While nearly every company discusses AI, few have successfully transformed it into a genuine productivity driver. Over the past two years, large models have iterated monthly, and AI Agents have become a buzzword chased by both capital markets and industry. However, within real enterprises, a different reality is equally common: numerous pilot projects and demos exist, but few AI applications are integrated into actual workflows to deliver sustained business results. Huang Xiaonan, Founder and CEO of DEEPZERO (SEHK: 02723), summarizes this issue in one sentence: the bottleneck in realizing enterprise AI value lies not in the models themselves, but in the methods of using AI.
In May 2026, DEEPZERO listed on the main board of the Hong Kong Stock Exchange, becoming the first "Enterprise Decision AI Agent" stock on the HKEX. Post-listing, the company proposed "reconstructing enterprise AI operations with Agentic Software + Agentic Service." This is not a simple product slogan but a comprehensive, operating-system-like methodology for enterprise AI implementation: advancing AI from a point tool into processes and simultaneously AI-fying the external service chain.
From Decision AI to the "Golden Wings"
Huang Xiaonan repeatedly emphasizes that DEEPZERO is not a company that pivoted to AI only with the generative AI boom. "Our business plan back in 2008 already stated that data and algorithms would change advertising delivery. Today, it's called AI-empowered decision-making. The terminology is different, but the core has never changed," she said in an interview. In her view, the company's initial focus on ad delivery was not fundamentally about "advertising" but about making high-frequency, complex, real-time marketing decisions: determining which content to show to which individual when a company faces hundreds of millions of users, thousands of creatives, and different times and channels. Such problems are impossible for the human brain to solve alone, requiring reliance on data, algorithms, and predictive models.
Publicly disclosed R&D investments corroborate this continuity. The prospectus reveals that DEEPZERO's AlphaDesk began development in 2011, and AlphaData in 2017, each handling intelligent ad delivery and intelligent data management capabilities, respectively. Prior to the track record period, the company had invested over RMB 300 million in these two platforms and continued to invest RMB 156.2 million from 2023 to 2025. These figures indicate that DEEPZERO's so-called "AI DNA" is not a post-IPO conceptual packaging but a continuous evolution from predictive AI, machine learning, and deep learning to large models and multi-agent systems. Huang Xiaonan mentioned in the interview that the company's English name, Deep Zero, was also inspired by Deep Learning and Alpha Zero, pointing to a technological belief in self-learning and continuous iteration through algorithms.
The significance of generative AI for DEEPZERO is not an ordinary technological upgrade but an industry-revolutionary amplification. "We believe this wave of generative AI is industry-revolutionary, not just a technological iteration," Huang Xiaonan said, likening it to "giving golden wings" to the data, algorithms, models, and client scenarios accumulated by DEEPZERO over the past seventeen years.
The value of these "golden wings" lies in enabling the company's previously hard-to-scale enterprise-level capabilities to achieve higher-efficiency productization for the first time. In the past, the company accumulated data, models, and industry expertise in high-value decision-making scenarios like digital advertising and CRM. Now, the language understanding, intent recognition, content generation, and Agent orchestration capabilities of large models allow these accumulations to extend to more scenarios like product innovation, GTM, social marketing, sales training, intelligent shopping guidance, and user operations.
Agentic Software: From Task to Job
Huang Xiaonan believes many enterprise AI projects fail to deliver significant value because they only complete Tasks ("point tasks") rather than Jobs (complex work in the full sense). An AI tool can help write a piece of copy, judge content compliance, generate a user tag, or answer a customer query. However, in real enterprise work, business objectives are typically not isolated actions but complete processes: insight, strategy, content, review, publishing, delivery, recycling, and post-analysis. Each step must connect to context, data, and organizational rules.
This is precisely the context for DEEPZERO's proposal of Agentic Software. Huang Xiaonan explains it with a vivid analogy: on the surface, it still looks like a software process, but robots are working beneath the process. When a human clicks a node, multiple Agents in the background may complete information reading, task decomposition, content generation, model judgment, compliance verification, and result feedback.
For example, in sales training scenarios, traditional systems often only provide questions and exercises. In contrast, DEEPZERO's Agentic Software first uses AI to generate training scenarios and customer personas, sets evaluation criteria, and then immerses sales personnel in simulated dialogues. During the dialogue, the system can instantly judge whether the sales pitch reduces conversion rates, improves customer sentiment, or conveys accurate product knowledge. Behind a seemingly simple software interface, over a dozen intelligent agents may be collaborating.
This is also the core logic behind the company's aim to replace traditional marketing software: not adding an AI assistant beside old software but embedding AI into the process itself, transforming software from a "collection of functions" into a "task completion system."
Agentic Service: AI-fying External Services
If Agentic Software addresses the AI-fication of internal "self-operation," then Agentic Service tackles the AI-fication of externally procured services. Enterprises do not complete all marketing work internally. Many aspects, such as social media marketing, content seeding, ad delivery, influencer selection, traffic boosting, and performance analysis, have long been handled by external agencies or service providers.
Huang Xiaonan argues that if a company only AI-fies its internal software while external services still rely on traditional human delivery, the value of AI remains incomplete. Therefore, DEEPZERO proposes Agentic Service: clients may not necessarily purchase and operate software but purchase service outcomes driven by AI.
Taking social media marketing as an example, the complete chain includes competitor analysis, strategy generation, content creation, KOL selection, traffic boosting, and performance analysis. Traditional agencies rely on human experience and execution, making service quality susceptible to fluctuations in team capabilities. The logic of Agentic Service is for experts to set the direction, with AI agents completing most of the execution and optimization, ultimately delivering results to the client.
This logic is consistent with DEEPZERO's past advertising delivery business. Advertisers care not about how the service provider delivers but whether they can acquire higher-quality new customers, achieve better conversions, and maintain more controllable KPIs under the same budget. The source of DEEPZERO's value is not reselling traffic but using AI to achieve equal or superior results at lower costs.
Palantir-like Enterprise AI Implementation Capability
Where is the moat for AI application companies? Huang Xiaonan's answer is not mysterious: vertical industry data assets, vertical industry know-how, and the capability to implement systems within enterprise scenarios. This is also what she emphasizes when discussing the commonalities between DEEPZERO and Palantir.
The market often uses Palantir to analogize a type of enterprise AI company: they do not merely sell standardized software, nor are they traditional consultancies. Instead, they deeply immerse themselves in clients' business environments, connecting data, models, processes, and organizational decisions to ultimately form sustainable, operational system capabilities.
Huang Xiaonan believes the similarity between DEEPZERO and Palantir lies not in serving identical industries but in sharing a similar methodology for enterprise AI implementation. First, there must be people who can truly understand the client's core business pain points. Second, product capability is essential; it cannot remain at purely customized project levels. Third, implementation and delivery teams are needed to truly embed AI products into client processes.
In Palantir's context, such roles are often called Forward Deployed Engineers (FDE). Within DEEPZERO, they are more inclined to call them AI Product Managers or AI Solution Teams. These individuals must understand business, products, and AI, capable of collaborating with clients to identify which needs have real value and which scenarios are suitable for AI solutions.
This is the hardest-to-replicate aspect of enterprise AI. Point tools can be quickly developed, and general models will continuously upgrade, but deeply understanding major clients' data structures, organizational processes, budget systems, and business objectives, and transforming AI from a demo system into a deliverable, reviewable, and sustainably iterative production system requires long-term accumulation.
For DEEPZERO, seventeen years of B2B service experience, marketing and sales scenario know-how, and the product and delivery system formed around AlphaData, AlphaDesk, and DeepAgent constitute the foundation of its Palantir-like capability. The competition in enterprise AI ultimately is not about who tells a better model story, but about who can integrate AI into real enterprise workflows and make clients willing to pay continuously. DEEPZERO aims to prove whether a company starting from advertising decision-making can become the AI operating system for enterprise marketing, sales, and user operations in the Agentic AI era.
Comments