Key Lessons from AI Commercialization Across Industries

After a four‑month hiatus, the author updates readers on his transition to a new department focused on AI commercialization at Alibaba Cloud. He defines AI commercialization as deploying Alibaba’s AI products and solutions across industries to solve real business problems, reduce costs, or create new business models, while also incubating new offerings based on market insights.

AI adoption gaps in industry

Understanding business needs is difficult because AI product teams often lack domain expertise and customers struggle to articulate requirements, leading to misaligned conversations.

Demonstrating business value is hard; mapping customer problems to AI capabilities frequently feels like two separate layers, and many requirements are hard to quantify, causing acceptance challenges.

Clients have inflated expectations of AI and insufficient awareness of its current limitations, while sales sometimes sell vague concepts without measurable KPIs, resulting in mismatched expectations.

End‑to‑end systems are complex for B2B customers; balancing standardization versus customization during product definition is a major challenge, as is integrating AI solutions with existing enterprise systems.

AI technology maturity remains low for many complex, data‑intensive scenarios, highlighting the skill gap between seasoned engineers and newcomers.

Differences between AI software and traditional SaaS

The author notes that SaaS enjoys a high gross margin (60‑80%) due to its "produce once, sell many times" model, whereas AI software typically sees lower margins (50‑60%). He cites a16z’s February 2023 article “The New Business of AI (and How It's Different From Traditional Software)” to illustrate two main challenges.

Hidden costs : AI projects incur substantial infrastructure expenses for offline model training, large‑scale data storage, and continuous model optimization, as well as operational costs for data ingestion, cleaning, quality assurance, model retraining, and building feedback loops. These costs often outweigh the revenue advantage of AI solutions.

Generalization difficulty : Unlike deterministic SaaS logic, AI models must generalize across long‑tail scenarios, handle data drift, and provide explainability—especially for mission‑critical applications. Balancing model effectiveness with broad applicability adds technical and cost burdens.

Data‑driven moat skepticism

The author questions the notion that massive industry data automatically creates a sustainable moat. He argues that acquiring large, high‑quality datasets in non‑internet sectors is costly and slow, and that beyond a certain point, additional data yields diminishing returns or even negative returns due to increased processing overhead.

What can be done?

Despite the challenges, the author sees AI commercialization as a major future trend with significant commercial potential. He recommends aligning with market needs, identifying gaps, defining problems clearly, and delivering concrete value. He emphasizes the importance of deep industry understanding, integrating AI with core business processes, and turning uncertainty into actionable solutions.