How Bernalillo County Saved $108,000 and Reduced Support Costs by 80% Using AI

BernCo saved $108,143.75 in net avoided agent costs over 18 months by deploying CustomGPT.ai to handle resident inquiries across web, phone, and email channels.

BernCo achieved a 4.81x return on investment, meaning every dollar invested in CustomGPT.ai generated approximately $4.81 in savings through reduced agent handling costs.

AI reduces government support costs by handling routine resident inquiries automatically, at a cost per interaction significantly lower than human agents. BernCo’s AI-handled interactions cost $0.99 each, compared to $4.59 for agent-handled contacts, an 80% reduction in cost per interaction.

RAG stands for Retrieval-Augmented Generation. Rather than generating answers from general language model training data, RAG-powered AI retrieves relevant content from a specific, curated knowledge base and grounds its responses in that material. This means the AI answers based on what the organization’s own documentation says, not what the model was trained to approximate.

Government agencies need AI that is accurate and accountable. RAG architecture ensures that AI responses are grounded in official, verified documentation rather than general training data. This prevents hallucinated or inaccurate answers, which in a government context can create legal and financial risk for residents who act on incorrect information. Learn how CustomGPT.ai prevents hallucinations.

BernCo’s AI handled 28,433 resident interactions over 18 months, representing nearly 25% of the 114,836 total resident contacts during that period.

No, and BernCo’s deployment was not designed to replace employees. AI handled routine, predictable inquiries that were consuming expert staff capacity. Human agents remained responsible for complex cases, appeals requiring judgment, and situations that fell outside documented processes. The result was a team able to focus its expertise on the work that actually required it.

The most effective approach is deploying RAG-powered AI trained on the agency’s own official documentation. This gives residents an intelligent, conversational interface to the information the agency has already produced, available 24/7, across the channels residents already use. BernCo’s experience demonstrates that well-implemented self-service can handle nearly a quarter of all resident contacts without human involvement.

Any government department that handles high volumes of resident inquiries benefits from AI. Property assessment, tax administration, permits and licensing, compliance, parks and recreation, and public health are among the most common high-volume contact centers. BernCo’s Assessor’s Office is representative of the type of operation where AI produces the clearest ROI. See CustomGPT.ai for government.

CustomGPT.ai allows government agencies to build AI assistants trained on their own official documentation, policy guides, process documentation, public records, and regulatory content. The no-code platform enables non-technical staff to build, deploy, and maintain AI agents without engineering resources. Multi-agent architecture supports specialized assistants for different audiences and use cases. RAG-powered accuracy ensures responses are grounded in verified documentation.

Start with high-volume questions. Ground AI in official documentation. Deploy specialized agents for distinct audiences. Expand across channels. Measure cost per interaction before and after. Use analytics to improve continuously. BernCo’s deployment demonstrates that each of these principles produces measurable results, and that together they create a support operation that delivers better service at significantly lower cost.

Identify the question types that generate the highest contact volume. Verify that accurate, official documentation exists to answer them. Deploy a focused AI assistant on the pages or channels where those questions occur most. Measure deflection rate and cost per interaction. Demonstrate ROI to leadership. Then expand.

BernCo’s phased approach, starting with the ACE Community Educator and expanding to specialized agents over subsequent weeks, is a repeatable model for any government agency navigating a first AI deployment. Start a free trial of CustomGPT.ai to see how quickly your team can be up and running.