# Sample Audit Delivery: Support Triage Workflow

Fictionalized public sample. Use the format and level of specificity, not the company details.

## Client

- Company: Sample DTC skincare store
- Segment: Shopify + TikTok Shop
- Audit owner: U-Sprint

## Executive Summary

The support inbox repeats the same order status, return policy, routine, and product-fit questions across Shopify and TikTok Shop. AI can safely draft replies from approved FAQ, shipping, return, and product documents, but refund approval, irritation/safety complaints, and angry customer cases should stay human-reviewed.

## Recommendation

Decision: Go for 999 USD/USDT pilot

Why:

- The workflow is narrow enough for a 72-hour build: triage and reply drafts for top support intents.
- The store can provide source docs and examples without sharing passwords or payment credentials.
- Human review boundaries are clear for refunds, safety complaints, and reputation-risk messages.

## Current Workflow

Inputs:

- Shopify order context
- TikTok Shop customer messages
- FAQ, return policy, shipping policy, and product pages
- Manual support replies

Current steps:

1. Customer asks about order, product fit, return, or routine.
2. Support owner checks the order or policy manually.
3. Support owner writes a reply from memory or old messages.
4. Edge cases are escalated informally.

Main bottlenecks:

- Policy-safe replies are inconsistent.
- Repeated questions are not tracked by intent.
- Product-fit and skincare claims need strict human review rules.

## Top Intents Or Work Types

| Rank | Intent / Work Type | Frequency Signal | Automation Fit | Risk |
| --- | --- | --- | --- | --- |
| 1 | Order status | Repeated shipping questions | High | Low |
| 2 | Return/refund policy explanation | Policy page + repeated tickets | Medium | Medium |
| 3 | Product routine or fit | Product catalog questions | Medium | Medium |
| 4 | Irritation or safety concern | Sensitive customer claims | Low | High |

## Safe AI Scope

AI can handle:

- Draft order-status explanations when order data is provided by the operator.
- Summarize approved return, shipping, and FAQ policy.
- Classify incoming messages by support intent.
- Draft review-mining notes for repeated objections.

Human review required:

- Refund approval or chargeback language.
- Medical, irritation, allergy, or safety-sensitive claims.
- Angry customers, public complaints, or reputation-risk replies.
- Any account, payment, or identity change request.

## Recommended 72-Hour Build

Day 1:

- Map top 5 support intents from examples.
- Lock source docs and escalation rules.
- Draft prompt pack and macro structure.

Day 2:

- Build reply-draft workflow and intent tracker.
- Create top 20 support macros.
- Add dashboard fields for source, intent, risk, owner, and next action.

Day 3:

- Test 12 representative customer scenarios.
- Tune risky cases into human-review lanes.
- Deliver SOP, launch checklist, and maintenance notes.

## Required Inputs

- 5-10 support ticket examples or screenshots.
- Current FAQ, return, shipping, and product docs.
- List of no-go claims and escalation owner.
- Exported examples or limited access, not passwords.

## Acceptance Criteria For Pilot

- Top 5 support intents classify correctly in test examples.
- AI drafts only from approved source docs.
- Refund, safety, and angry-customer cases route to human review.
- Handoff SOP shows who updates macros and source docs.

## Pilot Recommendation

Recommended next step: 999 USD/USDT pilot

Reason:

The workflow has clear examples, safe source docs, and a measurable handoff. The 299 USD/USDT audit fee is credited toward the pilot if approved within 7 days.

## Boundaries

This audit does not guarantee revenue, traffic, ad performance, platform approval, or legal/compliance outcomes. U-Sprint does not handle passwords, private keys, 2FA, or custody of client funds.