2303-UC Agentic Customer Success (ACeS)
Cybersecurity providers monitor each client across several platforms, and their Customer Success teams spend a large share of their week manually pulling data from those tools and turning it into client reports. Agentic Customer Success (ACeS) is a Phase 1 proof of concept, built with an industry sponsor, that automates the research behind those reports while keeping the results traceable and safe. ACeS combines data from Google Security Operations and Automox. A Planner turns each request into a fixed, validated plan. Read-only Collectors, written as plain code with no AI, load each client's data into Google BigQuery, where fixed SQL rules match devices across both systems and identify findings that neither tool shows on its own. Only then do specialized AI agents explain each finding and suggest a next step. The agents have no tools and no credentials, so they cannot act on a client system, and the raw data never passes through the AI. Every AI statement must pass an evidence gate that checks its numbers, devices, causes, and severity labels against the computed findings. Unsupported claims, such as calling a drop in log volume an attack, are removed before the draft is built. A person must approve every draft before a client-ready report is produced. ACeS has run end to end on live sponsor data. Automated security tests show that injected instructions cannot trigger actions, change a run's plan, or slip unsupported claims into a report. The project shows how constrained agentic workflows can make customer-facing security reporting faster, more consistent, and auditable.
Authors
- Amelia Black
- Helena Patino Cruz
- Amber O'Dwyer
- Elijah Guisto
- Steven Thai
Publication Details
- Journal
- DigitalCommons - Kennesaw State University (Kennesaw State University)
- Published
- 2026-11-20
- Primary Topic
- Network Security and Intrusion Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00