AI Agent Use Cases
Views —

AI Agent Use Cases

Added today

10All 221

1–20 of 221

Work & ResearchWalkthrough

Automation that classifies support emails and routes them to staff

n8n monitors a support inbox, assigns each email a category, summary, and priority, and sends it to the appropriate person's chat.

n8n

How it works

  1. The author shared an n8n workflow that monitors the support inbox of a small, family-run electronics shop.
  2. For each email, a single extraction call returns a category (billing, technical, sales, or other), a short summary, confidence, and a priority based on tone and urgency. The workflow sends these details to the responsible person's chat.
  3. Low-confidence emails go to a separate review channel instead of a specific person, with a note showing the category the model guessed.

Results

The author did not provide specific figures. They said the review channel reveals categories the model often confuses, helping them revise the prompt.

Notes

The author said the workflow JSON and four extraction prompts are available in their GitHub repository. The post does not report results from deploying it in the actual shop.

Source

Reddit Original post ↗

Example prompt · n8n

Please build an n8n workflow that classifies and summarizes incoming customer support emails and sends them to the appropriate person's chat channel.

Inputs: inbox to monitor [address], category list [for example: billing, technical, returns, other], responsible person and chat channel for each category [Slack, Telegram, etc.], and confidence rule [for example: send low-confidence items to a review channel]. First, check whether my n8n instance can connect to this inbox and these chat tools, and whether the email bodies may be sent to an AI model.

Implementation steps: 1) When a new email arrives, read its sender, subject, and body. 2) Use one AI call to return a category, a two- or three-line summary, confidence (high, medium, or low), and priority (based on tone and urgency) in a fixed format. Keep the category list inside the prompt so it can be changed later by editing the text. 3) For high or medium confidence, send the sender, subject, summary, and priority indicator to the responsible person's chat. 4) For low confidence, send the email to the review channel and add a note showing the category the model guessed. 5) Do not include any action that replies to or deletes emails.

Deliverables: ready-to-import workflow JSON, the full classification prompt, and example messages for the responsible person's chat. First, test with 10 past emails and show the classification results in a table. Review the emails sent to the review channel and summarize which parts of the prompt should be revised.