1.0 Document Control
Owner: Meah A. | Seat: CX Specialist
Dept: 4.0 - CX & Account Health
Last Updated: 2026-04-29
Last Verified: 2026-04-29
2.0 Operational Context
Purpose: This document is the daily QA audit playbook for CAR GUYS customer support. Defines how the QA reviewer audits closed tickets for accuracy, tagging, snooze handling, and logging — and how to deliver corrective feedback to agents through the Sandwich Method coaching protocol. The rules being audited against live in CX Standards and CX Daily Operations SOP; this SOP tells the reviewer how to test for them. When this SOP and another CX document conflict on what a rule says, the source document wins.
When to use: Every workday. The recurring Trello card on the 4.0 - CX & Account Health board mirrors this SOP and is the operational tool the reviewer checks off as they audit; this document is the canonical source of truth. If the audit process changes, update this SOP first, then update the Trello card from this SOP.
Designed for AI handoff: This SOP is currently run manually by the QA reviewer. It is designed so that an AI agent can run the same audit against the same rule references in the future. When that transition happens, the AI runs the full audit and drafts all coaching messages; the human role shifts to auditing the AI's flags, editing the drafted coaching where needed, and delivering it to the agent. The audit checks in §5.0 are phrased as discrete, rule-grounded yes/no questions specifically so an AI auditor can execute them. See §6.0 for the future-state workflow.
3.0 Audit Window
The audit window depends on the day of the week:
Mondays: Audit the full weekend (Friday–Sunday).
Tuesday–Friday: Audit the previous 24 hours.
If a Monday is missed, the next available day's audit must cover Friday through the most recent closed ticket.
4.0 Escalation Protocols
The audit produces two types of findings: response errors (Protocol A) and admin errors (Protocol B). Each has a different correction path.
4.1 Protocol A — Response Error
Definition: A deviation from standard response guidelines was identified — wrong product knowledge, wrong policy application, wrong voice register, overprocessing, missed questions, or skipped discovery. Requires coaching using the Sandwich Method.
One flag = one coaching message. If a ticket has multiple Protocol A flags, each is delivered as a separate Slack message. Do not bundle multiple flags into a single coaching message — each correction needs its own clear citation and corrected response.
Action 1 (KPI Tracker): Log the error in the Customer Support Report > QA Scorecard tab.
Action 2 (Slack): Post in
#4-1_customer-experienceusing the Sandwich Method format below.
Sandwich Method coaching template:
@[Agent Name] I reviewed your response to [Customer Name]'s ticket and need to walk through a few things.
See ticket: [Link]
One thing done well: [Insert]
The Corrections: [Cite the specific rule violated — e.g., "CX Standards §6.1 Discovery Before Diagnosis" or "CX Daily Operations SOP §3.5 Tag Preservation" — and explain the error]
Corrected Response: [Paste the exact response text the agent should have used]
Coaching principle: Always cite the specific document and section the rule lives in. This trains the agent to reference the canonical source in future tickets and removes ambiguity about whether something is policy versus preference.
4.2 Protocol B — Admin Error
Definition: An administrative or data-logging error was identified — missing tags, wrong snooze duration, missing log entry, or incorrect status. Requires an immediate fix and a quick notification, not full coaching.
One flag = one notification. If a ticket has multiple Protocol B flags, each is delivered as a separate Slack message.
Action 1 (Fix the Error): Fix the error immediately in GrooveHQ or the Customer Support Report.
Action 2 (KPI Tracker): Log the error in the Customer Support Report > QA Scorecard tab.
Action 3 (Slack): Post in
#4-1_customer-experienceusing the format below.
Quick notification template:
@[Agent Name] Quick heads up on [Ticket Link] — I updated the [Tags / Snooze / Log Entry], but logging this as a QA flag. Please double-check these moving forward. Thanks!
5.0 Daily Execution Checklist
The audit runs as a single sequential checklist. For each ticket in the audit window, run the four checks below in order. Any check that fails triggers the corresponding protocol.
Grading: Pass/fail per check. A ticket can fail multiple checks; each failure triggers a separate protocol action and a separate Slack message per §4.1 / §4.2.
Ambiguity rule: If a check is genuinely ambiguous — voice register debatable, customer intent unclear, or the scenario falls outside the rules in CX Standards or CX Daily Operations SOP — flag it as AMBIGUOUS rather than passing or failing autonomously. Meah makes the final determination. Ambiguous flags only count as QA errors if Meah confirms an actual rule violation after review.
QA Scorecard: The QA Scorecard tab in the Customer Support Report tracks every flag Meah confirms as a real QA error. Target is zero confirmed flags per audit window. Any confirmed flag — Protocol A or Protocol B — counts as one error against the target.
5.1 Open the Audit Queue
Open GrooveHQ Closed Folder.
Identify your audit window per §3.0 (Mondays = Fri–Sun; Tue–Fri = previous 24 hours).
Work through tickets oldest to newest.
5.2 Run the Four Audit Checks per Ticket
For every ticket in the audit window, run all four checks in order. Do not skip checks even if the first one passes.
Check 1: Accuracy Check
Read the agent's response in full. Test against the following rules. If any sub-question fails, the Accuracy Check fails and triggers Protocol A.
Did the agent answer every distinct question the customer asked? (CX Standards §7.10 Complete Loop Rule)
Did the agent match the correct voice register to the customer's emotional state? Angry Boat for angry/grieving customers, Confident/Collaborative for damage claims and worried customers, Direct/Helpful for neutral questions. (CX Standards §4.3 Voice Register by Scenario)
Did the agent match response depth to the question? Pre-sales/shopping → concise product map. Post-sales/how-to → detailed walkthrough. Direct yes/no question → one-line answer. (CX Standards §4.4 Response Depth)
If the customer reported a product not working, did the agent apply Discovery Before Diagnosis? Asked about prep process, surface material, or other discovery questions before recommending a fix. (CX Standards §6.1 Discovery Before Diagnosis)
If the agent recommended a product, is the product compatibility claim correct? Verify against the relevant Product Data Sheet. Never accept compatibility claims from memory — Plastic Restorer is not for leather, Super Cleaner is not for glass, Wheel Cleaner is not for rust. (Product Data Sheets in 02 — Product Data Sheets)
If the customer is unhappy, did the agent apply Framework F (Angry Boat) and offer the appropriate remedy? No discount coupons in lieu of replacements (Frictionless Replacement Rule). Refund offered without fine print per Zero Risk Policy. (CX Standards §7.6 Framework F, §5.1 Zero Risk Policy)
If the customer reported damage to property, did the agent apply Framework I (Damage Claims) instead of Framework F? Stage 1 Intake template sent. No remedy committed before documentation reviewed. Confident/Collaborative voice register applied, NOT Angry Boat. (CX Standards §7.9 Framework I, §5.3 Damage Claims Policy)
If the customer asked about a product CAR GUYS doesn't sell, did the agent recommend a competitor product without gatekeeping? (CX Standards §7.3 Framework C, §4.8 Recommending Third-Party Products)
If a claim showed red flags (story inconsistencies, documentation gaps, pattern flags), did the agent run the Claim Review Protocol before committing to a remedy? (CX Claim Review Protocol)
Did the agent avoid prohibited language? No corporate apologies ("we apologize for any inconvenience"), no "results may vary," no "I would feel terrible if," no specific dates or "coming soon" for unreleased products. (CX Standards §4.5 Language Rules)
Check 2: Tagging Check
Verify the tags applied to the ticket. If any sub-question fails, the Tagging Check fails and triggers Protocol B.
Are the correct tags applied for the ticket's outcome? Replacement sent →
Temperature Check. Waiting on customer info →Need info. Review requested →Asked for review. Damage claim →Product is Damaged / Leaking. (CX Daily Operations SOP §5.3 Finalize & Log Ticket)Have any historical tags been deleted? Historical tags are permanent breadcrumbs and must never be removed. (CX Daily Operations SOP §3.5 Tag Preservation)
Check 3: Snooze Check
Verify the snooze duration matches the ticket state. If any sub-question fails, the Snooze Check fails and triggers Protocol B.
If a replacement was sent, is the ticket snoozed for 10 days with
Temperature Checktag? (CX Daily Operations SOP §3.4 Temperature Check Protocol)If waiting on customer info, is the ticket snoozed for 48 hours with
Need infotag? (CX Daily Operations SOP §5.3)If the ticket was a second ghost (last message was a nudge or Temp Check and customer didn't respond), was the ticket closed instead of snoozed again? One Nudge Limit applies. (CX Daily Operations SOP §3.2 One Nudge Limit)
Check 4: Log Check
Open the Customer Support Report. If any sub-question fails, the Log Check fails and triggers Protocol B.
Is the ticket logged in the Customer Support Report? Every GrooveHQ ticket must be logged. No exceptions. (CX Daily Operations SOP §3.6 Mandatory Logging)
Does the log entry's Status (Open / Closed) match the actual ticket state in GrooveHQ?
Are all required fields populated? Complaint date, ASIN, order ID, customer name, email, complaint, quantity, category, corrective action, status, Last Follow-up Sent (if applicable), Date Closed (if applicable), Result (if applicable), GrooveHQ Rating (if any). (CX Daily Operations SOP §5.3 Finalize & Log Ticket)
5.3 Wrap-Up
IF ALL TICKETS PASS ALL CHECKS: Mark the Trello card as Complete.
IF ANY TICKET FAILED ANY CHECK: Confirm Protocol A or Protocol B was executed for each failure (KPI Scorecard logged + Slack notification sent). Then mark the Trello card as Complete.
5.4 Worked Example
The example below shows how the four checks apply to a real audit. Use it as a reference when interpreting the rules in ambiguous situations.
Ticket: #69505 — Bernardo asks: "What product should I use to clean and protect the faux leather sections of my dashboard, and the faded grey plastic trim around it?"
Agent's response: Recommended Plastic Restorer for both the faux leather AND the plastic trim. Provided the full Plastic Restorer application walkthrough (clean → apply → 10–30 min dwell → buff). Tagged the ticket appropriately, snoozed 48 hours, logged in the Customer Support Report.
Audit results:
Check 1 — Accuracy Check: FAIL. Two failures:
Plastic Restorer is not designed for faux leather (Product Data Sheet verification, CX Standards §7.1 Framework A). The agent should have stated CAR GUYS does not carry a faux leather protectant and recommended a competitor product.
Customer asked about two surfaces; agent only fully addressed one with a usage walkthrough. Faux leather was answered with the wrong product, not with a competitor recommendation. (CX Standards §7.10 Complete Loop Rule)
Additional issue: this was a pre-sales / shopping question. Customer asked "what product should I use" — agent responded with a full Post-Sales / How-To walkthrough. Wrong response depth. (CX Standards §4.4 Response Depth)
Triggers Protocol A — three separate flags, three separate Slack messages.
Check 2 — Tagging Check: PASS. Tags were applied correctly.
Check 3 — Snooze Check: PASS. 48-hour snooze with
Need infotag matches the ticket state.Check 4 — Log Check: PASS. Ticket is logged in the Customer Support Report with all required fields.
Outcome: Three Protocol A flags delivered as three separate Slack messages, each with its own rule citation and Corrected Response. All three count toward the QA Scorecard.
6.0 AI Audit Mode (Future State)
This SOP is designed to be runnable by an AI agent. The four checks in §5.2 are written as discrete, rule-grounded yes/no questions specifically so an AI auditor can execute them against the canonical reference set without ambiguity.
When AI audit mode is active, the daily workflow is:
AI runs the audit. Loads CX Standards, CX Daily Operations SOP, CX Claim Review Protocol, CX AI Routing Map, and relevant Product Data Sheets as its reference set. Pulls the audit window's closed tickets from GrooveHQ.
AI executes Checks 1–4 against each ticket. Marks each check pass, fail, or AMBIGUOUS. Documents the specific rule citation for every failure. For AMBIGUOUS markings, the AI documents what made the check ambiguous so Meah can quickly assess.
AI drafts the coaching message for every flagged ticket. For Protocol A failures, the AI drafts the full Sandwich Method message per §4.1, including:
One thing done well — a genuine positive observation from the agent's response
The Corrections — the specific rule violated with section citation, plus a clear explanation of the error
Corrected Response — the full reply the agent should have sent, written in CAR GUYS voice and grounded in the relevant rules For Protocol B failures, the AI drafts the quick notification message per §4.2.
AI compiles a daily QA Report for Meah. The report contains every flagged ticket with: ticket link, failed check(s), rule citation, AI's reasoning for the flag, and the drafted coaching message ready for Meah's review.
Meah audits the AI's work. She reviews each flagged ticket, confirms the flag is valid (or overrides if the AI got it wrong), edits the drafted coaching message if it needs adjustment, and delivers it via Slack.
Meah logs verified flags in the QA Scorecard and marks the Trello card as Complete.
Critical AI grounding rule for product compatibility claims: When auditing the Accuracy Check, the AI MUST verify any product compatibility claim in the agent's response against the relevant Product Data Sheet. Never accept compatibility claims from the AI's own training or memory — always cross-reference the data sheet. This is the safeguard that catches errors like recommending Plastic Restorer for leather or Super Cleaner for glass.
What the AI does:
Runs Checks 1–4 against every ticket in the audit window
Cites the specific rule violated for every failure
Drafts the full Sandwich Method coaching message for every Protocol A flag, including the Corrected Response in CAR GUYS voice
Drafts the quick notification message for every Protocol B flag
Compiles all flagged tickets into a daily QA Report for Meah
What the AI does NOT do:
The AI does not deliver coaching directly to the agent. Meah is the final touchpoint who reviews, edits, and delivers.
The AI does not modify tickets, tags, snoozes, or log entries. It flags issues; the human fixes them.
The AI does not log to the QA Scorecard. Meah logs verified flags after auditing the AI's report.
The AI does not override the Sandwich Method format or the Quick Notification format. Drafts always follow the §4.1 and §4.2 templates exactly.
Meah's role in AI audit mode: Meah's role shifts from running the audit to auditing the AI. She verifies the AI's flags are valid, confirms the rule citations are accurate, reviews the drafted coaching messages for tone and accuracy, edits where needed, and delivers the final coaching via Slack. The AI does the heavy lift; Meah is the editor and the voice that talks to the agent.