On April 9 this year, the General Office of the Guangdong Provincial Government issued a highly significant document: the "Guangdong Province Action Plan for Accelerating High-Level Application of Artificial Intelligence Across All Domains, at All Times, and in All Industries" (Yue Ban Han [2026] No. 50).
The phrase "all domains, all times, all industries" is not rhetoric. The document divides "AI+" into seven major directions. In the section "AI + traditional industries" alone, it writes scenarios one by one for twelve industries: electronics, chemicals, machinery, toys, home appliances, food, building materials, textiles, metallurgy, mining, shipbuilding, and energy. Toy makers, textile makers, and food factory owners can all find passages relevant to themselves in it.
After reading the document, entrepreneurs truly care about only two questions: How do I get the government's money? Where do I start with my own AI? These two questions are actually two sides of the same question.
First, look at the money pie
The scale of the supporting funding is unusually large compared with recent years:
In 2026, the provincial finance allocated 3.634 billion yuan to implement the post-event reward policy for technological transformation equipment, supporting equipment renewal and intelligent transformation; during the policy period, an additional 1 billion yuan was added to specifically support the digital transformation of small and medium-sized enterprises; in 2026, a fund for the development of the artificial intelligence and robotics industry was also newly established, with benchmark cases of large models and application solutions in the industrial field, supporting no more than 10 selected cases each year, with a maximum reward of 8 million yuan for each.
There are also supporting measures worth noting: the Provincial Department of Industry and Information Technology has built a manufacturing empowerment resource pool, categorizing and aggregating resources such as large models, industrial agents, datasets, computing power services, solutions, and consulting and diagnostic services; it has also launched the "Yue Zhi Fu" platform, providing inclusive "AI diagnosis" services, and is holding offline supply-demand matchmaking meetings for key industries such as electronic information, automobiles, home appliances, chemicals, and textiles.
In other words, from diagnosis and selection to funding, the government has laid out the service chain. The remaining key is: enterprises themselves must first prepare the qualifications to be "diagnosed".
The prerequisites for AI implementation are actually written in the document
There is a sentence in the action plan that is easy to skip over: relying on the 'Open Guangdong' platform, prioritize the opening of high-value data, encourage key fields such as industry to build high-quality industry datasets, and deepen the authorized operation of public data.
Policymakers understand it more clearly than many companies do—in the AI era, data is fuel. The document discusses datasets, data foundations, and data governance throughout.
This is the core judgment I want to make: For the vast majority of small and medium-sized manufacturing enterprises, the first step in implementing AI is not buying a large model, but getting their business data managed first.
The principle is simple. AI quality inspection requires production data to be online; intelligent scheduling requires order and capacity data to be online; predictive maintenance requires equipment data to be online. If data is not in the system, not standardized, and not connected, even the smartest model can only run idle.
Our daily observation from doing Odoo implementations is: how deeply an enterprise uses ERP basically determines how far it is from AI. When sales orders, purchases, inventory, production work orders, and financial vouchers all flow through the system, connecting an AI application (such as intelligent quoting or demand forecasting based on historical data) is a natural next step; conversely, if core data is still scattered across various departments' Excel files, the first step of any AI project must be to invest significant effort in catching up on data governance—this debt will have to be paid sooner or later.
Three-step recommendations for enterprises
Step one: first take clear stock of your assets. While the inclusive AI diagnosis offered by "Yuezhifu" is free or low-cost, do a diagnosis first to understand your current data situation and the scenarios most worth investing in. Do not launch projects based on gut feeling.
Step two: move the core business into the system. Prioritize bringing the main chain—sales, procurement, inventory, production, and finance—online and structured. Every penny spent in this step is both laying the foundation for AI and, in all likelihood, can also get on board the subsidies for technological transformation and digital transformation of small and medium-sized enterprises—as mentioned earlier, the 3.634 billion technological transformation pool and the 1 billion special fund for SME digitalization are real money.
Step three: choose a small entry point to try AI. Don't start with "comprehensive intelligentization" right away. Pick a scenario with a good data foundation and quick results—such as a customer service knowledge base, automatic document recognition, or sales forecasting—get one working, then replicate it. The benchmark cases in the document with an 8 million reward all grew out of specific scenarios, not out of grand plans.
A final honest word
"AI+" is the trend, and there is no arguing with that. But when it comes down to implementation, the gap between enterprises is not about who bought a large model first, but about whose business data is ready first.
AI is an express train, but the ERP road must be built first. No matter how fast the train, if the road isn't open, you can only watch the excitement from the platform.
If your company is at the stage of 'wanting to adopt AI but not knowing where to start,' feel free to reach out — helping you figure out the first step is much more important than rushing to sell you a system.
