
Taxpayer Service Chatbot
A state revenue agency deployed an AI chatbot to resolve taxpayer inquiries 24/7, cutting call volume and improving satisfaction.
A state revenue agency handled over 500,000 calls per year, with long wait times during tax season and high rates of abandoned calls. Most inquiries were repetitive questions about filing status, payment deadlines, refund tracking, and form downloads. The agency partnered with Vitosha to deploy an AI-powered chatbot on the state’s website and mobile app, providing instant, accurate answers and deflecting volume from the call center.
Overview
A state revenue agency handled over 500,000 calls per year, with long wait times during tax season and high rates of abandoned calls. Most inquiries were repetitive questions about filing status, payment deadlines, refund tracking, and form downloads. The agency partnered with Vitosha to deploy an AI-powered chatbot on the state’s website and mobile app, providing instant, accurate answers and deflecting volume from the call center.
Business Challenge
During peak tax season, average hold times exceeded 45 minutes and the abandoned call rate reached 35%. The agency’s website had comprehensive FAQ content, but taxpayers could not easily find answers to specific questions about their account status, refund timing, or payment history. Call center agents spent 60% of their time answering questions that could be self-served, while more complex inquiries faced delays. The agency needed a 24/7 solution that could handle natural language questions and integrate with backend systems for personalized responses, all within strict security and privacy requirements.
Solution
Phase 1: Knowledge Base Build — Curated and structured the agency’s extensive tax knowledge base, including FAQs, form instructions, filing guidelines, and policy documents, organizing content for semantic search and LLM retrieval. Phase 2: Chatbot Development — Built a Power Virtual Agents (Copilot Studio) chatbot using Azure OpenAI for natural language understanding, connected to Azure Cognitive Search for retrieval-augmented generation across the knowledge base, with Dataverse managing conversation context and escalation logic. Phase 3: Account Integration — Connected the chatbot securely to backend systems for authenticated taxpayers to check refund status, payment history, and filing deadlines, with Entra ID B2C handling authentication and session management. The chatbot escalated complex or sensitive cases to live agents through D365 Customer Service with full conversation history.
Results
The chatbot resolved over 40% of all inquiries end-to-end within the first year, handling more than 200,000 conversations annually. Average handle time dropped from 12 minutes (call) to under 2 minutes (chatbot). Call volume decreased by 40%, allowing the agency to redirect 15 agents from Tier 1 support to complex case resolution and compliance. Taxpayer satisfaction with digital services rose to 92%, and the chatbot achieved 24/7 availability with 99.9% uptime. During tax season, the chatbot handled 3x normal volume with no degradation in response quality, eliminating the seasonal spike in abandoned calls.
Conclusion
The AI-powered chatbot modernized taxpayer services, delivering instant, accurate responses around the clock. By deflecting routine inquiries, the agency reduced costs, improved satisfaction, and refocused skilled staff on complex cases that required human judgment.
Impact Metrics
40%
Call Center Reduction
24/7
Availability
92%
Satisfaction
Tech Stack
Other Case Studies

AI Model Risk Governance
A large bank implemented enterprise AI model risk governance, inventorying 500+ models and reducing validation cycle time.

AI Student Engagement System
A school district used AI and unified data to identify at-risk students early and boost engagement through automated, personalized interventions.

AML Monitoring Modernization
A global bank modernized its AML monitoring with ML-based alert triage, reducing false positives by 70% and SAR prep time.