Zhang Feng—the “Feng” behind one of the most talked-about AI transformation stories in China’s wholesale circuit—has a way of making the extraordinary sound mundane. “We just connected everything,” he told me over a video call, his laptop open behind him showing what looked like a dozen browser tabs. “ERP to this, WeChat to that. Suddenly everything talks to everything.”

What happened next is the kind of story that makes AI engineers roll their eyes and business owners lean in. In less than a year, Zhang went from barely knowing what an AI agent was to launching a paid product with hundreds of users—while simultaneously growing his core wholesale business by 30% annually.

The Starting Point: A Wife Who Couldn’t Sleep

It starts, as many small-business stories do, with someone working too hard.

Zhang’s wife handled procurement. Every few days she’d return from factory visits with anywhere from 300 to 400 new SKUs—clothing items in various sizes and colors. Each one had to be manually entered into the inventory system. Two hours, every single day, just for the入库 (stock-in) process.

“She was spending two hours every day just putting numbers into the computer,” Zhang said. “That’s not work. That’s tax on being small.”

The gap between where they were and where they needed to be wasn’t a gap at all—it was a canyon. Their upstream suppliers used System A. They used System B. Their downstream retailers used System C. None of them talked to each other. Every piece of data flowed manually, slowly, and with inevitable human error.

When Zhang first heard about AI agents—he calls it “menace” in a thick Guangzhou accent, mixing Mandarin and English the way local entrepreneurs do—he didn’t see it as a productivity tool. He saw it as a translator.

The Three-Stage AI Adoption Framework

What Zhang built over the following months wasn’t a single AI product. It was three distinct layers, each solving a different problem:

Stage 1: Internal Automation

The first problem to solve was data flow. When different systems don’t speak to each other, someone has to be the bridge. Zhang built that bridge with OCR.

Instead of manually typing in SKU data from purchase orders, his wife now takes a photo. The AI reads it, extracts all fields—item number, color, size, quantity, price—and populates the system automatically. What used to take two hours now takes twenty minutes.

“But that’s not the interesting part,” Zhang said. The interesting part is what came next.

He connected the ERP system to WeChat through an AI layer. Now, when a customer hasn’t ordered in 45 days—a “sleeping customer”—the system automatically sends a WeChat message with a personalized re-engagement offer. When a payment is overdue, the AI triggers a polite reminder. No more relying on the sales manager to remember who owes what.

“Before, you needed a dedicated person to do this,” Zhang explained. “Now it’s a button. A single button, and the AI does the rest.”

For a business with 15,000 retail store clients and only about 5,000 active in any given year, automated customer lifecycle management isn’t a nice-to-have—it’s the only way to scale without hiring a small army.

Stage 2: Empowering the Downstream

Zhang didn’t stop at his own business. He looked at his customers—those 15,000 small retail stores across China—and saw the same problem he once had: they were overwhelmed.

Every store owner knows they should be making social media content. Most of them have tried. A few have failed. Almost all have given up, blaming it on lack of talent, lack of time, or lack of “feeling” for what works.

Zhang’s insight was simple: the problem isn’t creativity. The problem is infrastructure. If you give small retailers a ready-made template system—templates for hooks, templates for scripting, templates for video editing—and then let AI handle the heavy lifting of adaptation and posting, suddenly the barrier to content creation drops to near zero.

He built what he calls a “hot content homework” system. Users pick a template, tweak a few details, and the AI generates a complete short video script with matching visuals. The system’s been running for about a month. They’ve already seen three accounts hit 1 million views, and one hit 10 million.

“The AI does the standardized work. Humans do the judgment work. That’s the split that makes sense.”

His own wife’s account—run with the old “spirit and inspiration” method—grows maybe 1,000-2,000 followers per month. Their test account, built with the AI framework from scratch, hit 30,000 followers in under a month.

Stage 3: Productizing the AI

Zhang didn’t plan to become an AI product founder. But somewhere between solving his own problems and watching his customers struggle, he realized the tools he was building had value beyond his own business.

His first product, launched under the brand name “几何” (Jihe, meaning “Geometry”), targets clothing wholesalers and retailers with an AI-powered product photography tool. Upload a single flat-lay photo of a garment, pick a model from a predefined list, and the AI generates a full professional-looking product shot—without the five-step process that traditional tools require (upload front, upload back, pick model, pick pose, pick background).

The tool costs 399 RMB per month (about $55). Users get a few thousand credits per month—enough to generate hundreds of product images.

Zhang has since added a hardware product: an AI-powered name card device that records and transcribes sales conversations in real-time. The device runs for 80 hours on a charge, automatically slices recordings, and syncs everything to a cloud dashboard where store managers can review the full customer journey—from first hello to final pitch.

The device sells for 899 RMB. He’s already produced 100 units.

What Actually Makes This Different

There are a thousand “AI for business” stories that end with a demo and a press release. This one is different for one reason: Zhang is making money.

Not pilot programs. Not proof-of-concept deployments. Not “we’re exploring partnerships.” Real, recurring revenue from real customers paying real money for real tools.

The key difference is in his approach to building. Zhang doesn’t start with “what can AI do?” He starts with “what’s broken in my business?” Then he asks “can AI fix it?” And if the answer is yes, he builds it, uses it, and only shows it to others after it’s been validated internally.

The one-button payment reminder system? Built for his own customers first. The product photography tool? Used by his own team for months before anyone else saw it. The name card device? Originally just to solve his own problem of not knowing what his sales staff were doing on customer visits.

This is not a “we disrupted traditional industry with AI” story. It’s a “we found the parts of our business that felt stupid and fixed them with AI” story.

The Honest Obstacles

Zhang is the first to admit the transformation wasn’t seamless. When I asked about setbacks, he pointed to the early days: “I was spending probably half my time on AI. Now I’m spending 90% of my time on AI. My wife thinks I’ve lost my mind.”

The learning curve was steep. Zhang used a combination of tools—primarily Claude (CC) and Codex—depending on the task. Codex, he says, has stronger aesthetic judgment. Claude has better code generation. Using them together requires learning when to switch and how to verify outputs.

He also noticed that not everyone who learns AI actually uses it. “Some people come to the course, get excited, go home, and then… nothing. They wait. They hesitate.” His theory on why: AI adoption requires a fundamentally different relationship with uncertainty. If you need to see proof before you believe, AI’s early outputs—imperfect, rough, obviously AI-generated—will stop you cold.

The Metaphor That Stuck With Me

Toward the end of our call, Zhang said something I haven’t been able to stop thinking about.

“AI isn’t just a productivity tool,” he said. “It’s a tool that widens your field of view.”

He meant it literally: when you can see further, you see different things. He didn’t set out to build a product company. He set out to solve his wife’s stock-in problem. But once he could see further—once the mundane tasks were handled—he noticed opportunities he’d never seen before. The downstream retailers struggling with content. The gap in product photography tools for下沉市场 (lower-tier markets). The invisible gap between what senior salespeople know and what junior ones can learn.

All of that was always there. But you need AI to clear the noise before you can see it.

Key Takeaways

• Start internal before external: Zhang validated every tool on his own business before selling it to others
• AI reduces friction, not creativity: The system doesn’t replace human judgment—it removes the boring parts so humans can focus on the important parts
• Productize after validation: Don’t go to market with an unproven tool; use it yourself first
• Small businesses can compete with AI: You don’t need a tech team; you need to be willing to learn and iterate
• The real value is visibility: AI doesn’t just speed up existing processes—it reveals new opportunities

FAQ

Q: How did Zhang Feng’s company grow from 5 square meters to 40M RMB revenue?

A: Starting in 2017 with a small stall in Guangzhou’s Thirteen Hongs market, Zhang and his wife grew through aggressive expansion (5m² → 15m² → 30m² → 75m²), weathering market relocations and the pandemic. The 30% annual growth rate continued even during COVID, reaching 40M RMB in 2024. Key growth drivers were expanding their B2B customer base from local retailers to 15,000 stores nationwide, and more recently, AI-powered efficiency gains and new productized AI services.

Q: What AI tools does Zhang Feng actually use in his wholesale business?

A: Zhang uses a combination of Claude (CC) and Codex for different tasks—Codex for aesthetic/creative work and Claude for code generation. His AI applications include: OCR-based inventory entry (2hr → 20min per session), automated WeChat customer outreach for sleeping clients and overdue payments, AI-generated video content templates for downstream retailers, and a product photography tool for generating professional garment images from flat-lay photos.

Q: How does the AI-powered product photography tool for clothing work?

A: The tool (brand name “几何”) simplifies the traditional product photography workflow from 5 steps (upload front, upload back, select model, select pose, select background) to just 2 steps: upload a flat-lay garment photo and select a model. The AI handles everything else—background generation, lighting, pose matching. Priced at 399 RMB/month with enough credits for hundreds of images, it targets smaller retailers who can’t afford professional photography but need consistent, trustworthy product imagery.

Q: What is the AI name card device and how does it work?

A: Priced at 899 RMB, this hardware device records sales conversations in offline environments (trade shows, store visits), automatically slices recordings into segments, and syncs transcriptions to a cloud dashboard. With 80-hour battery life and built-in connectivity, it helps B2B sales managers track what their team is actually saying to customers. The data then feeds into the AI system for coaching and performance analysis—identifying which salespeople close more and why.

Q: How can small traditional businesses start with AI adoption?

A: Zhang’s framework: 1) Identify your biggest time tax—the repetitive task that eats most of someone’s day; 2) Start with OCR or document processing to eliminate manual data entry; 3) Connect your existing systems (ERP, WeChat, CRM) before building new ones; 4) Use AI internally for 2-3 months before thinking about selling anything; 5) Focus on validation over elegance—it’s better to have a rough tool you use than a polished tool that sits unused.