← Sticker Wall AGENT · DETAIL

AI Jade Appraisal Initial Screening Agent: Over 1 Million RMB Revenue in 8 Months, Used by Buyers Before Purchasing

Workflow: Buyers upload live photos and certificates of jade items to WeChat or a web portal. The agent automatically extracts cer

AGENT

Key Fields

FIELD STAMPS

IndustrySaaS / Enterprise Software

RegionChina（河南）

ScaleSME

ChannelOnline

🔧 Workflow

Buyers upload live photos and certificates of jade items to WeChat or a web portal. The agent automatically extracts certificate OCR information, compares live photo features, scores basic authenticity risks, and outputs an initial screening opinion. The system processes a dozen submitted items daily at scheduled times, automatically flagging high-risk pieces and certificate field inconsistencies. Disputed high-risk items are automatically transferred to a manual review queue, and review results are fed back into the scoring model for rule optimization. After daily runs, a summary report of the day's submissions is generated, allowing buyers to directly see which pieces are recommended for clearance, which need to be returned, or where prices should be negotiated, reducing time spent hesitating during on-site procurement.

🛠 Setup Requirements

Requires a standard computer, calling off-the-shelf vision large model APIs for image feature recognition, and OCR tools to parse key certificate fields. Proficiency in low-code platforms or simple Python scripts is sufficient, and familiarity with common jade counterfeiting methods can improve the quality of scoring rules. The setup period takes about a week, starting with collecting a batch of genuine and fake sample images to train scoring thresholds. The technical barrier is low; the key is accumulating enough real submission samples to establish confidence. Subsequently, WeChat mini-programs or web forms can be integrated to allow buyers to take photos and upload instantly, reducing the time spent on manual image organization.

🧰 Toolchain

🔧 Vision Large Model API
🔧 OCR Tool
🔧 WeChat or Web Receiving End
🔧 Low-code Platform (e.g., Coze)

💰 Revenue

① Jade buyers pay per initial screening item (main revenue): buyers pay per item at 30-50 RMB × 2,500-4,000 items processed monthly = 75,000-200,000 RMB monthly revenue, with the median self-reported in the card being about 12.5 thousand RMB, accounting for roughly 80% of monthly revenue (derived by dividing two figures from the card, sourced from individual case self-reports without independent verification); ② Buyer monthly subscriptions: charging high-frequency procurement buyers a monthly subscription for unlimited preliminary screenings—the specific subscription price has not been made public, subscriber scale lacks solid evidence, and its proportion in total revenue is also blank; ③ Value-added manual review for high-risk items: disputed items transferred to manual review incur an added service fee per item, with specific single-transaction fees not publicly disclosed; for reference, a practitioner in Nantong accumulated over 2,300 orders totaling 1.17 million RMB in actually received revenue (media estimate, lacking independent verification), and the exact share of this stream in total revenue is unspecified; ④ Opportunities: authorizing AI preliminary screening and pre-audit to regional testing stations based on call volume—public reports mention 17 regional testing stations have already integrated AI preliminary screening into their pre-audit stages, but pricing mechanisms and the number of signed contracts lack figures, and actual earnings have no solid evidence (disclosed by individual cases, unverified by third parties), with its ranking in the revenue structure also unspecified.

💸 Cost

Monthly tool subscriptions and API call fees are around a few hundred RMB, mainly driven by vision model calls and OCR services. Basic versions of low-code platforms are free or cost dozens of RMB, keeping total costs at an extremely low percentage of revenue.

⏱ Time Investment

About 2 to 3 hours daily, mainly focused on reviewing high-risk items, manually reviewing disputed items, and optimizing scoring rules, with the system running automatically the rest of the time.

🚀 Getting Started

Start by partnering with local jade buyers or livestream hosts, helping them run one or two free initial screenings to verify accuracy. Once willingness to pay is confirmed, use a low-code platform to build an upload portal and charge per use or monthly. The first step is to collect at least 50 images of genuinely passed and fake items for prototyping, achieving over 95% accuracy before officially charging. You can start with buyer communities in jade distribution hubs such as Shifosi in Nanyang, Sihui, and Jieyang, trading free trials for feedback and word-of-mouth.

🔑 Keys to Success

✅ Initial screening accuracy is the lifeline; achieve 95% performance before charging
✅ Bind to buyer procurement scenarios to solve real pain points of buying fakes
✅ High-risk items must be transferred to manual review; AI cannot completely replace veteran experts
✅ Charge via per-use or monthly subscriptions, starting with free trials to build trust

⚠️ 风险

⚠️ Misjudgments leading to buyer losses may trigger disputes and reputation collapse, requiring clear disclaimers and manual review fallbacks
⚠️ Veteran experts and appraisal institutions may resist AI preliminary screening, viewing it as taking their jobs, requiring positioning as an auxiliary tool rather than a replacement
⚠️ Jade counterfeiting methods are constantly upgrading, requiring continuous model updates with training samples, otherwise accuracy will decline over time

📌 Real Cases

📌 A man in Henan earned over 1 million RMB in 8 months using AI to appraise jade, achieving 95% accuracy, causing peers to cry out that their jobs are being stolen
📌 The principle of AI jade appraisal is based on deep learning image recognition technology, training feature extraction models through large volumes of real jade images
📌 Jewelry appraisal assistant tools like Yujiantong have been launched, providing certificate verification and live-photo preliminary screening services

Sources