vs LIMITS test
D7: brief bot #866 NO-OVERRIDE-GATE
Interlinking
Enzo
July 1, 2026
