Last June, I compiled an article about JD’s AI supply chain case shared at Stanford University. My impression then was: this is a solid record of internal corporate transformation, worth learning from but somewhat distant from us.

Nine months later, my assessment has changed. Not because JD has done anything shocking, but because what they’ve done is becoming a “standard”—when AI-driven supply chains evolve from competitive advantage to infrastructure, companies that don’t keep up aren’t just “missing a tool,” they’re “missing a layer of infrastructure.”

Taiwan’s manufacturing and service industries need to seriously confront this transformation.

From Stanford Podium to Academic Paper: The JD Model Gets Formally Documented

First, a noteworthy signal. In 2025, Stanford Business School’s supply chain management guru Hau L. Lee and UC Berkeley professor emeritus, now HKU Vice President Zuo-Jun Max Shen, wrote JD’s AI supply chain practices into a formal academic paper, published on SSRN. The paper analyzes three specific JD cases using the classic “Triple-A” framework—Agility, Adaptability, and Alignment.

What does this represent? It means JD’s approach is no longer just “internal corporate optimization” but has been recognized as an industry template with universal value. The leap from corporate presentation to academic case study is significant.

Three Layers: Prediction, Explanation, Optimization

JD’s AI supply chain isn’t one system—it’s a stack of three-layer capabilities.

The first layer is intelligent prediction. JD has shopping behavior data from over 700 million users, real-time inventory from over 1,600 self-operated warehouses, plus external variables like weather, holidays, and geopolitics. According to JD’s published data, prediction models trained on this data improved accuracy by nearly 15%. More critically, they use “synthetic data”—deep learning simulations of successful historical transactions—to enhance training samples, allowing models to maintain stability when facing unprecedented market volatility.

The second layer is explainability. This is what many companies overlook when implementing AI. No matter how accurate predictions are, if business units don’t trust, understand, or dare use them, it’s all for nothing. JD breaks down every prediction result into “baseline value + contribution values of various factors,” making it comprehensible to sales, marketing, and logistics departments. They even provide three promotional strategies for individual products, each corresponding to different costs and expected benefits, letting brands choose for themselves.

The third layer is logistics optimization. Traditional logistics optimization requires specialized Operations Research (OR) background—complex modeling, difficult adjustments. JD has also bridged this layer with natural language—you input “deliver a thousand items to three warehouses at lowest cost,” and the system automatically converts this to mathematical models, pulls latest transportation costs, and generates solutions. It can also do What-if analysis: if a warehouse fails or a route is disrupted, what are the alternatives?

These three layers together don’t represent “AI helping you work”—they represent “AI helping you think.”

Nine Months Later: From Internal Tool to External Platform

When I wrote the original piece last year, JD’s AI supply chain primarily served itself. Now it has clearly moved toward platform-based export.

After the “Logistics Brain 2.0” upgrade, it’s no longer just an automation tool. Through global perception, model evolution, and human-machine collaboration mechanisms, it transforms workflow processes that previously relied on human experience into data-driven dynamic decisions. According to JD’s published data: operational standardization improved 15%, personnel-vehicle-cargo-facility scheduling efficiency improved nearly 20%, and human-machine collaboration efficiency improved over 20%.

In robotics, JD’s “Wolf Pack” series products—Smart Wolf, Sky Wolf, Ground Wolf, Flying Wolf, Lone Wolf, Wing Wolf—plus automated sorting walls have been deployed in over 500 warehouses globally according to JD. In July 2025, JD also released its self-developed unmanned light truck “JD Logistics VAN,” with 400km full-load range and L4-level public road autonomous driving capabilities.

More noteworthy is the business model transformation. JD Industrial (JDi) received Hong Kong Stock Exchange IPO approval in 2025, positioned as an “industrial supply chain technology and service platform,” with first-half revenue of 10.3 billion RMB, up nearly 19% year-over-year. This essentially packages JD’s accumulated AI supply chain capabilities into products for sale to other companies.

Meanwhile, JD Logistics’ international brand JoyExpress has officially entered Europe (UK, Netherlands, Germany, France), JoyLogistics is deepening its presence in the Middle East, signing strategic cooperation with the FII Institute covering digital transformation, logistics infrastructure, and cross-border commerce. JD made Gartner’s Global Supply Chain Top 25 for the second consecutive year.

The model is clear: first use AI to optimize your own supply chain, accumulate data and models, then transform this capability into an external platform. Supply chain is no longer just a cost center—it’s a profit center.

Why Taiwanese Companies Should Care

At this point, some might say: JD is a Chinese e-commerce giant with 700 million users and over 1,600 warehouses—what does this have to do with Taiwan’s SMEs?

The relationship is significant, but it’s not about imitating JD—it’s about understanding trends.

The first signal: AI supply chain capabilities are moving from “build” to “rent.” JD Industrial’s IPO, SaaS-based supply chain services, API-enabled logistics interfaces—all point to the same thing: future SMEs won’t need to build their own AI teams but will plug into others’ pre-built AI supply chain platforms. The question is: whose platform will you plug into?

The second signal: explainability determines whether AI can truly land. JD invested enormous effort in “making business units understand AI decisions”—this isn’t a technical problem, it’s a trust problem. Many Taiwanese companies’ AI implementations fail not because models are inaccurate, but because frontline people don’t trust or dare follow AI recommendations.

The third signal: supply chain competitive dimensions are changing. Previously, competition was about “whose logistics is faster, cheaper.” Now it’s about “whose predictions are more accurate, reactions faster, anomaly handling more automated.” When JD’s AI system can automatically reallocate routes the instant a warehouse fails, this isn’t an efficiency advantage—it’s a structural advantage.

For Taiwan’s manufacturing and supply chain service providers, the real question is: when your customers start interfacing with you using AI-driven supply chains while you’re still managing inventory with Excel, you’re not “slower”—you’re speaking different languages.

The Essence Seen Through JD

Back to the core of the JD case. When they shared at Stanford last year, they emphasized four success conditions: cross-disciplinary talent, leadership support, flat organizations, and massive data.

Looking again nine months later, I think the most important is actually just one: whether organizations are willing to let AI participate in decision-making, not just generate reports.

Many companies’ AI implementations stop at “AI does analysis, humans make decisions.” JD has reached “AI makes recommendations, humans choose options,” and in some areas even “AI makes decisions, humans monitor anomalies.” The gap between these three stages isn’t a technology gap—it’s a trust gap.

The JD case tells us: AI supply chain isn’t something you can achieve by buying software and hiring a data scientist. It’s a systematic transformation involving organizational culture, decision processes, and data governance. Technology is the easiest part, talent is harder, but hardest is getting an organization accustomed to “humans have final say” to learn to trust data and models.

This challenge—regardless of company size or industry—every organization must face.