Four AI projects delivered in one month for 120k: digging client business rules beats the model
Lucas｜AI自动化改造小红书 · 2026-09-04
Read original作者复盘 8 月交付 4 个 AI/自动化项目，总收入 12 万。项目包括微信销售 Agent、订单回款系统等。强调真正决定价格的是嵌进客户业务有多深，而不是技术差距。
Business model
In August the author delivered four AI/automation projects for a total of 120k. Examples include a WeChat sales reply agent, structured quoting from chat, and an order-collection system (1415 customers / 3187 orders). Clients pay for rules that are extracted and made runnable, not for which model is used.
Ordinary people can turn “help small businesses automate” into a service. Quoting, collections and sales replies are frequent pain points. Turning a client’s hidden rules into a working agent commands far higher fees than a plain chatbot.
Workflow
- Sit with the client and extract business rules (price, lead time, implicit agreements)
- Let the agent read chats/tables and structure them into checkable fields
- Human confirm key outputs then generate quote or collection views
- Iterate rules after real use
- Charge per project or monthly
Adaptation plan
Content and results come from the original poster; AgentFlow only breaks down the public summary and has not independently verified truthfulness or income. Added steps are untested suggestions and do not guarantee results. Page access limited (summary only).
Starter prompt
Written by AgentFlow; not the original author’s prompt. Not independently deployed in Muse or Grok.
The client is in [industry]. Their daily chats/orders contain product, specs, quantity, price, lead time, plus implicit rules like “use last price” or “switch to this size”. Please: 1) extract core fields from sample dialogues; 2) write a checkable rule list; 3) design a flow that generates PDF/Excel after human confirmation. Give a plan first—do not touch their live system yet.
Running costs and pitfalls
Content and results come from the original poster; AgentFlow only breaks down the public summary and has not independently verified truthfulness or income. Added steps are untested suggestions and do not guarantee results. Page access limited (summary only). No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.