English · 中文 · Framework · Case Study

GMV Max Diagnostic Framework

A data-driven, causal-analysis framework for diagnosing TikTok Shop GMV Max campaigns before changing budget, ROAS, creatives, or campaign structure.

Built from 4 months of operating a real TikTok Shop account and distilled into a reusable diagnosis workflow.

Use this when

Symptom	Bad default reaction	Better diagnosis path
$500 spent, no orders	Kill the campaign	Check timeline, spend ramp, traffic quality, and creative hook
ROI suddenly dropped	Change ROAS bid	Separate bid effect, campaign interference, and operational events
One video took off	Copy the creative blindly	Test whether hook rate, CVR, or audience selection caused the lift
Multiple campaigns fight each other	Add more campaigns	Diagnose cross-campaign interference first

Quick path

Read the 4-step diagnosis framework.
Compare it with the portable fans case study.
Export Campaign x Day data from TikTok Shop Ads Manager.
Rebuild the timeline before writing any prescription.

What is this

Most people running GMV Max:

$500 spent, no orders → add budget? kill it? swap creative? Gut feeling.
ROI dropped → adjust ROAS or change creator? Gut feeling.
A video took off → was it the creative or the ROAS bid? Gut feeling.

After 4 months in the trenches, I codified the decision process into:

4-step diagnosis (Step 0–3): Take temperature → Rebuild timeline → Systematic decomposition → Write prescription
4-layer decomposition (Layer ABCD): Single-campaign stability / Cross-campaign interference / Operational event causality / Creative effectiveness

Core principles:

The nature of GMV Max: You make decisions, TikTok executes.

The nature of diagnosis: Explain changes, don't evaluate states.

The nature of analysis: Use time to separate causality, not snapshots for correlation.

Counterintuitive Finding (anchored in 4 months of data)

ROAS bid doesn't improve creatives — it picks higher-value users for you.

Same video, across 2 campaigns:

Campaign	ROAS	ROI	Hook Rate	CVR
Campaign A	12	12.5x	51%	9.8%
Campaign B	3.2	1.8x	57%	12.8%

The ROAS=12 campaign delivered 7x higher ROI — but Hook Rate and CVR were actually higher in the low-ROAS campaign.

→ ROAS bid doesn't improve creative performance to lift ROI. It lifts ROI by letting the system select higher-LTV users.

Implication: Hook Rate is an intrinsic creative property (stable across campaigns) → optimize creatives to lift Hook. CVR is controlled by ROAS bid → optimize bidding to lift CVR. Two independent levers — don't conflate them.

Framework Documentation

Full 4-step diagnosis + Layer ABCD framework (English)
Anonymized case: portable fans shop 4-month retro (English)
完整框架 (中文)
脱敏案例 (中文)

Practical resources from SCTCamp

GMV Max not spending checks campaign status, target ROI, and budget use before you change settings.
Daily review checklist shows how to record spend, orders, and decisions in a review sheet.
Online diagnostic report demo lets you inspect the report and sample campaign export without an account.

How to use this

You're an individual seller / small team

Read frameworks/ads-diagnosis.md and walk through Step 0–3 on your own account.

You'll need:

Your GMV Max Campaign × Day data (export from TikTok Shop Ads Manager)
Your operations log (when you built/adjusted/killed campaigns)
Excel or a Python notebook

You want to run it with Claude Code / Cursor

The framework is designed for AI agents. Clone this repo, point Claude Code at frameworks/ads-diagnosis.md, feed it your data — it can output diagnostic reports straight from Step 0–3.

(A turnkey automation Skill is in the works — will be open-sourced as a separate repo.)

You want deeper methodology / 1-on-1 account audit

Paid consulting / full SOP package / monthly enterprise retainer — contact below.

About me

Education: TJU + NYU CS
Big-tech product background: Former PM at a major Chinese internet company (ad attribution + monetization systems)
Cross-border operations: TikTok Shop US top-category operator + Amazon / DTC sites / US-domestic brand management + 4 months consulting on a friend's GMV Max account
Founder experience: Co-founder, CastarAI / AIvocado
Currently: AI builder × global expansion community × consulting × enterprise AI transformation

Last year I publicly introduced the GMV Max AB-chain strategy on Douyin (300k+ views). This year, I'm releasing this automated diagnostic framework.

Contact

Douyin (China): @SCT出海Scott (scotti1i)
YouTube: @ScottGlobalAI
WeChat: scotti1i
Email: [email protected]

License

MIT

gmv-max-diagnosis-framework