JIA JINGQIU
EN中文
CV PDF ↓
CROSS-BORDER AI GROWTH LEAD · MULTIMODAL AI PRODUCT MANAGER

I turn cross-border commerce decisions into AI products teams can use every day.

I turn hands-on cross-border commerce experience into AI products that execute the work, then apply them in live growth operations across Amazon, TikTok Shop and Shopify.

From product data, market research and selection through Listings, images and video, creators, advertising, inventory and fulfilment, community, SEO, GEO and social distribution, AI handles daily data collection, decisions, generation, execution and result capture. I own the growth goals, multimodal capability design and operating rules, plus consequential decisions such as price, budget, account permissions and publication.

My work includes 8 public sites · Amazon and TikTok Shop projects · product and brand videos · a public AI-video test set.

SEE HOW I DIAGNOSE GROWTH CONTACT ME DOWNLOAD CV
STARTING FROM ZERO

From zero to one

I connect store, advertising, ERP, competitor, content and customer data. AI collects the numbers, finds what changed, prepares or performs approved work, and checks the result. People still approve price, budget, payment, account access and public release.

AMAZON · DAILY OPERATIONS AI

Find which ASIN, campaign or keyword needs attention today

Listing, ads, rank, inventory and profit are read together—not in separate reports.

DATA IN
Business Report, ad search terms, Search Query Performance, Softime / SellerSprite rank, ERP inventory and margin, reviews and returns.
AI RUNS
Join the data by ASIN and date; compare Sessions, CVR, CPC, ACOS, TACOS, organic rank, stock turns and contribution profit; then rank Listing, keyword, budget and replenishment work.
ONE EXAMPLE
When CVR falls, check price, new review themes, search-term mix, main image and video, recent Listing edits, stock and delivery together. Produce the issue summary, copy changes, five video hooks and suggested negatives; a person approves account changes.
VIEW OPERATIONS CASE → VIEW AMAZON CODE ↗
TIKTOK · MULTIMODAL AI

Different creators should not shoot the same product the same way

I adjust the script to each creator’s audience, content style and filming habits, so they receive a brief they can pick up and shoot.

DATA IN
Product pages, source images, reviews, creator history, video reports, two- and six-second views, clicks, product visits, orders, refunds, inventory and fulfilment.
AI RUNS
Lock the exact SKU, turn complaints and differences into hooks, write creator-specific briefs and storyboards, then check generated video for product deformation, wrong actions, broken continuity and unsupported claims.
ONE EXAMPLE
Amazon negative reviews → five video hooks. A product page → an 18-shot storyboard. A creator profile → a shoot-ready brief written for that person. An AI video → product, action and continuity checks.
VIEW CREATOR METHOD → VIEW TIKTOK CODE ↗
SHOPIFY · GEO + CONVERSION AI

Trace ChatGPT orders to the page, SKU, profit and returns

Traffic alone does not show whether a product should scale.

DATA IN
Shopify orders, SKUs and refunds; GA4 landing and checkout events; Search Console, ads, on-site search, support questions, reviews, product pages, schema, feeds, stock and margin.
AI RUNS
Connect source → landing page → SKU → paid order → profit and refund; separate confirmed ChatGPT referrals, AI-assisted orders and unknown sources; then rank PDP, checkout, GEO and SEO work.
ONE EXAMPLE
After a ChatGPT order signal, trace the exact landing page, SKU, order value, profit and refund. Turn repeated questions into product comparisons, FAQs, a Hero Video, blog posts and pages that search and answer tools can understand, then check later sessions and orders.
VIEW GEO CASE → VIEW SHOPIFY CODE ↗
B2B · EDM & ACCOUNT DEVELOPMENT

Understand what the company sells and who buys—then decide which products and proposal to bring

This is account-based EDM: each company and buyer receives relevant email content, not the same message blasted to a purchased list.

DATA IN
Product and supply capability, target-company websites, catalogues, news and public social activity, likely buying roles, plus sample, quote, order and reorder status in Feishu or CRM.
AI RUNS
Collect, deduplicate and score accounts; map product fit, price band, buying role and risk; prepare an account brief, email or LinkedIn draft, one-page proposal and tailored product visual or video. Feishu tracks review, sending and follow-up.
ONE EXAMPLE
Brazil distributor sample data → company, category and likely buying role → priority → matched SKU, sample and quote pack → a follow-up suggestion awaiting review. Nothing is sent before approval, and no relationship is fabricated.
VIEW B2B METHOD → VIEW B2B CODE ↗

Seven AI modules share the same store data and action history.

A problem found by any module enters one task queue. After execution, GMV, margin, rank, returns, stock and replies return by SKU, channel and date. The next run keeps, changes or rolls back the action—it does not end with a report.

01 · PROFIT MODULE

Recalculate contribution profit by SKU × channel every day

DATA
Orders, price, discounts, product cost, platform fees, ad spend, FBA / 3PL, freight, refunds and returns.
AI
Align currency, date and attribution; calculate contribution profit; flag where GMV rises while margin falls or a required cost is missing.
WRITE BACK
Loss-making SKUs, price or budget proposals and missing-cost alerts enter the task queue. Price and budget changes wait for approval.
02 · REPEATABILITY MODULE

Separate stable demand from a one-off hit

DATA
Search terms and rank, videos, creators, landing pages, orders, returns, repeat buyers and promotion dates.
AI
Compare the same SKU across weeks, creators, creative versions and channels; separate search demand, creative strength and creator fit from a promotion spike.
WRITE BACK
Reusable keywords, hooks, creator traits and stop conditions return to the Listing, content, creator and ad queues.
03 · CONVERSION MODULE

Locate where traffic fails to become an order

DATA
Sessions, Listing or PDP versions, offer and price, reviews, add-to-cart, checkout, payment, delivery promise and stock.
AI
Rebuild the path from visit to payment; compare page, price, traffic mix, review, stock and delivery changes; keep facts separate from suspected causes.
WRITE BACK
Create the exact Listing, PDP, offer or checkout test, with owner, expected signal and approval state. Later CVR, profit and returns decide whether it stays.
04 · RETENTION MODULE

Find why customers reorder, leave or return the product

DATA
Customer, first SKU and source, time to second order, support topics, refunds, returns, email and community response.
AI
Group outcomes by SKU, source and reason; distinguish a product problem, wrong expectation, late delivery, missing replenishment reminder or cross-sell gap.
WRITE BACK
Create product, content, FAQ, support, replenishment or cross-sell work. The next run checks repeat purchase, refund and support-contact changes.
05 · CONCENTRATION MODULE

Expose dependence on one ASIN, creator, account or buyer

DATA
GMV and profit by ASIN, channel, campaign, creator, video, ad account, buyer and traffic source.
AI
Calculate concentration and model what happens if the top source pauses; distinguish dependence on volume from dependence on profit.
WRITE BACK
Queue backup keywords, creative, creators, channels or customer development in the right order instead of blindly opening more channels.
06 · SUPPLY & CASH MODULE

Set replenishment priority before stock or cash becomes critical

DATA
Stock, sales velocity, lead time, purchase orders and inbound, payment terms, freight, ad plan, refunds and returns.
AI
Calculate stock cover, likely stockout window and cash required; connect a proposed ad increase with inventory and payment timing.
WRITE BACK
Create replenishment and budget calendars plus exception alerts. Purchase orders, payments and final price remain human approvals.
07 · EXECUTION & STORE MEMORY

Turn the six diagnoses into work—and remember what happened

DATA
Alerts from the six modules, account permissions, rule thresholds, action history, owners, approvals, previous results and rollback conditions.
AI
Rank P0 / P1 / P2 work, merge related symptoms into one task, prepare the exact change, and run only authorized low-risk actions through APIs, MCPs or CLIs. High-risk actions wait for a person.
WRITE BACK
Record who changed what, when, why, the before-and-after result and whether it was kept or rolled back. This becomes store memory for the next run.
ONE DAILY RUN

AI-automated daily tasks

Every task keeps its input, evidence, owner, approval state, expected signal and actual result. Missing data remains missing; a draft never becomes an executed action by wording alone.

01 · 08:00Pull authorized data

Store, ads, ERP, analytics, competitors, content, creators, support and CRM.

02 · ALIGNAlign the records

Map ASIN / SKU × channel × date; normalize time, currency and attribution; keep missing values visible.

03 · DIAGNOSERun the seven modules together

Profit, repeatability, conversion, retention, concentration, supply and execution share evidence instead of producing seven isolated reports.

04 · QUEUEBuild one action queue

Each task contains the affected object, evidence, priority, owner, proposed change, expected signal, approval and rollback condition.

05 · EXECUTEApprove, then execute

Authorized low-risk work can run through APIs, MCPs or CLIs. Price, budget, payment, claims, account access and publication wait for a person.

06 · READ BACKCheck and remember

Read GMV, margin, CVR, ACOS, rank, returns, stock and replies; keep, adjust or roll back the action; save the result to store memory.

EXAMPLE · AMAZON SWEATPANTS ASIN

A hero sweatpants ASIN shows how the modules work together. When CVR falls, CPC rises and stock cover shortens, the system checks search terms, ad placement, price, new sizing reviews, Listing changes, delivery and inventory together—not as separate alerts. It prepares suggested negatives, a size comparison, Listing and video changes, plus a replenishment alert in one queue. After a person approves the account, content and purchase-order actions, the next run compares CVR, ACOS, rank, contribution profit and stock cover, then keeps, changes or rolls back each action.

FOUR COMMON BUSINESS SITUATIONS

Four business situations I see most often.

DATA → DECISION → EXECUTION → READ-BACK Store, advertising, ERP, competitor, social and supplier data → AI judgment → Skills, APIs and CLIs → Amazon, TikTok Shop, Shopify, CJ, Meta, Feishu and DingTalk → GMV, margin, CVR, returns, ranking and inventory feed the next action.
PARAMETERS → QA → WRITE-BACK

One SKU, from ERP data to Amazon and storefront publishing

The model is not left to improvise from one prompt. Every SKU enters the task with source product images and structured fields. Generation, upload and public release are separate steps, and the system reads the platform result back after each write.

SOURCE DATAAn ERP MCP connector reads authorized SKU, variant, cost, stock, dimensions, packaging, logistics and existing Listing fields. Amazon SP-API provides the marketplace's current Listing state. SellerSprite supplies third-party keyword, rank and competitor trends; public or authorized high-performing content is used only to study composition, setting and pacing. Conflicting fields stop for review.
MODEL PARAMETERSEach SKU sends source product images plus structured JSON. product_truth locks the SKU and variant, colour, material, dimensions, packaging, logo, set quantity, allowed claims and forbidden changes. style_reference carries only visual direction. Task fields specify platform, language, aspect ratio, output count and whether to validate, save as a draft or wait for release.
BATCH IMAGE CONTROLFiles stay tied to SKU, variant and channel from generation onward. A vision model runs the first check against the source images for shape, colour, logo, packaging text, set quantity and use. Clear failures are regenerated; low-confidence, OCR-disputed and high-risk results go to human review before entering the upload queue.
PLATFORM WRITE-BACKShopify CLI or a custom CLI calls the authorized Admin GraphQL API to create or update an unpublished product, upload and order media, write alt text and bind variants. A single Amazon SKU can be pre-validated through Listings Items API; batches use JSON_LISTINGS_FEED, with image URLs submitted as Listing attributes. A person confirms the final write and public release.
RESULT READ-BACKThe system stores the relevant SKU or ASIN, Shopify product and file IDs, Amazon feedId, processing state and errors. It then checks Amazon Listing issues and feed reports, and waits for Shopify media to be READY before checking variant binding and channel visibility. Only the requested state closes the task; ACCEPTED or DONE alone does not prove that a product is live or searchable.
SEE THE FULL DATA FLOW →
AI RUNS DAILY OPS

AI reads the Amazon store each day and starts the next action

It combines Listing, traffic, ads, ranking, stock and profit, ranks anomalies, executes work within agreed rules, and checks the result the next day.

DATA INSessions, CVR, CPC, ad spend, ACOS, TACOS, organic and paid rank, sales, margin, stock, returns, reviews, keywords and competitor changes.
AI RUNSMerge and normalize store, advertising, ERP and competitor data; detect anomalies, explain likely causes and rank actions. Low-risk work runs inside approved rules; anything beyond a threshold enters the approval queue.
PLATFORMSAmazon Seller Central · SP-API · Amazon Ads · Softime / SellerSprite · ERP / StoreClaw · Feishu / DingTalk CLI · an authorized ZiNiao workspace.
READ-BACKThe next daily run checks whether rank, CVR, CPC, ACOS, sales, margin and stock improved, then updates the priority list.
HUMAN GATEPrice, budget increases, bids beyond limits, negative keywords, bulk publishing, replenishment and irreversible account actions.
SEE THE AMAZON WORKFLOW →
AI RUNS PRODUCT TESTS

With no product, AI runs from sourcing to the first order

AI screens suppliers and exact variants, builds the store draft and content, distributes tests, and follows the order and delivery state.

DATA INExact CJ, DSers and AliExpress variants, cost, stock, warehouse, weight, freight, delivery and returns; competitor price, reviews, ads, trends and social content.
AI RUNSReject weak products by margin, stock, freight, return risk and competition; map the exact SKU; create the Shopify draft, PDP, pricing scenarios, video, creator brief, social posts, blog and GEO pages; then check fulfilment.
PLATFORMSQoderWork CN · SellerSprite · CJdropshipping · DSers · AliExpress · Shopify · TikTok Shop · PayPal · Meta / Instagram · YouTube · LinkedIn · X · Facebook · Feishu / DingTalk CLI · ERP / 3PL.
READ-BACKTraffic, add-to-cart, checkout, payment, order, supplier acceptance, tracking, delivery and returns decide whether a product scales or stops.
HUMAN GATESamples, supplier terms, payment and tax accounts, test budget, domain and public store opening.
SEE THE DROPSHIPPING WORKFLOW →
LOCK THE SKU BEFORE GENERATION

One SKU, four separate routes for Amazon A+, TikTok, Instagram and DTC video

QA is only the last gate. Before a model runs, I lock the product, the buyer question, the platform format, the shot plan and what each reference is allowed to control. Batch output is then easier to keep on-product, continuous and repairable.

INPUT PACKThe exact SKU link, official images, variants, dimensions, packaging, existing Listing, reviews and platform rules become one structured pack. Sources, unknowns and conflicts remain visible; the model cannot fill in unsupported functions or claims.
PRODUCT VS REFERENCEThe Product Anchor locks shape, colour, material, logo, packaging, parts, scale and supported use. High-performing TikTok or Instagram content can inform the hook, composition, pacing and creator voice only; it cannot rewrite the SKU or be copied as creative.
PLATFORM BRANCHESAmazon A+ / Listing uses five 16:9 proof frames to answer buyer doubts about material, size, parts or operation. TikTok uses a six-shot 9:16 chain with an in-market hook, creator point of view and supported product action. Instagram / Reels turns authorized samples into composition and pacing direction. A 60/90-second DTC film freezes one six-beat, 18-shot story across three boards.
MODEL HAND-OFFReferences do not share jobs: the product image controls the product; when a fixed person is needed, an authorized identity image controls that person; interaction references control hands, occlusion and scale; the previous-page exit controls the next pose, camera and scene. One script is frozen first, then Part 1 → 2 → 3 runs in sequence. Each call receives only its shots and matching references, not one giant prompt.
FINAL GATEEvery candidate remains tied to its SKU, platform, run and version. The actual image or video is checked for product geometry, logo and text, hand contact, action order, person and scene continuity, aspect ratio and claim safety. Clear failures return to the affected stage; low-confidence and high-risk cases require human review. QA stops bad output from leaving—it does not choose the creative direction.
SEE THE 18-SHOT CONTROL METHOD →
SELECTED EXAMPLES · HOW I THINK AND EXECUTE

Three examples: a cross-border project, AI product design and delivery.

WATCH THE VIDEO WORK →
CROSS-BORDER DTC PLANNING · LUOLAI AUSTRALIA

LUOLAI Australia: move high-ticket bedding from product files to Shopify, Amazon and overseas content.

AUSTRALIATARGET MARKETAU.LUOLAI.COM · June 2026 planning snapshot
SHOPIFYDTC STOREFRONTEnglish storefront for Australian customers
25 SKUPLAN SNAPSHOTMulberry silk · goose down · bedding bundles
RAASAMAZON CLIENT WORKProject details are recorded in my résumé
01 · MARKET & SKU

Decide who to sell to and which SKU goes first

DATA IN
Product, material, price, stock, freight and returns; Australian seasonality, search terms, competitors and audience signals.

ACTION
Choose the local-Australian or Chinese-diaspora priority, hero SKU, channel and the assumptions that still need testing.

02 · STORE & CVR

Turn high-ticket trust into page changes

DATA IN
Shopify and GA4 funnel, mobile behavior, product-page exits, size and material questions, shipping, returns and reviews.

ACTION
Produce a ranked list for the homepage, PDP, trust proof, CTA, cart and checkout—not a generic redesign.

03 · CONTENT & SEARCH

One product file, different assets for each channel

DATA IN
Material, use, care, customer questions, Amazon keywords, competitor gaps and Australian search intent.

ACTION
Amazon Listing and A+, Shopify Hero and PDP, Instagram or TikTok video briefs, plus blog, SEO and GEO pages.

04 · MEDIA & CREATORS

Test the winter audience before scaling spend

DATA IN
Meta and Google performance, creative hooks, audience segments, creator replies, CPA and ROAS.

ACTION
Create the test matrix, channel-specific assets, creator briefs and the budget queue that still requires approval.

05 · READ-BACK

Let GMV, margin, returns and stock set the next move

DATA IN
GMV, CVR, AOV, ROAS, cart and checkout loss, stock, returns, email revenue and repeat purchase.

ACTION
Rank the next product, page, content, media, stock and retention actions, then check the result again.

THE AI-AUTOMATED CROSS-BORDER ROUTE Shopify, GA4, Amazon, Softime or SellerSprite, advertising, creator and inventory data → AI collects and judges → content and task queues are created → approved low-risk work runs → GMV, margin, CVR, returns and stock feed the next daily decision. People approve price, budget, claims, account access and public release.

My responsibility: market and SKU prioritization, Amazon and DTC channel roles, store and content planning, creator and media testing, fulfilment checks, and the operating rhythm for reading results back into the next decision. The internal launch document remains private.

SEE RÉSUMÉ RECORD → OPEN AMAZON SELLER RECORD ↗
LONG-TERM GROWTH ASSETS

Explain why customers buy and what they ask after purchase, then use community, editorial content, search and ChatGPT to keep bringing in new customers.

Start with real support questions, reviews, returns and site search, then judge the work by branded search, organic visits, AI referrals, repeat purchase and orders.

Brand

Turn the buying reason into one consistent promise across product, page, video, packaging and service.

Read: branded search · direct visits · price acceptance · repeat purchase

Community

Use support, reviews, members and creators where the product benefits from learning, sharing, replenishment or co-creation.

Read: active customers · useful UGC · referrals · cross-category purchase

Editorial

Turn repeated questions from reviews, support, returns and site search into answers that remain useful.

Read: qualified entry pages · assisted orders · support deflection

SEO

Capture existing search demand through category, comparison, compatibility, care and use pages.

Read: indexing · search queries · qualified organic visits · orders

GEO

Give AI answer tools accurate product information, clear comparisons, sources, authorship and update dates.

Read: AI referral → page → SKU → order → return and profit
COMMUNITY · CREATOR · LAUNCH OPERATIONS

AI turns community questions, launch work and creator collaborations into a daily operating queue.

WHAT CHANGESAuthorized conversations and platform records are sorted into members, questions, creators and tasks, instead of being reread by hand or answered with one broadcast message. A person still approves public replies, incentives, outreach, rates, usage rights and publication.

AUTHORIZED SIGNALS → DAILY QUEUE → CHANNEL ACTION → RESULT WeCom and Feishu conversations, Discord onboarding and forums, support, reviews, creator content, campaign status and orders → AI removes duplicates, groups issues, sets priority and calls the relevant Skill → approved work runs in the right channel → joins, replies, content, visits, orders, returns and repeat purchase set the next task.
01 · LAUNCH & RANKING DAY

Put launch replies, owners and the Discord hand-off in one queue

In a Product Hunt launch I worked on, the team used a dedicated Discord invite, a welcome path, reply references, a launch-day roster, limited admin roles and status collection. Under the current rules, that operating method should direct supporters to try the product and leave genuine feedback—not solicit votes.

BEFORE
Lock the demo path, FAQ, maker roles, direct launch link, Discord entry, welcome tasks, response ownership and escalation rules.
AI RUNS
Read authorized questions and status updates, merge duplicates, rank urgent product issues, assign an owner and prepare source-linked reply notes for review.
PLATFORM RULE
Product Hunt prohibits AI-generated and mass comments and acts against fake or coordinated voting. The maker writes or approves every public reply; AI never votes, creates accounts or comments automatically.
READ BACK
Direct-link visits, Discord joins, welcome-task completion, real questions, trials and next-week retention—not the ranking alone.
02 · DAILY COMMUNITY

Answer what is stuck first, then schedule the day's activity

Discord holds ongoing learning, member support and creator exchange. WeCom carries seed-user and customer conversations. Public communities help reveal recurring questions; they are not automatically treated as a private contact list.

SETUP
Use Discord Onboarding questions to assign roles and channels, Server Guide tasks for the first action, Forum tags for durable questions and Scheduled Events for product clinics, office hours and creator showcases.
AI RUNS
Prepare an unanswered-question queue, newcomer follow-ups, repeated objections, a moderator hand-off and one activity brief from authorized messages, support, reviews and returns.
ACTIVITIES
Run newcomer tasks, weekly Q&A, product-use clinics, UGC prompts, roadmap polls, creator demos and customer-story reviews. Each activity starts from a real member question.
READ BACK
Visitors, communicators, new-member retention, unanswered time, useful UGC, referrals, support reduction, repeat purchase, opt-outs and complaints.
03 · CREATOR OUTREACH

One creator gets one outreach note and one shootable brief

The same SKU is framed differently for each creator's audience, format and speaking style. Exact product facts stay fixed; the hook, scene, demonstration and CTA change with the creator.

FIND
Use TikTok One's 2026 Creator AI Search and creator projects, Meta Creator Marketplace, plus public or authorized creator history. Compare audience market, format, comments, past partnerships, product fit and risk.
AI PREPARES
Remove duplicates, score fit and produce a creator-specific opener, product angle, shot list, reference frames, proof points, forbidden claims and follow-up timing.
FEISHU QUEUE
One row per creator records owner, channel, SKU, evidence, approval, sample, due date, follow-up, reply, content status and stop reason. The CLI writes and updates the queue; approval controls sending.
READ BACK
Reply, agreement, sample delivery, script approval, post, authorized ad use, visits, orders and refunds update the next shortlist. Rates, samples, contracts, rights and sending remain human decisions.
04 · CREATOR FEEDBACK → MODEL TASK

When a creator says “change eight seconds,” the model must know what cannot move

Recent creator feedback handled in WeCom included a small edit regenerating the whole video, an unwanted background change, unclear speech, a pet identity changing and a voice edit changing the visual. These are product inputs, not comments to archive.

CAPTURE
Keep the source video, prompt, requested change, elements that must stay fixed, authorized identity or product references, model, run, version and the creator's exact feedback.
COMPARE
Compare only the requested region and tag whole-video regeneration, background drift, identity mismatch, unclear speech, broken timing or an unnatural cut.
REPAIR
Return to the affected shot or generation stage, lock the last accepted frame, product, person and voice, and rerun the smallest possible section. Low-confidence changes go to review.
USE AGAIN
An accepted repair becomes a rule in the next creator brief, a model comparison case and a product backlog item. Time to an acceptable result and repeated failures matter more than raw output count.
ONE JOB PER CHANNEL

The source material is shared; the operating method is not.

AI first decides whether a signal belongs to member support, public discussion, creator work, a launch, B2B development or consented follow-up. It then prepares the format, owner and measurement that fit that channel.

DISCORDOnboarding, forums, events and retention

Onboarding assigns roles and channels; Server Guide gives the first three to five tasks; Forum tags hold questions; Scheduled Events run clinics and showcases. AI builds the moderator queue; where Server Insights is available, it shows visitors, communicators and new-member retention.

WECOM · FEISHUSeed users, customers and task ownership

WeCom carries customer questions, launch coordination and creator feedback. Feishu keeps one auditable row per creator, issue or activity with owner, approval, follow-up and stop reason. AI summarizes and assigns; it does not mass-send private messages.

REDDIT · FACEBOOK GROUPSListen to the question before the brand answers

AI groups recurring public questions and maps them to a SKU, FAQ or content task. Reddit replies stay human, specific and disclosed—never automated AI comments. Facebook Groups use questions, polls, live sessions and events for member research and support.

TIKTOK · INSTAGRAMFind creators, prepare briefs and scale authorized work

TikTok One and Meta Creator Marketplace provide current creator discovery, project and permission tools. AI prepares the match and materials; people confirm rates and rights. Approved creator work can enter Spark or Partnership Ads, then read visits, orders and returns back to the creator record.

YOUTUBEKeep asking after the video is published

AI turns comments into a poll, Community prompt, Live Q&A queue and the next video brief. The channel owner moderates and replies. Community participation, returning viewers, qualified site visits and product questions guide the next topic.

LINKEDIN · X · EMAIL · WHATSAPPB2B topics, launch distribution and consented follow-up

AI researches the company and buyer role before preparing a tailored post, newsletter, event invitation or EDM sequence. Email, WhatsApp and Telegram follow-up only uses consented contacts; opt-out stops the sequence immediately. Replies, meetings, referrals and repeat purchase decide the next message.

This section comes from my work with Discord and overseas communities, creator segmentation in the TikTok Singapore ecosystem, Product Hunt launch coordination, and recent creator feedback handled through WeCom. The public pages below show the Feishu queue, creator-outreach kit and the path from community questions to GEO and commerce content.

SEE THE FEISHU QUEUE → OPEN THE OUTREACH KIT → SEE THE GEO RETURN PATH →
PUBLIC TOOLKIT

This workflow brings product data, reviews, keywords, Listing drafts, paused ad setups and reports into one daily review. A person approves every action before anything changes.

PROBLEMTeams review Listings, ads and profit in separate tools
WHAT I DIDConnected six workflow steps, designed a daily check and recorded what happened after approved actions
OUTPUTSA public method, working tools and a detailed interface design
RUN STATESix skills and the daily review are ready; seller-authorized data can be connected to produce Listing and advertising actions for approval
SEE THE TOOLKIT → SEE THE DAILY REVIEW →
DELIVERED VIDEO

Starting from one verified product page, I created storyboards, a production script, a finished film and a step-by-step production demo, with visual checks to keep the garment accurate.

PROBLEMAI fashion videos can change garment details, fit or movement
WHAT I DIDChecked the product page, wrote three storyboards and a production script, then reviewed every shot
OUTPUTSA 23.17-second finished film and a 40.77-second production demo
DELIVERYThree storyboard sheets, a production script, the finished film and a source-to-output walkthrough
SEE THE CASE → PUBLIC SITE ↗
AULUOLAI AUSTRALIA DTC PLAN + AMAZON RECORD AMZAMAZON STORE, LISTING AND WORK RECORDS 3TIKTOK SHOP LINKS VIDEOPRODUCT AND BRAND VIDEO WORK
THE DETAILED WORK BEHIND THE RESULT

31 specific tasks behind the work.

The 31 items below cover the full growth cycle: product research, multimodal content, advertising, store operations and growth. Each item opens a related case or usable tool and states its current operating status.

5WORK AREAS
31SPECIFIC TASKS
1:1LINK FOR EACH
SEE ALL 31 TASKS AND EXAMPLESREAL EXAMPLE · WORKING TOOL · IN PROGRESS
01

Product Research

DECIDE WHAT TO SELL
  • Product ResearchMARKET · USER · SKU

    Market, competitor, customer, price and SKU evidence assembled into a decision brief.

    VERIFIEDGitHub ↗
  • Amazon OpportunityASIN · VOC · KEYWORD

    ASIN, review and keyword evidence; market sizing and final Go/No-Go remain explicit inputs.

    TOOLKITCase ↗
  • Competitor ResearchCHANGE · CONTENT · OFFER

    Public-source collection and normalized competitor signals for offer and content decisions.

    TOOLKITGitHub ↗
  • Product Validation & Unit EconomicsDEMAND · MARGIN · GATE

    Demand tests, contribution margin, CAC ceiling, MOQ exposure and explicit validation gates.

    TOOLKITGitHub ↗
  • Pricing StrategyWTP · PRICE LADDER · MARGIN

    Competitor price ladders, willingness-to-pay tests and margin-aware price boundaries.

    TOOLKITGitHub ↗
  • Sourcing & Factory MatchingTREND · MOQ · CAPABILITY

    Trend clusters translated into supplier criteria, factory capability and MOQ fit.

    TOOLKITGitHub ↗
02

Content Generation

TURN PRODUCT DETAILS INTO CONTENT
  • UGC ScriptHOOK · PROOF · CTA

    Mobile-first scripts based on real product use, customer objections and believable creator actions.

    VERIFIEDGitHub ↗
  • StoryboardSCENE · ACTION · CONTINUITY

    Scene-by-scene boards with product anchors, actions, framing and continuity constraints.

    VERIFIEDGitHub ↗
  • AI Product PhotographyIMAGE-FIRST · SKU FIDELITY

    Product-image-first lifestyle scenes; this is AI imagery, not a physical studio shoot.

    TOOLKITGitHub ↗
  • Brand Film Planning30 / 60 / 90 SEC

    Continuous brand-film treatment, three boards and an execution script for longer-form production.

    VERIFIEDCase ↗
  • Exact-SKU Video Production & QAGENERATE · REPAIR · ACCEPT

    Image/video generation, continuity repair, file assembly and explicit accept/reject checks.

    VERIFIEDCase ↗
  • Product Detail Control & Tool SelectionDETAILS · TOOLS · PAST MISTAKES

    Confirm product details, choose how to make the asset and check past mistakes before production.

    VERIFIEDGitHub ↗
03

Advertising

DIAGNOSE BEFORE SCALING
  • Meta Creative ReviewCTR · CVR · CPA

    Read-only ad-row review compares CTR, conversion rate and CPA with account medians, then ranks the creative and budget questions that need attention first.

    TOOLKITGitHub ↗
  • TikTok Creative ReviewHOOK · HOLD · ACTION

    Offline funnel review compares two-second, six-second, click and action stages without turning missing metrics into zero.

    TOOLKITGitHub ↗
  • Amazon Listing & Ads OperationsCTR · CPC · CVR · ACOS

    Replaces the narrow “CTR Review”: listing, keywords, SP launch, budget, negatives and report diagnosis.

    TOOLKITCase ↗
  • Hook OptimizationVARIANT · TEST · LEARN

    Reviews seconds 0–3, designs one-variable hook tests and ranks like-for-like results; causal conclusions come from controlled test results.

    TOOLKITGitHub ↗
  • Attribution & Experiment DesignUTM · EVENTS · A/B

    Tracking plan, UTM governance, capture events, experiment queue and success criteria.

    TOOLKITGitHub ↗
04

Operations

FIX THE BUYING PATH
  • Shopify Store Audit & Launch QAHOME · PDP · CART · CHECKOUT

    Real browser QA across mobile, merchandising, trust, policies, checkout and launch blockers.

    VERIFIEDGitHub ↗
  • TikTok Shop AuditCATALOG · FULFILLMENT · REFUND

    Read-only audit turns authorized product, order, return and finance exports into prioritized business issues without writing to the shop.

    TOOLKITGitHub ↗
  • Review MiningVOC · PAIN · OBJECTION

    Clean and segment reviews into pains, motivations, objections, questions and keyword language.

    VERIFIEDCase ↗
  • Refund AnalysisREASON · SKU · COHORT

    Local CSV analysis ranks refund reasons, SKUs, channels and cohorts; rates appear only when a complete all-orders denominator is declared.

    TOOLKITGitHub ↗
  • Conversion DiagnosisFRICTION · TRUST · OFFER

    Trace purchase-path friction into concrete page, policy, offer and trust fixes.

    VERIFIEDGitHub ↗
  • CRO AuditHERO · OFFER · PROOF · CTA

    Score the persuasion path and turn gaps into prioritized experiments, not generic redesign notes.

    TOOLKITGitHub ↗
  • Catalog & Merchandising OperationsSKU · PRICE · FEED · QA

    SKU screening, checked product records, pricing rules, image briefs and draft catalog outputs.

    TOOLKITGitHub ↗
05

Growth

FIND AND RETAIN DEMAND
  • Creator OutreachSCORE · CONSENT · FOLLOW-UP

    Score creators, draft outreach, enforce consent gates and manage a follow-up queue.

    TOOLKITGitHub ↗
  • Affiliate DiscoveryFIT · ECONOMICS · RISK

    Find creators and score fit, likely economics and brand risk from public information; outbound contact stays disabled by default.

    TOOLKITGitHub ↗
  • SEO ContentINTENT · SCHEMA · READBACK

    Search-intent content, metadata, structured data, hreflang and public-page readback.

    VERIFIEDSite ↗
  • Email Flow ReviewWELCOME · ABANDON · RETAIN

    Read-only Klaviyo and Shopify Email snapshot review checks seven lifecycle flows, overlaps, triggers, status and evidence gaps without sending.

    TOOLKITGitHub ↗
  • B2B Account DiscoveryCOMPANY · ROLE · NEXT STEP

    Find and score companies, map public presence and form buyer-role hypotheses.

    TOOLKITGitHub ↗
  • Instagram / Reels Trend IntelligenceHOT · RISING · ANGLE

    Normalize public or authorized samples into viral signals and SKU content angles.

    TOOLKITCase ↗
  • Consent & Claims ComplianceRIGHTS · CLAIMS · PLATFORM

    Consent checks, claim boundaries and platform constraints embedded before outreach or generation.

    TOOLKITGitHub ↗

STATUS KEY — VERIFIED: delivered work or a real case · TOOLKIT: working code and inspectable outputs · BUILDING: data connections or operating records are still being completed.

§01 — HOW I CHECK AI VIDEO CMV-24 / PTFT
CHECKING PRODUCT ACCURACY IN AI VIDEO

24 real videos, reviewed for what went wrong.

CMV-24 records the problems that make an AI product video unusable: the product changes shape, an action looks impossible, scenes do not match, the selling point is unclear or the video creates a platform risk. I use those records to choose the right tool for each job and reject results that should not be delivered.

product_shape hand_object_contact action_logic scene_consistency commercial_readability compliance_risk
WHAT I CHECKEDn = 24
01
Similar past mistakes
find problems seen in similar products
02
Best tool for the job
choose a method that fits the product
03
Video quality check
spot product, action and platform risks
04
Usable result ranking
keep the result that is accurate and channel-ready
17
TIKTOK/UGC
3
INDIE HERO
4
AMAZON
// public artifacts
cmv24_metadata.csv · label_taxonomy.md
cmv24.schema.json · training_protocol.md
paper.md

Watch the work.

§02 — VIDEO WORK

Play 17 finished videos here, including brand films, a client-requested video and work made for TikTok, Reels, Shopify and Amazon.

01 / 17
Jaxon Lane — Bro Mask ritual
PLAYABLE ASSETS · n=17 SELECT ASSET · WATCH VIDEO

How I make one product video.

§03 — PRODUCT PAGE → READY-TO-USE VIDEO
01 · INTAKE
Collect the SKU, audience, channel and goal.
02 · CLASSIFY
Define category, format and communication objective.
03 · STRATEGY
Generate creative hypotheses from customer pain points.
04 · STORYBOARD
Break the asset into scenes, actions and proof moments.
05 · CHOOSE TOOLS
Choose generation and editing tools that fit the product.
06 · WRITE INSTRUCTIONS
Describe the product, person and scene clearly so details stay consistent.
07 · CHECK
Check the product shape, actions, realism and whether the selling point is clear.
08 · TEST VERSIONS
Prepare versions for each channel and decide what success looks like.
09 · KEEP WHAT WORKED
Save what worked, what failed and what to change next.

Practical guides and project notes

§04 — METHODS AND LESSONS

These notes show the business problem, the data and tools I used, what I produced and how the next operating step works. Open any card to read the full version.

Three merchant journeys you can follow: turning a supply-chain product into a launch, running a daily Amazon diagnosis, and testing a dropshipping product through first-order fulfilment. Each shows the source data, decisions that need approval, what was executed and what happened next.

Published here: 2026-07-24 Canonical: jiajingqiu.com/cross-border-ai-operating-system/ First detailed build: Amazon daily diagnosis Current state: product blueprint and working toolkits; account connectors are added per project

Six evidence-linked skills for source collection, review analysis, keyword governance, Listing drafts, paused ad initialization and report analysis.

Published here: 2026-07-21 Canonical: jiajingqiu.com/amazon-listing-ads-skill-suite/ Current state: working suite and a draft case; marketplace execution begins after account authorization Implementation: password-protected

DTC Daily Ops Router

TEAM OPS DESIGN

Automated product selection with QoderWork CN and SellerSprite, daily store and competitor intelligence, and AI-personalized creator activation managed through DingTalk and Feishu CLI—with enterprise seat, audit, browser-session and credit controls.

Architecture published: 2026-07-22 Canonical: jiajingqiu.com/dtc-daily-ops-router/ Scope: automated selection · creator activation · 8 daily gates · team governance Current state: design specification and working modules; account writes follow permissions and approval

ChatGPT Commerce GEO

FIELD METHOD

A measurable commerce GEO method for turning ChatGPT referral signals into SKU-level attribution, product-detail checks, crawl and feed checks, answer-ready pages, experiments, automated selection and creator activation.

Published here: 2026-07-22 Canonical: jiajingqiu.com/chatgpt-commerce-geo/ Scope: attribution · product details · crawl/feed · conversion · automation Current state: attribution, page and feed checks are ready; results are tracked through referral, order, return and profit data

Code and practical tools

§05 — HOW THE WORK IS BUILT

Each card shows a small tool I have built for ecommerce work. Open one to see what it does, what it needs as input and what it produces.

meta-creative-review

Read-only Meta creative triage using account-level CTR, conversion and CPA baselines.

PUBLICTYPESCRIPT ↗ tiktok-creative-review

Offline TikTok hook, hold, click and action funnel review.

PUBLICPYTHON ↗ hook-optimization

Controlled first-three-second variants and directional result ranking.

PUBLICSKILL ↗ tiktok-shop-audit

Read-only catalog, fulfillment, refund and finance audit for TikTok Shop exports.

PUBLICPHP ↗ refund-analysis

Local refund Pareto and rate analysis with explicit denominator checks.

PUBLICJAVASCRIPT ↗ affiliate-discovery

Creator discovery with fit, likely economics and brand-risk scoring based on public information.

PUBLICPYTHON ↗ email-flow-review

Read-only lifecycle flow coverage, overlap, trigger and evidence review.

PUBLICPYTHON ↗ crossborder-ai-fde-kit

Product research, sourcing, marketplace operations and commerce-agent patterns.

PUBLICWORKFLOW NOTE ↗ indie-store-growth-ai-kit

Shopify/CRO audits, tracking plans and prioritized experiment queues.

PUBLICWORKFLOW NOTE ↗ creator-outreach-ai-kit

Creator scoring, outreach drafting, consent checks and follow-up operations.

PUBLICWORKFLOW NOTE ↗ b2b-account-radar-ai-kit

Company discovery, account scoring and buyer-role research.

PUBLICWORKFLOW NOTE ↗ jingqiu-DTC

DTC storyboards and creative production based on checked product details.

PUBLICPYTHON ↗ storyboard-tiktok-video-skill

Product-image-first TikTok scripts and storyboard contracts.

PUBLICSKILL ↗ storyboard-dtc-product-creative

DTC product-creative storyboards with SKU, action and continuity constraints.

PUBLICSKILL ↗ storyboard-amazon-listing-demo

Amazon product-led proof storyboard adapted for listing communication.

PUBLICSKILL ↗ amazon-skill

An Amazon SKU content process with source checks and clear execution rules.

PUBLICSKILL ↗ ins2reel

Product-image-first lifestyle imagery and short-form asset workflow.

PUBLICSKILL ↗ cmv-24-ptft

A commerce video review set covering product accuracy, common errors and accept/reject decisions.

PUBLICDATASET ↗
VIEW COMPLETE GITHUB PROFILE ↗

Commerce work by channel

§06 — INDEPENDENT SITES · AMAZON · TIKTOK SHOP

My cross-border work is grouped by channel. The same brand appears more than once when my role genuinely covered independent-site, Amazon and TikTok Shop operations.

01 · INDEPENDENT SITES

Independent sites

Eight public DTC, brand and B2B websites already documented in the portfolio.

trix.loveBrazil B2B sourcing and lead capturePUBLIC SITE ↗ comfelie.comUS / UK brand and conversion sitePUBLIC SITE ↗ jmtrick.comFashion DTC and B2B lead capturePUBLIC SITE ↗ fizzandrizz.comUS jewelry brand sitePUBLIC SITE ↗ xsoulofficial.comUS beauty-device storefrontPUBLIC SITE ↗ shoplifelong.comErgonomic-product storefrontPUBLIC SITE ↗ twinklestarshop.netUS home-and-garden catalog sitePUBLIC SITE ↗ bencmateworld.comUS pet-care storefrontPUBLIC SITE ↗
02 · AMAZON

Amazon

Brand Stores, product pages and client engagements from my Amazon work.

COOLGYMRepresentative listing · ASIN B0G4JGCG4DPRODUCT PAGE ↗ GIANTGUYAmazon brand product resultsBRAND RESULTS ↗ TwinkleStarRAAS merchant support · Amazon Brand StoreBRAND STORE ↗ BENCMATERAAS merchant support · Amazon Brand StoreBRAND STORE ↗ LIFELONGRAAS merchant support · Amazon Brand StoreBRAND STORE ↗ XSOUL Laser Hair RemovalRAAS merchant support · Amazon Brand StoreBRAND STORE ↗ LUOLAILuolai Amazon RAAS client engagement · details in my résuméRÉSUMÉ ↗ ComfelieUS / UK brand-growth role · Amazon conversion scopeRÉSUMÉ RECORD ↗ BrisiraCross-channel Amazon content and TikTok Shop launch workflowOPEN SHOP ↗
03 · TIKTOK SHOP

TikTok Shop

Storefront links and project records from my TikTok Shop operations, consulting and platform work.

FizzandRizzUS TikTok Shop · interim content and store-growth consultingOPEN SHOP ↗ ComfelieUS TikTok Shop · operations, content, live and creator growthOPEN SHOP ↗ Brisira Co.US TikTok Shop · onboarding, launch and cold-start supportOPEN SHOP ↗
GIANTGUYProject consulting · TikTok Shop registration, document review, platform AM coordination and store cold startPROJECT RECORD
TikTok Singapore Official EcosystemCreator ecosystem operations and brand support (TikTok Shop SEA) · creator onboarding, brand live-commerce launch, TSP coordination, Ads training and merchant supportROLE RECORD

My project roles: the brands and stores above come from work I led, supported as a consultant, handled through RAAS client service, or completed within the TikTok Shop ecosystem. Each record states my specific role and channel responsibility.

→ CONTACT
Email jingqiuguilai@gmail.com Phone +86 133 4185 8992 WhatsApp Open WhatsApp