DingTalk's AI assistant pulled one company's ticket response from hours down to seconds — and the price wasn't a self-built large model, but the AI capabilities bundled with off-the-shelf OA (office automation) tools plus low-code (a visual, drag-and-drop way to build systems). This near-zero-cost AI deployment path makes us reconsider how quickly Chinese enterprises can actually roll out AI.
What This Is
The author used DingTalk's AI assistant (an AI module built into DingTalk, backed by Alibaba Cloud's large models underneath) to build a ticket support system. It breaks down into three parts.
First, the knowledge base. Historical tickets, FAQs, and product manuals get fed in; when answering, the AI retrieves relevant material first, then hands it to the LLM to generate the reply. This is what the industry calls RAG (Retrieval-Augmented Generation — in short, look things up before answering rather than answering from memory).
Second, ticket templates. Fields are forced into structured formats, with required items capped at 3–5, so the AI can accurately understand what the user is asking — improving data quality at the source.
Third, the feedback loop. Every AI reply gets a human rating, which feeds back to optimize the knowledge base — forming a "use → evaluate → improve" cycle.
The whole thing runs inside the DingTalk ecosystem, using an "AI filters first, humans decide later" collaboration model. Companies don't need to spend extra.
The Industry View
Supporters call it pragmatic. SMBs don't need to train their own models — borrowing the AI capabilities bundled with OA platforms can get their first Agent (an AI assistant that autonomously completes multi-step tasks) scenario running. Ship the efficiency numbers first, argue strategy later.
The objections deserve more attention. First, data flow — tickets routinely contain customer information and internal process details. Feeding them into a third-party AI module (even DingTalk's own) means enterprise IT needs to reassess compliance risk. Second, vendor lock-in: if DingTalk works today, switching to Feishu or WeCom tomorrow won't be cheap. Third, ticketing is a highly structured scenario — one of the easiest for AI to handle well. Success here can't be extrapolated to "AI transforming the whole company."
Impact on Regular People
For enterprise IT: Low-code + AI assistants are rewriting the cost math of building systems in-house. The "build everything ourselves vs. borrow the platform's bundled capabilities" tradeoff will come up more and more often, and the logic for allocating tech budgets has to change.
For individual careers: Repetitive Q&A in customer service and technical support roles will get absorbed by AI first — but these jobs won't disappear. They'll just shift from "all-human answering" to "reviewing AI answers + handling exceptions."
For consumers: Next time you call customer service, you'll likely get an AI reply in seconds, with complex issues handed off to humans. Faster response, yes — but also more templated answers, and not necessarily a better experience.