A company with an annual revenue between several hundred million and over one billion yuan, equipped with ERP, financial, OA, and CRM systems plus a basic IT team, already recognizes the necessity of AI. However, securing approval for a multi-million yuan budget in the first year is a significant hurdle. A more realistic figure a boss might offer is a pragmatic one: "Here's 1 million yuan to experiment for a year, show me what you can create."

One million yuan is not a "large model construction fee"; it's better described as seed funding for the enterprise's AI transformation. This amount is both substantial and limited. For a company with annual profits of 20 million yuan, it represents 5% of its earnings, a significant addition to a typical IT budget of 3-4 million yuan. Yet, it's insufficient to absorb the cost of a major procurement error. Spending 300,000 to 400,000 yuan on a platform, plus another 200,000 to 300,000 on knowledge bases, training, and consulting, would quickly exhaust the budget without clarifying where AI can truly create value. Therefore, the objective for this first year shouldn't be to "build an enterprise large model" or meet a quota for deploying agents. The primary goal is to foster a subtle but critical cultural shift: where initially only a few young employees use AI occasionally, a year later, dozens or even hundreds of staff naturally pause to consider if AI could handle a task before starting it manually. This budget is insufficient for a complete AI technology stack, but it's ample to establish a company's first mechanism for continuously discovering, learning, filtering, promoting, and absorbing AI. The core aim is to transition the company from merely "knowing AI is important" to "understanding how AI should be used." This discussion assumes the company already has a digital foundation—with knowledge workers, industry software, and an IT team—but lacks a formal AI unit. For those without basic information systems, a large portion of the budget must first address these foundational gaps. For companies with existing infrastructure, the most critical characteristic of this 1 million yuan is that it must preserve optionality. The AI landscape evolves rapidly—models, prices, office software, agent tools, and AI coding capabilities are all in flux. A cutting-edge product today might be a standard feature tomorrow, and capabilities that seem bespoke now could be integrated into mainstream platforms like DingTalk, Feishu, WeCom, or WPS within months. A small budget is ill-suited for being locked into a closed platform or a multi-year contract. The first year is for seeding: experiment broadly, learn quickly what works and what doesn't, and redirect funds toward initiatives that show real value. A key principle in budget allocation is that the largest expenditure should be on people, not models or platforms. Software can be purchased monthly, and cloud resources can be scaled, but without a dedicated individual whose primary focus is enhancing the company's AI proficiency, resources will likely be squandered on fragmented purchases.

The first priority is to hire an "AI Coach." Most companies have AI-savvy individuals scattered across departments, but their efforts are uncoordinated. The problem is a lack of ownership for AI outcomes. An AI committee that meets monthly to discuss trends is ineffective. What's needed is a dedicated, hands-on "AI Application Manager" or "AI Innovation Manager"—an internal AI coach. This role differs from an AI leader in a large conglomerate; it requires someone who is hands-on and versatile. This person must track external tools, discern what's worth adopting, embed themselves in sales, procurement, finance, and R&D to observe real work, deliver practical training, build simple prototypes using AI coding or low-code platforms, and know when to escalate complex issues to the IT team or external vendors. The coach's success is measured not by doing AI work for the company, but by making the company less dependent on them. Their operating principles should be: enable employees to do tasks themselves, purchase mature tools rather than building them, and only commission custom development for tasks that offer a true competitive advantage. This is reminiscent of the early days of personal computers in business; companies didn't need to develop Windows, but they needed people to evangelize the benefits of word processors over typewriters. AI is going through a similar adoption phase, but at a much faster pace. Many employees don't lack access to a chatbot; they lack guidance on how to integrate AI into their existing workflows.

Pairing the AI Coach with 2-3 "AI-native" interns is a strategic investment. Allocating roughly 200,000 yuan for interns might seem extravagant, but it's a highly effective configuration for a modest budget. These young professionals, having grown up with AI as a default tool, can partner with experienced employees to uncover valuable applications. They bring fresh perspectives on what can be automated, while veterans provide the crucial real-world context of why processes exist. Instead of a separate "AI task force," the best approach is to pair interns with business units—one with sales, one with procurement, one with finance—to observe and identify opportunities for AI integration. They can also alleviate the AI coach's workload by continuously testing new products, creating short tutorials, and building rapid prototypes to validate ideas. Their output should be reusable assets like templates, workflows, and evaluations, transforming one-off tasks into organizational knowledge. If the company expands its AI efforts, these interns could be offered full-time positions, creating a team that is both AI-fluent and understands the business, rather than hiring externally.

The strategy is to prioritize mature AI tools and integrate them into existing workflows before any custom development. The array of off-the-shelf AI solutions for office work, documentation, meetings, research, and content creation is vast and improving. For a small budget, leveraging these commoditized tools is far more economical than building a proprietary platform. There's no need to create a new "AI portal" that employees would have to remember to visit. The focus should be on enhancing the tools they already use daily, like WeCom, DingTalk, or Feishu. The AI tool strategy should be tiered: a foundation for all employees using built-in features of current office suites; a second tier for 30-50 heavy users in roles like sales operations or financial analysis, who receive access to more powerful agents or workbenches; and a third tier for specialized roles like developers with AI coding tools. Equally important is actively managing software licenses—reclaiming unused seats and upgrading for active users, rather than letting subscriptions sit idle. The AI coach and interns should maintain a lightweight "AI tool radar" to track what's been tested, its cost, and its effectiveness, creating a valuable internal repository of AI knowledge.

AI training should be a sustained effort to shift work habits, not a one-off event. A standard two-day training session on AI concepts rarely translates into changed behavior. The budget should be allocated to continuous learning: bi-weekly 30-minute internal sessions on a single practical task, weekly "AI clinics" where employees bring their own work problems, an internal IM group for sharing tips and failures, and monthly showcases from departments on how they've improved a process. Training content should be tailored: management needs to understand capabilities and ROI, middle managers need to see how AI changes workflow and oversight, and general staff need concrete examples for their daily tasks. The ultimate goal is to create a new reflex—where employees automatically consider if AI can help with a task. A significant challenge is employee hesitancy to share productivity gains for fear of increased targets or job loss. The company must signal that sharing efficiency improvements is rewarded, not punished, otherwise AI adoption will remain hidden and fail to become an organizational capability.

After tool adoption, the focus should shift to perfecting just 1-2 business use cases. Around the six-month mark, the company should identify promising tasks for deeper integration, but this shouldn't be decided through a formal "AI scenario request" form. The AI coach should act as an internal consultant, observing work and questioning why processes are slow or information is hard to find, rather than just collecting requests. The goal is to find a few high-value tasks that can be transitioned from an individual's experiment to a team tool and eventually to an enterprise application with proper integrations. This "promotion mechanism" avoids both the rigidity of a traditional IT project and the chaos of a hastily launched pilot. System modifications should be minimal, solving only the immediate integration needs without launching a massive AI platform or data governance initiative. The advantage of a small budget is its agility; it's better to make many small, inexpensive mistakes than one large, costly error, as cheap failures enable rapid learning.

The most common obstacles to a successful 1 million yuan AI initiative are often organizational, not technical. These pitfalls include an "AI leader" who is still overloaded with their prior duties; business units that don't allocate time for experimentation; friction between the new AI role and the existing IT department; business teams treating AI as just another IT project; the AI coach becoming a general-purpose "AI helpdesk" for every minor request; interns being relegated to basic content creation; purchasing expensive licenses for everyone that go unused; employees using unauthorized "Shadow AI" tools with company data; redundant AI purchases when existing software already has the features; a constant chase for new tools without deepening any one application; employees hiding their productivity gains; unrealistic expectations of immediate financial returns; vendors consuming most of the budget on a "comprehensive" solution that leaves the company dependent; and a company's AI capability vanishing if a key person leaves. A deeper issue is "AI-ing" an old process without redesigning it—adding an AI step to a report but keeping all the old review layers. The fundamental question should be: if AI existed from day one, would this work be designed this way?

At the end of the year, the measure of success should be the capabilities and assets the company has built. The focus should be on having a person who truly understands the business and tracks AI developments; interns who grasp how the business functions; 5-10 employees who proactively solve problems with AI; and dozens more who have integrated AI into their daily routine. The company should possess a repository of internal tutorials, tested workflows, tool evaluations, and documented case studies, including failures. This is the foundation for the next budget cycle, enabling the management to make informed decisions about where to invest next—whether it's expanding a proven tool, hiring an AI application engineer, or deepening a specific integration. The return on this first-year investment is not just immediate financial gains, but a significant reduction in the probability of making a large, costly mistake in the future. The true value of 1 million yuan is not in purchasing a system, but in cultivating the company's first capacity for continuous AI learning. This capability is what will guide the effective allocation of larger budgets in the future.

The job description for the AI Application Manager (or AI Innovation Manager) should focus on business impact, not just technical expertise. This role reports to the CEO or CIO and is responsible for promoting AI tool adoption, discovering business use cases, and building the company's AI literacy. Key duties include tracking external AI tools, running internal evaluations, providing ongoing training, embedding with business teams to identify optimization opportunities, building lightweight prototypes, mentoring interns, cultivating AI champions, and managing software licenses. The ideal candidate is an avid AI user with 3-8 years of experience in a business or IT role. They must be hands-on, able to build simple prototypes, but, most importantly, they need to understand business processes, communicate effectively, and train others. Candidates who are pure researchers, strategists who don't engage with tools, or salespeople without practical experience are a poor fit for this stage. The interview should involve a practical, real-world task, such as analyzing customer data or building a solution, to assess their problem-solving approach and their ability to explain their methods to a non-technical person. The first-year goals should be to validate a set of AI tools, establish usage habits among core staff, develop a group of AI champions, create internal resources, and successfully implement 1-2 business use cases, providing clear direction for the next year's AI investment.