About Email Categorization Using AI

Email categorization allows automatically determining the category of incoming emails using artificial intelligence and forwarding them to operators for quick processing. This reduces the need for manual processing and improves response speed to client requests.

Context and Problem

In many cases, you need to automatically process incoming emails:

Large number of emails requires quick processing
Different types of requests require different operators or topics
Manual categorization takes a lot of time
Quick response to client requests is needed

AI categorization solves this task by allowing automatically determining email category and forwarding it to the appropriate operator.

Key Concepts

Email Processing Stages

Stage 1: Receiving Email

receive_email_message block automatically reads new emails
Email text is saved in email_message
Email subject is saved in email_subject
Sender address is saved in email_to

Stage 2: Categorization Using AI

fastline_completion block uses AI to analyze text
AI determines the best request category
Result is saved in completion (or specified variable)

Stage 3: Forwarding to Operator Panel

Email with category is forwarded to operator panel
Category corresponds to topic (subject) in operator panel
Topic must have an alias that matches AI value

Approach Options

AI Categorization vs Manual Categorization

AI Categorization:

✅ Pros: Speed, automation, scalability
❌ Cons: Possible errors, prompt configuration needed

Manual Categorization:

✅ Pros: Accuracy, control
❌ Cons: Slowness, does not scale

Why we use AI: AI allows automatically processing large numbers of emails and quickly forwarding them to operators, which improves response speed to client requests.

Adopted Solutions

Using fastline_completion

Action fastline_completion is used to analyze email text and determine category. The prompt should:

Describe categorization task
Specify available categories
Return only one category

Topic Alias Matching

The value returned by AI must exactly match the topic alias in the operator panel. This ensures correct assignment of email to the corresponding topic.

Implications for Users and Implementation

For Integrators

When configuring categorization, it's important to:

Configure receive_email_message — for receiving emails
Create AI prompt — with category descriptions and instructions
Configure topics in operator panel — with aliases that match AI categories
Check categorization accuracy — configure prompt for better results

Common Errors

Error: AI returns incorrect category
Problem: Insufficiently detailed prompt or incorrect instructions
Solution: Improve prompt, add examples and clear instructions

Error: Email is not assigned to topic
Problem: Topic alias does not match AI value
Solution: Check topic alias matching with values returned by AI

Error: Categorization works slowly
Problem: Slow model used or large prompt
Solution: Use faster model or optimize prompt

Usage Examples

fastline_completion Configuration

{
  "model": "gpt-4o-mini",
  "temperature": 0.3,
  "prompt": "You need to categorize client emails. Based on the provided message '{{email_message}}' and subject '{{email_subject}}', determine the best matching topic from the following: consultation, complaint, clarification. Return only one topic without additional explanations.\nImportant: choose only one category. If none of them fit, return 'Other' (other)."
}

operator_panel__connect_to_operator_with_msg Configuration

{
  "subject_alias": "{{completion}}"
}

Explanation: Connecting Chat with Operator — forwarding to operator panel