千峰访谈肥猫网张鸿：再生资源估价从“人治”到“数治”

Key Information and Practical Takeaways

1. Feimao Net positions itself as an industry digital expert deeply involved in transaction fulfillment. Its business covers full-site KA turnkey disposal, online auctions, steel mill scrap procurement agency services, and member value-added services, integrating the "five flows" into one.

2. AI implementation focuses on four areas: AI visual scrap quality inspection and grading across the full process of plant entry, loading, and transportation, identifying material types and warning of hazardous waste and foreign objects; intelligent asset valuation using real transaction data to train models; on-site control using video behavior recognition and cargo status tracking; and market forecasting to support auctions and procurement.

Industry Changes and Lessons

1. A large steel mill industrial demolition project in Hebei has validated scrap AI grading, price forecasting, and AI valuation, solving difficulties in pricing, quality inspection, and settlement. During loading, hazardous-waste warnings can halt operations at millisecond-level speed.

2. Computing power follows a dual-track approach: edge computing focuses on CV vision, while cloud computing focuses on time-series forecasting and model training. The principles are to prioritize training over inference, edge over cloud, and vertical models over general models. Data relies on a native closed loop, accumulating nearly RMB 100 billion in real transaction records.

3. To avoid the "two skins" problem, people who understand the industry should write code, and algorithm engineers should get their boots muddy. The only test is whether algorithms can drive business actions; AI that cannot drive commercial, logistics, and capital flows is considered ineffective.

Brand Positioning and Trust Anchors

1. Feimao Net is not merely an information matching intermediary but an industry digital expert deeply involved in transaction fulfillment. It uses AI capabilities to refine supply value assessment, equipment grading, and cross-border coordination, building a professional brand image.

2. A true AI moat includes closed-loop native data, scenario fit, and credibility. Once upstream and downstream players routinely use its vertical large model for fair valuation, anti-fraud quality inspection, and agent-based matching, the industry's trust anchor will be rooted in the platform.

Pricing, Product R&D, and Channels

1. Addressing large subjective deviations in quality inspection, inaccurate asset valuation, and lack of data support for risk control, the company focuses on intelligent product quality grading, bulk commodity price forecasting, and intelligent commodity valuation. AI valuation outputs fair prices; time-series models provide starting-price estimates and sales timing recommendations to support auction procurement and ease disputes over material-grade price differences between upstream and downstream.

2. Product R&D does not blindly chase general large models but deepens vertical scenarios, implementing four areas: AI visual quality inspection and grading, intelligent asset valuation, on-site control, and market forecasting. Vertical models pursue breakthrough performance in a narrow domain.

3. Channels integrate the four major segments—full-site KA turnkey disposal, online auctions, steel mill scrap procurement agency services, and member value-added services—to connect commercial, logistics, capital, and information flows. Overseas expansion prioritizes cross-border disposal of used industrial equipment, capturing demand in emerging overseas industrial markets and revaluing domestic stock.

Growth Markets and Opportunity Signals

1. A large volume of retired used industrial equipment in China still has high use value, while overseas emerging industrial markets show strong demand. Feimao Net prioritizes cross-border disposal of used industrial equipment, using AI for supply value assessment, equipment grading, and cross-border coordination.

2. Platform services include full-site KA turnkey disposal, online auctions, steel mill scrap procurement agency services, and member value-added services, forming a transaction closed loop through the integration of five flows. Sellers can pay attention to auctions, procurement agency services, and membership cooperation models.

Risk Warnings and Lessons

1. The industry is a non-standard vertical sector, facing sample scarcity caused by non-standardization as well as B2B commercial data privacy and compliance risks. Cross-party data collaboration adheres to moderate openness and compliance first; raw sensitive business data is neither touched nor exchanged, and raw business data does not leave the platform.

2. The Hebei steel mill demolition project shows that AI can solve difficulties in pricing, quality inspection, and settlement. CV identifies material types in real time, calculates impurity deductions, and provides hazardous waste and foreign object warnings, halting loading at millisecond speed, which can improve on-site efficiency and reduce customer complaints and disputes.

3. Lessons include prioritizing training over inference, edge over cloud, and vertical models over general models. Let people familiar with the industry write code, and send algorithm engineers to dismantling sites and weighing stations. The only test is whether algorithm outputs can drive business actions, preventing technology and business from becoming two separate layers.

Production Quality Inspection and Design Requirements

1. Scrap AI visual quality inspection and grading covers the full process of plant entry, loading, and transportation, precisely identifying material types and providing real-time warnings on hazardous waste and foreign objects, turning non-standard goods into standardized components. Intelligent on-site control analysis uses surveillance video for behavior recognition and dynamic tracking of cargo status.

2. AI valuation can extract features such as equipment model, condition, and wear level, and combine them with time-series price models to assess equipment value. During loading, Feimao Yikong IoT cameras identify material types, calculate impurity deductions, and warn of hazardous waste and foreign objects.

Business Opportunities and Digitalization Insights

1. Attention can be paid to whole-plant asset disposal, industrial demolition projects, steel mill scrap procurement agency services, and cross-border disposal of used industrial equipment. The large steel mill project in Hebei fully validated scrap AI grading, price forecasting, and AI valuation, indicating commercial space for full-site disposal.

2. Digitalization should not upload all video streams to the cloud. Lightweight CV inference models can be deployed to local IoT servers on site, with edge computing focused on vision and cloud computing focused on training iteration: prioritize training over inference, edge over cloud, and vertical models over general models.

3. Algorithm engineers must get their boots muddy by visiting dismantling sites and following trucks to weighing stations to understand on-site interference such as wet scrap and oil-stain reflections. Digitalization should be judged by whether algorithms can drive commercial, logistics, and capital flows; AI that cannot drive business actions is ineffective.

Industry Trends and New Technologies

1. Recyclable resources are a non-standard vertical industry and are transitioning from rule by experienced masters to digital governance. AI and overseas expansion are jointly driving the transformation of business modalities. B2B platforms are upgrading from providing tools to delivering results, and industry agents are becoming digital business experts that can inspect goods, grade them, forecast prices, match counterparties, and schedule logistics.

2. New technologies include AI visual scrap quality inspection and grading, intelligent asset valuation based on real transaction data, intelligent on-site control analysis, and multi-factor market forecasting. Computing power adopts a dual-track approach: edge computing focuses on CV vision, while cloud computing focuses on time-series forecasting and model training.

Customer Pain Points and Solutions

1. Customer pain points include large subjective deviations in quality inspection, inaccurate asset valuation, and lack of data support for risk control; cross-border transactions lack assessment, dismantling, price protection, and transaction fulfillment links; project sites also face difficulties in pricing, quality inspection, and settlement.

2. Solutions apply AI vision across the full process of plant entry, loading, and transportation to identify material types in real time, calculate impurity deductions, warn of hazardous waste and foreign objects, and halt loading at millisecond speed. Before listing, time-series models output starting prices and sales timing; CV extracts equipment features and combines them with price models for valuation.

3. Data solutions emphasize native closed loops, moderate openness with compliance first, and iron-walled tiered desensitization; raw business data does not leave the platform. Service providers can see the implementation value of vertical models, IoT hardware, and edge deployment.

Platform Needs and Latest Practices

1. Industry pain points are non-standardization and reliance on manual governance, subjective quality inspection deviations, inaccurate valuation, and lack of data for risk control; cross-border transactions lack assessment, dismantling, price protection, and fulfillment links; disputes over material-grade price differences between upstream and downstream require platform intervention.

2. Feimao Net positions itself as an industry digital expert deeply involved in transaction fulfillment rather than simple information matching. Through four major segments—full-site KA turnkey disposal, online auctions, steel mill scrap procurement agency services, and member value-added services—it integrates the five flows and uses AI visual quality inspection, intelligent valuation, on-site control, and market forecasting to improve fulfillment.

Operations Management, Cooperation, and Risk Avoidance

1. Cooperation and operations can center on member value-added services, online auctions, steel mill scrap procurement agency services, and full-site disposal. Industry agents can automatically match counterparties and schedule logistics, upgrading the platform from a tool to a result-delivery service.

2. Data operations adhere to a native closed loop, with nearly RMB 100 billion in real transaction records forming a moat. Cross-party collaboration is moderately open and compliance first, sharing data value rather than data itself; iron-walled tiered desensitization is established, and raw business data does not leave the platform.

3. Risk avoidance should guard against sample scarcity in non-standard sectors and B2B privacy compliance risks. Computing power should prioritize training over inference, edge over cloud, and vertical models over general models. Avoid the disconnect between technology and business, let people familiar with the industry write code, and use whether algorithms can drive business actions as the test.

New Industry Dynamics and New Issues

1. Recyclable resources are moving from manual governance to digital governance, with AI and overseas expansion bringing a transformation of business modalities. Retired used industrial equipment in China is being revalued in emerging overseas industrial markets, and cross-border disposal has become a new direction. Industry agents are pushing B2B from providing tools to delivering results, and the value of vertical large models lies in breakthrough performance in a narrow domain rather than general omnipotence.

2. New issues include sample scarcity caused by non-standardization, B2B commercial data privacy and compliance risks, general large models' lack of industry understanding and high token and computing costs, and the fact that industry digitalization projects often look technologically impressive but are difficult to implement in business terms.

Business Models and Governance Implications

1. The business model is an industry digital expert deeply involved in transaction fulfillment, relying on full-site KA turnkey disposal, online auctions, steel mill scrap procurement agency services, and member value-added services to form a closed loop integrating the five flows. Data competitiveness comes from a native closed loop, accumulating nearly RMB 100 billion in real transaction records and on-site images and videos.

2. The AI moat consists of three lines of defense: closed-loop native data, scenario fit, and credibility. The dual-track computing approach prioritizes training over inference, edge over cloud, and vertical models over general models.

3. Data governance implications are moderate openness and compliance first, not touching or exchanging raw sensitive business data, iron-walled tiered desensitization, and keeping raw business data off-platform. Implementation implications are to let people who understand the industry write code, ensure algorithm engineers get their boots muddy, and test AI by whether it can drive business actions.