Back to Case Studies
ENTERPRISE AGENTIC AI AMADEUS 2025 to 2026

SOC Copilot: Agentic AI for Security Operations

Agentic copilot that auto-investigates incoming security alerts, assesses risk, and tells analysts what to do next, grounded in approved SOC playbooks.

89%
Active analyst adoption across global operations
10-15 min
Research time on hard alerts, cut from about a day
~₹50L / yr
Estimated operational analyst time saved annually
SOC Copilot End-to-End Agent Architecture
TheHive Alert ~300 alerts/wk Power Automate Event Trigger Copilot Studio Agent TheHive data reasoning + Playbook RAG Risk Assessment Context + Citations Analyst Action Fast Triage Weekly Senior Analyst Sampling & Quality Control Loop

Problem

The enterprise Security Operations Center (SOC) handles approximately 300 alerts each week through this workflow. Routine investigations such as risky logins required 30 to 60 minutes each, while complex, hard investigations (roughly 10% of total volume) routinely took up to a full working day per alert.

Analysts faced alert fatigue, repetitive manual pivoting between disparate consoles, and varying levels of playbook adherence across rotating shifts. As alert volume scaled, security leadership needed to absorb volume without adding headcount.

Discovery & User Research

In October 2025, I initiated analyst shadowing sessions across multiple security tiers to understand where triage time was actually lost. The findings revealed that analysts did not struggle with taking final mitigation actions; they lost hours gathering context across systems, cross-referencing internal SOPs, and synthesizing initial findings.

I structured a proof of concept and presented the discovery insights at the Scaled Agile PI planning session (PI 25.4), aligning engineering, architecture, and security leadership on an agentic solution.

Key Architectural Decision

Should we adopt the general enterprise Microsoft Copilot with prompt grounding, or build a purpose-built SOC Copilot?

We chose to build a purpose-built SOC Copilot in Microsoft Copilot Studio. General enterprise copilots pulled broad ungrounded context and lacked operational integration with ticketing infrastructure. A purpose-built product allowed us to:

Shipping Strategy: Two-Stage Evolution

Release 1 (January 2026): Grounded Copy-Paste Assistant

Instead of delaying launch while awaiting an enterprise cloud tenant migration, we shipped a v1 RAG assistant directly in Microsoft Teams. Analysts pasted alert payloads and received instant playbook-grounded guidance. The usage telemetry and positive feedback proved clear operational value, securing executive buy-in and funding for the complete tenant migration and integration.

Release 2 (March 2026): Connected Agentic Pipeline & Web UI

Following the cloud tenant migration, we integrated TheHive alerts directly into the agent via Power Automate. When an alert arrives, the agent automatically reasons over severity, observables, indicators of compromise, and analyst notes, cross-referencing verified playbooks.

User feedback on the Teams interface showed it felt slow and opaque. I led the move from Teams to a dedicated, responsive web application that exposes the agent's step-by-step reasoning, tool invocations, and playbook citations in real time. Transparency resolved skepticism, and adoption held steady.

Product Management Role

I owned the product end-to-end, partnering with two engineers:

Results & Measurable Impact

What's Next

Our upcoming roadmap focuses on deepening connected security data sources: