E-commerce
June 28, 2026
“They were promised a 15% discount.” “The previous agent told me it was OK even after the deadline.” “Finance refuses the refund, nobody can find any trace of the gesture.” Three tickets, zero internal notes, and a margin that evaporates because every exception has become a non-reproducible oral rule.
GetVocal points out that non-negotiable rules must exist as deterministic controls with auditable logs (session, version policy, reason code), not as forgotten Slack instructions (GetVocal, decision maps 2026). Heeya insists: the quality of internal documentation directly determines the quality of bot and agent responses (Heeya, automation SAV 2026).
This guide #237 addresses the documentation of support decisions: exceptions, commercial gestures, policy deviations. No existing content covered this area. Distinct from response base (#102) (phrasing) and AI review (#210) (bot validation): here, trace every human exception with approval and reason code.
Summary
Why do undocumented exceptions cost so much?
An untracked support exception recurs at random, amplifies among agents, and becomes a permanent customer expectation.
Three leakage mechanisms
Oral precedent: "last time we accepted" without a written policy
Double gesture: chat discount + auto credit note without a ticket link
Impossible audit: finance disputes, support has no proof
Typical DTC cost
Brand with 2,500 tickets/month, 8% with untagged gesture or waiver: estimated extra cost of 3 to 7% of support margin (duplicate refunds, returns outside of policy, customers who systematically renegotiate). Structured documentation reduces the unapproved exception rate to under 2%.
Goal #237
DEC-* registry, reason codes, approval workflow, monthly review, sync bot guardrails.
How does it differ from the answer base, AI governance, and commercial gestures?
Four registers, four roles in support ops.
Response database (#102)
REP database (#102): standard formulations and escalation criteria. The #237 tracks what comes out of the file on a case-by-case basis.
AI Governance (#142) and proofreading (#210)
AI Governance (#142): bot framework. Proofreading (#210): auto-response validation. The #237 covers human decisions and agent exceptions.
Escalation matrix (#193)
Escalation matrix (#193): third parties and ceilings. The #237 records when a third party has authorized a waiver.
Commercial gestures (#238)
The #238 will address when to offer a discount, credit note, or gift. The #237: how to document each gesture granted or refused.
What to document: policy, decision and exception?
Three levels not to be confused in your after-sales decision registry.
Policy (standard rule)
30-day return, refund within 5-10 days, customer return shipping. Canonical source: T&Cs page + REP sheet. Agent Red recommends one policy per paragraph, definitive language "we do / we do not" (Agent Red, business policies).
Decision (standard applied case)
Compliant return accepted, tracking sent, refund initiated. Ticket traceability + macro is sufficient.
Exception (policy derogation)
D+38 return accepted, 10% discount excluding promo, partial refund without product return. Obligation: DEC sheet + reason code + approver if threshold is exceeded.
Simple rule
If an agent hesitates "can I?", it's an exception to be documented. If the REP sheet answers unambiguously, it's a standard decision.
What structure is used for a DEC-* decision sheet?
Each DEC exception support file follows the same template, ticket or Notion.
Required fields
DEC-ID: DEC-2026-00482 (year + sequential)
Ticket / order: #T-8842, #ORD-55102
Reason code: EXC-RET-LATE, EXC-REF-PARTIAL (section 5)
Waived policy: REP-RET-003 30-day return
Decision made: return accepted D+38, free shipping
Impact amount: €47 excl. tax margin
Customer justification: 1 factual sentence
Agent + approver: name, tier, date/time
Internal helpdesk note (template)
"DEC-2026-00482 | EXC-RET-LATE | Policy REP-RET-003 | Decision: return accepted D+38, free return shipping €8.90 | Impact €47 | Reason: package received D+32, loyal customer 4 orders | Approved: [Supervisor], [date]."
Where to store
Gorgias/Zendesk internal note + Notion/Airtable registry line exportable to finance. A single source of truth for monthly stats.
What reason codes should be adopted for exceptions and goodwill gestures?
After-sales exception reason codes allow reporting without reviewing 400 tickets.
Recommended Taxonomy (12 MVP codes)
EXC-RET-LATE: late return policy exception
EXC-RET-WORN: used product return (discretionary exception)
EXC-REF-PARTIAL: partial refund without full return
EXC-REF-GOODWILL: goodwill gesture without proven fault
EXC-SHIP-FREE: free shipping or free resend
EXC-SHIP-EXPEDITE: complimentary shipping upgrade
EXC-DISC-PCT: % discount outside of public promotions
EXC-CREDIT-STORE: exceptional store credit
EXC-REPL-FREE: product resend without billing
EXC-WARR-EXT: warranty extension
EXC-OPS-ERROR: proven warehouse / ops error
EXC-VIP: high-value customer exception
Mirror helpdesk tags
Ticket tag exception_[code] in addition to the DEC note. Dashboard: top 5 codes per month = policy signal to clarify or product to correct.
What approval workflow is required before granting an exception?
The exception approval workflow aligns agent limits and audit trails.
Thresholds per tier (DTC example)
Tier 1: gesture < €15 or return postage only, auto-approved if EXC-OPS-ERROR is proven
Tier 2: €15-80, supervisor approval in helpdesk
Tier 3: > €80 or EXC-REF-PARTIAL, lead + mandatory DEC note
HITL Pattern
Microsoft Agent Governance Toolkit describes the require_approval pattern: pause action, human approves or rejects, immutable log (Microsoft AGT, approval 2026). In human helpdesk: ticket status "pending_approval" + @mention supervisor.
Documented rejection
A rejection is also tracked: DEC-REFUSED + reason "out of policy, alternative proposed: 5% credit on next purchase". Avoids the customer quoting another agent.
How to integrate documentation into Gorgias, Notion, and the bot?
The decision documentation must live where the agent works, not in some forgotten Google Doc.
Helpdesk (Gorgias / Zendesk)
Internal macro DEC-TEMPLATE pre-fills the fields. Rule: no Shopify Admin refund > threshold without a DEC note + exception tag. Shopify Flow: exception_logged tag if custom field is filled.
Notion / Airtable Registry
Finance view: month, code, amount, agent, approver, SKU. Monthly CSV export for accounting. Bidirectional link ticket ID ↔ DEC-ID.
Bot and guardrails
The bot never promises any exception not listed in the policy. GetVocal: refund ceilings in deterministic nodes, not in LLM prompt alone. See AI governance (#142). Bot corpus = standard policies; DEC registry = humans only.
Sync REP → DEC
When an exception occurs 5×/month (same code), update the REP sheet or public policy, rather than multiplying ad hoc DECs.
Which monthly review to convert exceptions into rules?
The monthly exceptions review transforms the log into continuous improvement.
60 min ritual (4 participants)
Support lead: top 5 reason codes, volume, total amount
Ops: EXC-OPS-ERROR root cause, warehouse action
Product: recurring SKUs EXC-RET-WORN or EXC-REF-PARTIAL
Finance: exceptions > threshold, double-action anomalies
Post-review decisions
Policy update: frequent code → new REP rule
Training: agent with abnormal exception rate
Adjusted ceiling: tier 1 too low or too high
Product: PDP sheet, packaging, sizing
Review KPIs
Objective: reduce the volume of exceptions for the same codes by 10% to 20% for two consecutive months through a clarified policy, not through systematic customer refusal.
What edge cases and documentation errors should be avoided?
Five customer service documentation anti-patterns cost more than having no register at all.
Exception in the customer chat
Promising a discount in public without an internal note = customer quotes the chat, next agent denies it.
DEC without approver
Tier 1 grants €60 "to calm things down": financial audit impossible, toxic precedent.
Double gesture
Shopify discount + manual promo code on the same ticket: link DEC to a single gesture, check order history.
Catch-all code
EXC-OTHER for 40% of cases: useless for reporting. Force choice among 12 codes or create an explicit 13th one.
Dead register
Notion never consulted: monthly review cancelled three times = return to oral exceptions. Assign register owner as for the REP database (#102).
Which KPIs should be used to manage support exceptions?
Measure the DEC registry as a controllable cost center.
Volume and Cost KPIs
Exceptions / tickets: target < 8%, alert > 12%
Average amount per EXC-* by code
DEC without required approver: target 0
Repeat exception same customer 90 days: signal of abuse or unclear policy
Quality KPIs
Ticket reopen post-action: customer disputes the DEC
Codes → policy updates: conversion into a written rule
DEC entry delay: < 24 hours after action
Dashboard
Columns: DEC-ID, code, amount, agent, approver, SKU, reopen Y/N. See support dashboard.
How does Qstomy integrate into the decisions registry?
Qstomy applies standard policy in bot mode and escalates out-of-scope requests with structured context.
Documentation features
Intent exception_request: routes to human, no auto-resolution
Payload handoff: cited policy, requested amount, order history
Bot limit: no refund/discount beyond tier 0
Suggestion reason code for agents from verbatim
DEC export to Notion/Airtable webhook post-approval
Quantified DTC scenario
Home decor brand, 11% of tickets with resolution, 34% without DEC note prior to project. DEC template + 12 codes + tier thresholds + monthly review. After 10 weeks: DEC entered 96%, amount of unapproved exceptions −58%, post-resolution reopens −31%, 3 codes converted into EPR policy.
Explore Shopify integration, AI customer support, request a demo.
Which operational playbooks should be launched this week?
Playbook 1: DEC template (3 h)
Create Gorgias DEC-TEMPLATE macro + Notion section 4 columns. Team briefing 15 min.
Playbook 2: reason codes (2 h)
Adopt 12 section 5 codes, helpdesk tags, disable EXC-OTHER except with mandatory note.
Playbook 3: approval thresholds (2 h)
Align tier #193 with section 6 ceilings. Document in Notion visible to agents.
Playbook 4: 30-day retroactive audit (4 h)
Scan Shopify refunds/discounts for 30 days, cross-reference tickets without DEC, estimate leakage.
Playbook 5: recurring monthly review (1 h/month)
Fixed calendar, registry owner, top codes → actions section 8.
Useful links
Documenting an exception does not slow down support: it prevents paying for the same mistake twice. When every action has a DEC-ID, a reason code, and an approver, exceptions stop being margin drains and become signals to refine your customer service rules.
Enzo
June 28, 2026
