Run voice and chat support assistants with memory
AT&T uses BigQuery and Gemini Enterprise for Customer Experience to deploy highly contextual AI agents, resolving 20% more calls autonomously. Task: AT&T deployed BigQuery for intent routing and Gemini Enterprise for Customer Experience to power voice and chat agents with 60-day memory. This replaced rigid rule-based systems, allowing agents to recall customer history across channels and resolve inquiries autonomously. Human agents were freed to handle complex issues, though specific human-in-the-loop workflows for the AI are not detailed.; tools: Gemini. Human role: The original post does not say how a person stayed involved.; limitations: The source does not specify how human oversight is integrated into the autonomous resolution workflow or the exact cost per interaction. Success relies on deep integration with BigQuery for historical data access and Gemini for conversational context. Security and monitoring protocols are mentioned but not detailed, requiring robust internal governance for production deployment..
Tools
Published on Agent Work Atlas:
Overview
AT&T uses BigQuery and Gemini Enterprise for Customer Experience to deploy highly contextual AI agents, resolving 20% more calls autonomously.
Workflow
AT&T deployed BigQuery for intent routing and Gemini Enterprise for Customer Experience to power voice and chat agents with 60-day memory. This replaced rigid rule-based systems, allowing agents to recall customer history across channels and resolve inquiries autonomously. Human agents were freed to handle complex issues, though specific human-in-the-loop workflows for the AI are not detailed.
Reported result
The implementation resolved 20% more customer inquiries autonomously and achieved approximately $150 million in annual savings. Net Promoter Scores improved due to reduced repetition and faster resolution. The system processes 40 billion tokens daily across 150 generative AI solutions, though these metrics reflect the broader AskAT&T platform rather than just the customer-facing agents.
Human role
The original post does not say how a person stayed involved.
Limitations
The source does not specify how human oversight is integrated into the autonomous resolution workflow or the exact cost per interaction. Success relies on deep integration with BigQuery for historical data access and Gemini for conversational context. Security and monitoring protocols are mentioned but not detailed, requiring robust internal governance for production deployment.
Try it
I am building a customer support agent for [replaceable business context, e.g., an e-commerce store] using [replaceable tools, e.g., BigQuery and Gemini]. My goal is to reduce repetitive customer explanations by implementing a 60-day memory feature. First, analyze my sample customer interaction logs in [replaceable file path/URL] to identify common intent patterns. Then, draft a system prompt for the AI agent that uses this history to personalize responses. Before deploying, run a dry test on 5 sample tickets and show me the proposed responses. Do not send any messages or make changes until I approve the output. If data is missing, ask me for specific customer history fields. Ensure the agent escalates to a human if confidence is low.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.