Customer support teams often face high volumes of repetitive email requests, including information inquiries and invoice-related complaints. These processes are typically manual, time-consuming, and prone to human error, leading to inefficiencies and inconsistent service quality.
This project was developed to address these challenges by leveraging automation and AI-powered classification to streamline the handling of customer emails. The solution integrates tools such as AI Builder, Power Automate, Excel, and OneDrive to create a structured, scalable workflow.
As shown in the workflow diagram, the system automates the full lifecycle of a request—from email reception to classification, validation, task creation, and resolution, ensuring improved operational efficiency and service consistency.
This project presents an end-to-end automated workflow designed to manage customer support emails, specifically focusing on information requests and invoice complaints.
The system is triggered whenever a new email is received and uses Natural Language Processing (NLP) to classify the request type, detect sentiment, identify urgency, extract key data (e.g., TIN), and generate a structured summary.
Key functionalities include:
- Automated email classification (Request Type, Sentiment, Language, Urgency);
- Intelligent request summarization for operator support;
- Automatic storage and processing of invoice attachments;
- Validation rules for complaints (e.g., invoice within 12 months);
- Integration with Excel for structured data logging;
- Task creation for operational tracking;
- Urgency-based alerts via Microsoft Teams;
- Approval workflow with customer feedback loop (including satisfaction survey).
The solution significantly reduces operational workload, minimizes human error, and accelerates response times, ultimately improving customer satisfaction
Prompt used: You are a natural language processing assistant. Your task is to analyze the text of a received email and identify the following information: Request Type Classification: Determine whether the email is an "Information Request" or a "Complaint".
Use the following criteria:
- If the text contains questions, requests for details, or explanations, classify it as an "Information Request".
- If the text contains dissatisfaction, criticism, reported issues, or complaints, classify it as a "Complaint".
Sentiment Analysis: Classify the overall sentiment of the email as "Positive", "Negative", or "Neutral".
- Positive: The text shows satisfaction, gratitude, or praise.
- Negative: The text shows dissatisfaction, frustration, or complaints.
- Neutral: The text has no clear emotions or uses objective language.
Language Identification: Identify the language in which the email was written. Return the language in ISO 639-1 format (e.g., "pt" for Portuguese, "en" for English, "es" for Spanish).
Tax Identification Number (TIN) Identification: Search for the Tax Identification Number in the email text.
- If found, return the identified number.
- If not found, set it as "N/A".
Urgency: Yes or No: Determine whether the email indicates urgency. Use the following criteria: If the text contains words such as "urgent", "immediate", "priority", or mentions tight deadlines, set it as "Yes". Otherwise, set it as "No".
Input:
Expected Output: the following fields in the format:
- RequestType: "Information Request" or "Complaint".
- Sentiment: "Positive", "Negative", or "Neutral".
- Language: ISO 639-1 code of the identified language and the language name in European Portuguese.
- TIN: Identified number or "N/A" if not found.
- Urgent: "Yes" or "No".
Prompt used: “Structure the task body based on the following email in European Portuguese.
Identify the customer, include a clear and objective summary of the reported situation, indicating whether it is an information request or a complaint.
Specify any relevant data, such as the customer’s Tax Identification Number (TIN), and detail any additional information or inconsistencies present, such as missing or incorrect attachments.
- Highlight elements that facilitate the operator’s work, such as urgency, message details, or documentation. Ensure a structured and clear presentation to facilitate understanding.”
Based on the implementation and identified limitations, the following improvements are recommended:
- Fully integrate AI Builder outputs into the workflow to enable automated validation of invoice data (e.g., Issue Date filtering);
- Enhance data consistency by including all extracted invoice fields (especially Invoice Date) in the Excel database;
- Improve system robustness by adding error-handling mechanisms for missing or invalid attachments.
You can consult the full project presentation here: