← Back to work
Case Study · 2025 – present

Merchant Onboarding Portal.

How I turned a flood of overseas-merchant support tickets into a self-serve, AI-guided onboarding flow — and helped a Nanjing platform land cleanly in North America.

Huarui Cloud Nanjing · Hybrid Nov 2025 – present Operational Business Analyst
15+
Pre-launch bugs surfaced
~30%
Reduction in support tickets
4-stage
Register / Verify / Train / Live
EN / 中文
Bilingual delivery
Illustrative flow 1 2 3 4 REGISTER VERIFY TRAIN LIVE SUPPORT TICKETS LAUNCH
Onboarding as four product stages, with support-ticket volume trending down after launch.
Python SQL AI content generation Video UX Product analytics Bilingual coordination
Problem

The friction wasn't in the product — it was in the seam between two continents.

Huarui Cloud was scaling a cross-border B2B/B2C marketplace out of Nanjing into the North American market. Overseas merchants were landing on a platform designed and documented in Chinese, then handing the hardest questions — registration paperwork, verification requirements, payout setup, go-live checks — to a China-based support team that couldn't answer them at Pacific-time speed.

I joined the North American site app-building team in November as the operational business analyst. By the end of my first week I could see the same three questions cycling through the ticket queue four times a day.

Approach

Treat onboarding as a product, not a document.

Before writing anything new, I sat inside pre-launch QA for the North America app and ran structured tests against the merchant flow — surfacing 15+ blocking bugs the engineering team fixed before we opened the doors.

Then I ran the ticket log through a Python + SQL pipeline to see which questions actually got asked, in which order, and at which stage they broke. Four clusters emerged. Those four clusters became four product stages: Register → Verify → Train → Live.

What I built

Two shipped surfaces — one for the merchant, one for the ops team.

An AI-powered onboarding portal. Each of the four stages carries an interactive video tutorial — script and voiceover generated with AI, delivered in both English and Chinese — that walks a merchant through the exact screens they'll see, in the exact order they'll see them. Merchants can rewatch any step without re-queuing a support ticket, and stage-completion is tracked in a lightweight dashboard my ops counterparts in Nanjing watch in real time.

An automated order-tracking dashboard in Python and SQL. The most common post-launch ticket wasn't a bug — it was "where's my order?" I built a query layer over the fulfillment tables and wrapped it in a merchant-facing status view, so the answer arrives without a human in the loop. Common status inquiries now resolve themselves; the ops group spends its hours on the exceptions instead of the norms.

When analytics reveal the same friction twice, you're looking at a product spec, not a report.
Outcome

Fewer tickets, cleaner expansion, a shared vocabulary across time zones.

Reflection

The leverage was closing the loop.

My job was to see the pattern in the tickets, but the real leverage was turning the analysis back into a shipped surface merchants could use themselves. A dashboard would have documented the pain; a portal removed it.

Being bilingual mattered less than being willing to sit inside both time zones and translate not just language but intent — what a Nanjing product manager meant, what a North American merchant heard, and what belonged in the video versus in the code.

Next

See how the same loop shows up on the analytics side.

Sales-engineering intelligence, retail seasonality, and a live quant bot — all built with the same analytics-into-product instinct.