祝融说。

In the previous three parts, we have systematically completed a panoramic depiction of the computing platform from top-level design and operations system construction to core technical support. This is like having built an advanced, efficiently managed "information highway." However, the value of a highway lies not in itself, but in the "cars" running on it — the applications that truly carry cargo, transport people, and create economic activities. If the traffic on the highway is sparse, then no matter how good the road, it is a waste of resources.

Therefore, starting from this part, we enter the "second half" of computing platform construction, and the most critical stage of value realization — value realization and ecosystem expansion. Our focus will shift from "how to build a good platform" to "how to use the platform well," from focusing on the achievement of technical indicators to focusing on the creation of business value, from serving internal users to building an open and win-win ecosystem. This is the necessary path for the computing platform to leap from a "cost center" to a "value center" and "innovation center."

This part is divided into three chapters, respectively exploring application promotion, cost operations, and ecosystem construction. As the opening chapter, Chapter 9 focuses on computing application promotion and knowledge precipitation. The birth of a new platform is often accompanied by user cognitive gaps and usage inertia. How to make potential users "know" of the platform's existence, "understand" its capabilities, "be willing" and "good at" using the platform — these are the core propositions of application promotion. At the same time, during the promotion and application process, a large amount of successful experience, lessons from failures, technical know-how, and solutions are generated. How to systematically distill these precious "tacit knowledge" scattered in individual minds into "explicit assets" shared by the organization is the core challenge of knowledge management. These two efforts together constitute the two key engines for the "soft landing" of platform value.

Multi-Channel Application Promotion Strategy

No matter how good a product is, if it is "raised deep in an inner chamber, unknown to the world," its value cannot be realized. For a new thing like the computing platform, we must adopt proactive, multi-channel, three-dimensional promotion strategies to break information barriers, ignite user enthusiasm, guide applications onto the platform, and ultimately form a virtuous cycle of "the platform is good to use, users love to use, and applications flourish." Our promotion strategy follows the classic AIDA model — Awareness, Interest, Desire, Action — and designs different promotional "combinations" for each stage.

"Spread the Word": Expand Awareness, Build the Brand (Awareness)

In the initial stage of promotion, the primary goal is to let as many potential users within the entire company as possible — including business personnel, IT personnel, and management — know that "we now have a unified, powerful AI computing platform."

Authoritative Release and High-Level Endorsement

• Company-level formal document: Issue a formal red-head document from the group company to all levels, announcing "Notice on the Activation of the Company's Unified AI Computing Platform." This document carries the highest authority and is the "identity certificate" for the platform's legitimacy. The document will clearly state the platform's positioning, objectives, management units, and activation time.
• High-level leadership endorsement: At the company's annual work conference, digital transformation conference, and other high-level meetings, invite senior company leaders (such as the deputy general manager in charge of informatization) to specifically speak about the strategic significance and construction achievements of the AI computing platform. Leadership attention and endorsement are the strongest drivers for top-down promotion.

Online and Offline Three-Dimensional Publicity

• Online publicity matrix:
  • i State Grid column: On the company's internal "i State Grid" portal website, open a "Smart Computing Empowerment" column, regularly publishing news updates on platform construction, technical interpretations, policy notices, etc.
  • Corporate WeChat public account/video account: Produce exquisite promotional articles, short videos, and science animations, introducing the platform's core functions and value in a more vivid and lively format. For example, produce a 60-second short video quickly showcasing the entire process of a user from applying for resources to submitting a training task.
  • Promotional posters and e-handbooks: Design a series of visually striking promotional posters to display in public areas such as office areas and elevator lobbies. At the same time, produce a well-illustrated "Computing Platform User Quick Start Manual" (electronic version), widely distributed via email and WeChat work groups.
• Offline publicity activities:
  • Platform launch/inauguration ceremony: Host a grand platform launch event, inviting leaders from various units, technical backbone, and media representatives to attend, creating momentum and forming a "tipping point."
  • "Computing Platform Open Day": Regularly hold open day events, inviting employees to visit the intelligent computing center's physical machine room, experiencing the awe of a large-scale computing cluster firsthand, enhancing their intuitive understanding and sense of pride in the platform.

"Cater to Their Interests": Spark Interest, Precise Targeting (Interest)

After the broad publicity, we need to target different groups with language they "understand and are interested in," precisely conveying the specific value of the platform to them, sparking their interest in learning more.

Value Refinement and Presentation for Different Roles

• For business experts:
  • Promotional focus: Don't talk about technical details; focus on "what problems AI can solve for your business." We organize "AI Empowering Business" series salons, inviting industry experts and internal pioneers to share successful AI application cases in transmission, substation, marketing, safety supervision, and other fields.
  • Promotional language: "In the past, it took a veteran a week to look through drone inspection photos; now, AI models can automatically identify defects in minutes with even higher accuracy."
• For algorithm engineers/developers:
  • Promotional focus: Focus on the platform's "hardcore" technical capabilities and improvement to development efficiency. We hold events such as "Technology Open Day" and "Developer Meetup."
  • Promotional language: "Still troubled by failing to get GPUs and spending days configuring the environment? Our platform provides a one-click JupyterLab environment, a massive pool of A100/H100 resources, supporting PyTorch 2.0 and the latest parallel computing frameworks. Ignite your ideas instantly."
• For managers:
  • Promotional focus: Focus on the platform's management value — cost reduction and efficiency improvement, resource coordination, security, and compliance. We provide dedicated presentation materials and briefings for the management of each unit.
  • Promotional language: "Through a unified computing platform, we can increase the company-wide AI computing resource utilization from below 20% to over 60% (per internal operations statistics), avoiding substantial duplicate investment and operations costs."

"Delivery of Training to Your Doorstep": Systematic Empowerment Training

After interest is sparked, there must be follow-up empowerment training to help users move from "wanting to use" to "knowing how to use."

• Tiered and categorized training course system:
  • "Literacy" universal class (for all employees): "Basic Knowledge and Application Prospects of Artificial Intelligence"
  • "Introduction" hands-on class (for beginner users): "Computing Platform Usage Guide: From Entry to Practice"
  • "Advanced" class (for senior developers): "Large Model Distributed Training and Performance Optimization in Practice," "In-depth Analysis of TensorRT Inference Optimization"
• Flexible and diverse training formats:
  • Headquarters centralized training: Regularly organize offline centralized training courses, providing systematic learning and in-depth exchange opportunities.
  • Provincial roadshow: Organize "Smart Computing Service Light Cavalry," going deep into provincial companies for roadshow presentations and on-site Q&A, bringing services to the frontline.
  • Online learning platform: Record all courses as videos and upload them to the company's online learning platform for employees to learn anytime, anywhere.
  • Establish a certification system: Launch certifications such as "Computing Platform Certified Engineer" (Junior/Senior), testing users' learning outcomes through exams and practical projects, and providing incentives.

"Using Points to Drive the Whole": Set Benchmarks, Create Desire (Desire and Action)

The power of example is infinite. Rather than us saying over and over how good the platform is, let a successful user speak from experience. By creating and promoting benchmark cases, we can push users from "interested" to "eager to have" and ultimately motivate them to "take action."

"Open Competition": Computing Application Innovation Contest

• Activity format: Regularly hold an "'AI Computing Cup' Artificial Intelligence Application Innovation Contest" open to the entire company.
• Collect challenges: Solicit real business pain points from frontline business departments as competition topics.
• Provide resources: Provide all participating teams with free, ample computing resources and technical support.
• Selection and awards: Invite internal and external experts to form a judging panel, comprehensively evaluating competition works on "technological innovation," "business value," and "promotion prospects," and provide generous material and honor rewards (such as cash prizes, promotion bonus points, project incubation funds).
• Activity value:
  • Discover applications: Through the contest, a large number of highly promising innovative applications can emerge.
  • Discover talent: Can discover and cultivate AI talent within the company.
  • Create buzz: The contest itself is an excellent promotional event, and the winning works are the best "living advertisements."

Establish Seed User and "Early Adopter" Programs

• Selection and invitation: In the early stages of platform construction, we proactively select and invite teams with strong technical capabilities, good business scenarios, and high willingness to cooperate as the platform's "seed users."
• Key support: We provide "nanny-style" VIP services for seed users, including dedicated technical support experts, priority resource assurance, and in-depth performance tuning, ensuring their projects are successful from the first battle.
• Feedback to the platform: The success of seed users provides us with the first batch of valuable benchmark cases. At the same time, the feedback and suggestions they raise during early use are the most valuable input for platform optimization and iteration.

Through these multi-channel, phased, and rhythmic promotional strategies, we have successfully transformed the computing platform from an unfamiliar technical term into an "innovation source" known to everyone within the company and coveted by core users, paving the way for large-scale application migration onto the platform.

Benchmark Cases: Compilation and Promotion of Computing Operations Results

Benchmark cases are the most powerful "ammunition" in the promotion strategy. They are not empty propaganda but visible, tangible demonstrations of results. We have established a standardized process from discovery, packaging, to promotion, systematically transforming operational results into influential "stories."

Selection Criteria for Benchmark Cases

A good benchmark case should possess several of the following characteristics:

• Significant business value: Can genuinely solve business pain points, bringing quantifiable efficiency improvements (e.g., defect identification rate improved by XX%, labor costs saved by XX yuan).
• Technologically representative: The applied technical solution (such as large model fine-tuning, multimodal recognition) has a certain degree of advanced and representativeness, reflecting the platform's technical support capability.
• Replicable and promotable: The successful experience and model of this application have strong reference significance and promotion value for other units or similar scenarios.
• Storytelling and spreadability: The project has a compelling story behind it (such as how the team overcame difficulties, how business experts sparked ideas with technical experts), making it easy to understand and spread.

Packaging and Compilation of Benchmark Cases

For selected benchmark cases, we invest professional resources in deep "packaging," transforming them from a technical project into an engaging "success story."

Standardized Case Content Structure

We write "Computing Platform Application Benchmark Cases" according to a unified template:

• Case overview (one-page summary): Summarize the core highlights of the case in the most concise language and data.
• Business background and challenges: Describe in detail the difficulties and pain points faced by the business before the project launch.
• Solution:
  • Business logic: Clearly explain how AI solves the business problem.
  • Technical architecture: Introduce the model, algorithm, and technology stack adopted.
  • Platform support: Highlight the key role played by the unified computing platform (such as how much computing power it provided, what technical problems it solved, how much it improved training efficiency).
• Application results: Use detailed data to quantify the business value and economic benefits brought by the project after going live.
• Experience summary and outlook: Extract the key successful experiences of the project and look forward to its future development and promotion plans.

Multi-Media Presentation

• Well-illustrated case collection: Compile multiple benchmark cases into a volume, creating an exquisite "State Grid AI Computing Application Benchmark Case Collection" as important material for external exchange and internal learning.
• Personnel interviews and feature films: Conduct in-depth interviews with core members of the benchmark project, filming and producing high-quality feature short films, using personal stories to enhance the appeal of the cases.
• Interactive online demos: For applications suitable for showcasing, assist the development team in creating an interactive online demo system, allowing users to experience the application's effect firsthand.

Promotion of Benchmark Cases

Packaged benchmark cases need to be widely promoted through multiple channels to maximize their value.

• Internal promotion:
  • Case sharing sessions: Regularly hold "Benchmark Case Sharing Sessions," inviting project leaders to share their experiences and success journey firsthand.
  • Internal media coverage: Conduct in-depth coverage of benchmark cases in media such as i State Grid and the company's internal publications.
  • Incorporation into training courses: Use benchmark cases as "living teaching materials," integrating them into our training courses.
• External publicity:
  • Industry conference speeches: Encourage and support benchmark project teams to give speeches and share at top technical conferences in the power industry or AI industry, enhancing the company's technical influence in the industry.
  • Authoritative media releases: Cooperate with national-level media such as Xinhua News Agency and People's Daily, as well as industry vertical media, to publicize benchmark cases of significant social value (such as AI empowering new energy consumption, ensuring large grid safety).
  • Award participation: Actively organize benchmark projects to apply for national, provincial, and ministerial-level awards for scientific and technological progress and management innovation, using authoritative honors to prove the platform's value.

Through the systematic operation of benchmark cases, we have not only successfully promoted the platform but, more importantly, established a cultural atmosphere of "using AI to solve problems" across the entire company, inspiring more teams to ride the wave of AI application innovation.

Knowledge Management: Building a Computing Operations Knowledge Base

During the construction, operations, and promotion of the computing platform, we have accumulated a vast amount of valuable knowledge assets. If this knowledge merely stays in the minds of a few experts, scattered in the documents of various projects, or buried in massive chat records, then it is "dead" and "one-time use." The core task of knowledge management is to systematically collect, organize, store, share, and reuse this knowledge, building a self-evolving "organizational brain," achieving knowledge distillation and reuse, and ultimately improving the entire organization's AI capabilities.

Positioning and Top-Level Design of the Knowledge Base

• Positioning: The "encyclopedia" and "navigation map" of the computing world
  • Single Source of Truth: Our goal is to make the knowledge base the primary inquiry entry point and the most authoritative source of answers for all questions about the computing platform within the company.
• Top-level design: Multi-level, multi-dimensional knowledge classification system
  • We design a clear, easy-to-search knowledge classification system:
    • By life cycle: Planning and design, deployment and implementation, operations management, application development, optimization and improvement.
    • By knowledge type:
      • Rules and regulations: Such as "Computing Resource Management Measures," "Service Level Agreement."
      • Operation manuals: Such as "Platform User Manual," "Cluster Operations Manual."
      • Technical solutions: Such as "Large Model Inference Optimization Plan," "Best Practices for Distributed Training."
      • Best practices/case library: That is, the aforementioned benchmark cases.
      • FAQ/fault knowledge base: Frequently asked questions and fault handling experience.
      • Training materials: PowerPoint slides, videos, and experiment code for all training courses.
    • By user role: Provide personalized knowledge navigation views for different roles such as business personnel, developers, operations personnel, and managers.

Knowledge "Input": Building a Co-Construction and Sharing Knowledge Production Mechanism

The vitality of a knowledge base lies in the continuous updating and enrichment of its content. We cannot rely on just a few people to write it; we must establish a mechanism that incentivizes all users and employees to participate in co-construction.

• Diverse knowledge sources:
  • Expert writing: Systematically written by the platform's core architects, operations experts, and technical support engineers to produce core technical documents and operation manuals. This is the "skeleton" of the knowledge base.
  • Operations process precipitation:
    • Ticket-to-knowledge: Require technical support personnel to take a few minutes after closing a representative service ticket to anonymize and standardize the solution, converting it into an FAQ or fault handling article.
    • Meeting minutes to knowledge: Minutes of major technical review meetings and fault postmortems, after being organized, can become valuable records of design decisions and lessons learned.
  • User contribution (UGC):
    • Award-writing contests: Regularly hold award-writing activities such as "My Computing Usage Tips," encouraging users to share their usage techniques, pitfall avoidance guides, and performance optimization experiences.
    • Open editing and comments: The knowledge base platform supports users in commenting on, correcting, and supplementing existing articles.
• Quality assurance review and release process:
  • Establish a knowledge base editorial committee: Composed of technical experts from various fields, responsible for reviewing the technical accuracy and writing standardization of newly submitted knowledge articles.
  • Standardized templates: Provide standardized writing templates for different types of knowledge (such as fault troubleshooting, operation guides), ensuring the knowledge is structured and readable.
  • Version control: All knowledge articles support version control, allowing their historical modification records to be traced.

Knowledge "Output": Creating an Intelligent, Convenient Knowledge Consumption Experience

Knowledge produced must be conveniently and quickly findable and usable to realize its value.

• Powerful search engine: The core of the knowledge base platform is a powerful search engine, supporting keyword search, synonym search, fuzzy search, and capable of personalizing search results based on users' historical behavior and roles.
• Intelligent recommendations:
  • Scenario-based recommendations: When a user submits a service ticket, the system can automatically recommend related solutions from the knowledge base based on the ticket content, achieving "ticket self-service" and reducing the pressure of manual support.
  • Proactive push: Based on users' subscription preferences and roles, regularly push the latest, possibly interesting knowledge articles via email and WeCom.
• Multi-channel access:
  • Integration into the service portal: Integrate the knowledge base as a core module of the computing service portal.
  • Integration into development tools: Through APIs or plugins, integrate the knowledge base's search functionality into the IDE or command-line tools commonly used by developers.
  • Intelligent Q&A robot: The core "brain" of the aforementioned intelligent Q&A robot is our knowledge base.

Knowledge Management Operations and Incentives

Knowledge management is not a one-time project but a long-term, living operational effort.

• Establish a dedicated Knowledge Manager (KM) position: Responsible for the daily operations, content planning, and community atmosphere building of the knowledge base.
• Establish an incentive mechanism:
  • Points and honor system: Users earn points for contributing knowledge, evaluating knowledge, and sharing knowledge. Points can be redeemed for gifts, or used as a reference for awards and promotions. We regularly recognize "Knowledge Contribution Star."
  • Incorporate knowledge contribution into performance assessment: For the platform operations and technical support teams, the number and quality of knowledge base article contributions are included as part of their performance assessment (KPI).
• Continuous data analysis and optimization:
  • We regularly analyze the backend data of the knowledge base. For example, which articles are searched most, read longest, and rated highest? Which keywords yielded no search results? These data can guide us to continuously optimize the content and structure of the knowledge base, making it increasingly "understand" the user.

Chapter Summary

In Chapter 9, we explored the "soft power" construction for the value realization of the computing platform. Through multi-channel application promotion strategies, we solved the problem of "good wine but deep alley," successfully spreading the platform's capabilities far and wide and igniting users' enthusiasm for applications. Through the systematic handling of benchmark cases, we established models of success, presenting the platform's value in the most intuitive and trustworthy way. Finally, through the construction of the computing operations knowledge base, we distill valuable practical experience into organizational wisdom that can be inherited and reused, building a solid knowledge foundation for the long-term and sustainable development of the company's AI capabilities.

These three tasks may seem less "hardcore" than technology R&D, but they are the key link connecting technology and business, present and future, individuals and the organization. Only by completing application promotion and knowledge precipitation have we truly finished the "last mile" of the computing platform's value loop, laying a solid user and cognitive foundation for the deeper exploration of cost operations and value measurement in the next chapter.