千峰访谈数阔黄树煜：数据、AI与自动化连成一条链路之后

Overall: This article explains how Shukuo connects data, AI, and automation into an end-to-end pipeline, and how software will increasingly be invoked directly by AI agents.

1. End-to-end pipeline: data acquisition → AI understanding → business judgment → automated execution. Companies can use AI to collect data, detect anomalies, and push alerts, while humans focus on key decisions such as repricing and replenishment.

2. Quantifiable cases: Xiaolv AI reduced product selection-to-launch from a three-person workflow to a one-person operation, raising new listings per store per week from roughly 10 to 50. In auto parts, client Sangshu uses RPA and AI to monitor Amazon prices, reviews, rankings, and competitor promotions.

3. How to judge AI value: not by the number of models or token consumption, but by whether it reduces manual work, improves conversion, and shortens decision cycles.

4. Future software form: users will not need to learn complex menus; they will set goals for agents in natural language, and agents will call tools to complete tasks such as review analysis, competitor selling-point research, and product syncing.

Overall: The most valuable takeaway for brands is using consumer insights and competitor data to drive product, pricing, and operations decisions.

1. Consumer insight: Yunting CEM processes reviews, customer-service conversations, surveys, and social media content daily. Large models can directly understand what users say, why they are dissatisfied, and how demand is changing, helping brands adjust product and communication strategies.

2. Price and competition monitoring: Cross-border auto parts client Sangshu uses RPA to continuously collect Amazon prices, reviews, rankings, Buy Box status, and competitor promotions; AI filters noise and escalates alerts by priority, enabling operators to adjust prices and replenish stock in time.

3. Consumption trends and launch efficiency: In the Xiaolv AI case, AI assesses keyword trends, category size, competition, and margin potential, raising new listings per store per week from 10 to 50, showing that data-driven selection can respond to demand more quickly.

4. Data competitiveness: what brands truly lack is not models but real internal business data and changing external market data; sustaining high-quality, real-time, business-relevant data is the long-term advantage.

Overall: This article offers sellers practical takeaways on e-commerce operations automation, product selection and listing efficiency, cross-border platform monitoring, and new opportunities created by AI agents.

1. Replicable efficiency gains: Xiaolv AI uses seven or eight digital employees to automatically collect keyword trends, competitor activity, sales, and ad-spend data in the early morning. AI evaluates trends and margin potential, then RPA assembles and publishes product listings, reducing a three-person workflow to one person and increasing new listings per store per week from 10 to 50.

2. Cross-border platform risk response: Sangshu's pipeline uses RPA to collect Amazon prices, reviews, rankings, Buy Box status, and competitor promotions; AI filters anomalies and pushes tiered alerts to operators, helping sellers react faster to pricing changes, replenishment, and negative reviews.

3. Growth markets and opportunities: Shukuo focuses on North America, Europe, Japan, and South Korea. Many AI agents and applications globally lack real-time, structured, sustainably obtainable external data, creating new demand for cross-border e-commerce data services.

4. Risk note: vertical large models are not necessarily essential; what matters is the combination of large models with accurate data, industry know-how, and agents. Sellers should evaluate whether a tool can actually complete the end-to-end task.

5. Collaboration model: capabilities are being opened to AI agents through API, MCP, and Skill. Sellers will soon be able to converse directly with agents to complete competitor analysis, negative-review attribution, and product syncing.

Overall: The key takeaway for factories is to use data and AI to identify real product demand, then use automation to standardize e-commerce listing and operations processes.

1. Product-demand insight: Yunting CEM analyzes reviews, customer-service conversations, surveys, and social media content daily. Large models can directly understand why users are dissatisfied and how demand has shifted, enabling factories to use external market data for product improvement and design.

2. Product selection and launching opportunities: In the Xiaolv AI case, AI analyzes keyword trends, category size, competition, and margin space, then automatically assembles product information and visual assets for publishing. New listings per store per week increase from 10 to 50, showing that production can iterate quickly based on real-time demand.

3. Digitalization and e-commerce implications: a practical pipeline is data acquisition → AI understanding → business judgment → automated execution. Even without in-house development capability, factories can integrate these capabilities into production and e-commerce processes through API, MCP, and Skill.

4. Cross-border path: exporting digital products does not require building an overseas team; the keys are product standardization, an online end-to-end commercial pipeline, and market-by-market operations. Factories going cross-border in e-commerce can adopt this model to reduce marginal expansion costs.

Overall: This article helps service providers understand the new technology stack in the AI-agent era, customer pain points, solution design, and how to turn existing capabilities into callable services.

1. Industry trend: software is shifting from menus, pages, and buttons to a flow in which users state goals, agents understand tasks, call tools, and return results or execute directly. Software must not only serve humans; it must also be discoverable, understandable, and callable by agents.

2. New technology levers: Shukuo exposes data collection, VOC analysis, and RPA execution to AI agents through MCP, Skill, and API. Service providers can package existing capabilities as tools and embed them into customers' AI platforms and workflows.

3. Customer pain points: what customers lack is not models but real internal business data and real-time external market data. Many AI applications are short of real-time, structured, and sustainably obtainable external data—a clear supply gap.

4. Solution examples: Sangshu uses RPA to continuously collect data, AI to filter anomalies, classify issues, and route alerts, and bots to send tiered alerts to operators. Xiaolv AI uses seven or eight digital employees to compress the selection-to-launch process from a three-person workflow to one person.

5. Measuring value: the success of an AI project is not judged by token consumption or model count, but by whether it reduces manual work, raises conversion, and shortens decision cycles.

Overall: The lesson for platforms is that AI agents are becoming new users; platforms must satisfy the need to be discovered, understood, and called by agents, while also providing merchants with real-time data tools.

1. New platform demand: many AI agents and applications globally lack real-time, structured, and sustainably obtainable external data. If platforms can open or aggregate data capabilities, they can create new service opportunities.

2. Platform approach reference: Shukuo is opening data, analytics, and automation capabilities through API, MCP, and Skill for various AI agents to call directly. Platforms should consider a similar open strategy rather than locking users inside their interfaces.

3. Merchant operations and management: sellers need to monitor price, reviews, rankings, Buy Box status, and competitor promotions on platforms such as Amazon. Platforms can use data interfaces or automation tools to help merchants handle anomaly alerts and reduce operational risk.

4. Risk avoidance and ecosystem recruitment: vertical large models do not necessarily create a moat; the real moat lies in data, business understanding, tools, and the ability to execute tasks. When building an ecosystem, platforms should prioritize service providers that fill gaps in these capabilities.

Overall: This article describes a new trend in which AI agents reshape the software industry and cross-border expansion models, and provides concrete cases and judgments for industrial internet research.

1. New industrial trend: AI plus globalization is not simply adding two business lines; it recombines existing data collection, VOC, and RPA capabilities into a pipeline of data acquisition → AI understanding → business judgment → automated execution. Software moves from a human-facing interface to an agent-callable capability.

2. Business-model innovation: digital products expanding overseas do not need to build sales and service teams in every country. Instead, they turn core capabilities into foundational services that the global AI ecosystem can directly call, lowering marginal expansion costs. Typical markets are North America, Europe, Japan, and South Korea.

3. Role of data as a factor: companies truly lack internal real business data and external real-time market data. Future AI competitiveness depends on the ability to continuously supply high-quality, real-time, business-relevant data.

4. Industry agents and B2B judgment: be cautious about vertical large models; foundation models will converge. The more valuable combination is large models plus accurate data, industry know-how, and agents. The moat in industrial AI lies in data, business understanding, tools, and completing tasks end to end.

5. Policy and platform implications: pay attention to interoperability standards such as API, MCP, and Skill; encourage service providers to open capabilities; regulate cross-platform data collection and use to avoid data monopolies and competitive risks.