50% lower MTTR with an agent that recommends the fix and runs it on approval

An incident recommendation agent for a global engineering software enterprise that analyzes real-time incident data against historical resolutions, surfaces fix paths from internal knowledge, and executes validated steps with a human in the loop.

Outcome

MTTR down 50%, alert noise down 30%, and root cause analysis running automatically, with the manual investigation phase after RCA handled by the agent.

See the results

Talk to an amp'd specialist

At a glance

Client & industry

Global engineering software enterprise, with incident knowledge spread across tickets, documents, forums, and runbooks

Engagement

Agentic incident recommendation, from knowledge lookup to auto-resolution with human approval

Maps to below services

MLOps & Scalable ML Platforms, Agentic AI & AgentOps

Where it started

Every incident meant hunting across fragmented tools before anyone could start on the fix.

Business problem

Engineers spent critical time searching ITSM, CMDB, monitoring, and chat logs just to understand a problem. Without unified context, even simple incidents turned into multi-hour investigations, and MTTR climbed.

Technical barriers

MTTR was dominated by alert noise, manual root cause analysis, and slow escalation chains. After RCA, the fix path still had to be found by hand across internal knowledge.

Agivant's role

Agent engineering, hybrid search and knowledge graph design, and human-in-the-loop remediation workflows

What Agivant built

Agivant built the recommendation agent to analyze real-time incident data against historical resolutions. An AI intelligence layer handles issue detection, resolution matching, duplicate finding, and RAG with Gemini models. It draws on a hybrid search index of keyword and semantic vectors and a knowledge graph of tickets, content, and their relationships, fed by documents, forums, runbooks, ITSM tickets, and observability data through real-time and batch ingestion. Validated remediation steps run once a person approves. Value in days and weeks.

Alert-triggered workflows

Incidents start the agent automatically through alert-triggered workflows in the orchestration layer.

Resolution matching

Compares each live incident against historical resolutions to find the closest proven fix.

Automatic root cause analysis

Identifies the root cause automatically before a fix is recommended.

Knowledge lookup

Surfaces fix paths from internal knowledge sources, replacing the manual investigation phase after RCA.

Validated remediation steps

Recommends remediation steps drawn from proven resolutions and internal knowledge.

Human-in-the-loop execution

Executes the recommended steps once an engineer approves them.

Noise and duplicate reduction

Issue detection and duplicate finding cut alert noise before incidents reach the agent.

How it is built

Orchestration	Models	Search	Knowledge graph	Knowledge sources	Ingestion	Controls
Alert-triggered workflows, Agent Engine with A2A protocol	Gemini, RAG engine	Hybrid index, keyword and semantic vectors	Tickets, content, and relationships, on Spanner Graph and BigQuery	Documents, forums, runbooks, ServiceNow and Jira tickets, Cloud Logging	Application Integration, Dataflow, Document AI, real-time and batch sync	Engineer approval before execution

Before

Engineers hunting across fragmented tools, noisy alerts, manual RCA, slow escalation chains, and fix paths found by hand.

After

Incidents matched to historical resolutions, root cause captured automatically, and validated fixes executed after human approval.

Proof in production

An alert fires and the agent checks the incident against historical resolutions in the knowledge graph and hybrid index. It confirms the root cause, pulls the matching fix path from internal knowledge, and lays out validated remediation steps. An engineer approves, and the agent executes them.

Measured results

Faster resolution with less noise, and engineers approving fixes the agent has already researched.

50%

MTTR reduction

30%

Less alert noise

Automatic

Root cause analysis

Human-approved

Every executed fix

The repeatable pattern

Reusable asset

The recommendation agent is one of the ticket-lifecycle agents in Agivant's agentic IT support platform, working alongside the triage, investigate, lifecycle, and support assist agents. Every build amplifies the next.

Challenges overcome

Matching each live incident to the right historical resolution through heavy alert noise, and keeping every executed fix under engineer approval.

Enterprise movement

Foundational

Operational

Path to autonomous