AI for ITSM — GenAI-Powered IT Service Management

The Challenge: GenAI Pilots That Never Reach Production

Most enterprises have run a GenAI pilot for IT service management — a chatbot proof of concept, an LLM experiment on top of a knowledge base — that never made it past a demo. The gap is rarely the model; it's the absence of someone who understands both the AI tooling and the operational reality of ITSM well enough to connect the two.

Ticket classification needs to integrate with real ServiceNow workflows. Incident co-pilots need to respect existing escalation paths and stakeholder communication norms. Knowledge assistants need to be grounded in actual runbooks, not generic training data. Without that operational grounding, GenAI for ITSM stays a pilot indefinitely.

There's also a trust problem. Operations teams have seen enough hype cycles to be skeptical of AI tools that promise to 'automate everything.' Adoption only happens when the AI is positioned correctly — as something that removes the repetitive 80% of triage work so people can focus on the 20% that genuinely needs human judgment, with a clear, auditable trail of what the AI did and why.

My Approach: Production AI, Not Just Prototypes

I build GenAI tools for ITSM the same way I've run service delivery for two decades — with a bias toward measurable operational outcomes over technical novelty. My flagship implementation, Smart Desk, is an AI-powered IT support triage system built on Claude API, n8n, and ServiceNow that classifies, prioritizes, and routes tickets automatically — achieving roughly 42% ticket auto-resolution and saving approximately 2.5 hours per day in manual triage effort in the live prototype (a benchmark, not a named-client result).

Beyond triage, I've built AI Incident Co-Pilot, a GenAI tool that assists operations teams with real-time incident analysis, resolution guidance, and auto-drafted stakeholder communications, and an Enterprise AI Knowledge Assistant — a RAG-powered conversational system (Claude API, vector search, deployed on HuggingFace Spaces) that lets enterprise teams query IT knowledge bases and runbooks in natural language, reducing knowledge retrieval time and improving first-contact resolution.

Every implementation follows the same principle: the AI augments a defined ITSM workflow rather than replacing the workflow itself. Tickets that the system isn't confident about still route to a human. Incident communications drafted by the AI are reviewed, not auto-sent. That governance layer is what makes these tools survive contact with a real operations team, rather than getting shelved after a pilot.

Tools & Technologies

Engagement Process

1. Workflow Audit

Identify which ITSM workflows — ticket triage, incident response, knowledge lookup — have the highest manual effort and the clearest definition of 'correct' for an AI to learn from.

2. AI Tool Design

Design the specific GenAI implementation (triage classifier, incident co-pilot, RAG knowledge assistant) around your existing ServiceNow data and escalation structure.

3. Build & Integrate

Implement using Claude API, n8n, and direct ServiceNow integration, with human-in-the-loop checkpoints for anything the AI isn't confident about.

4. Pilot, Measure, Scale

Run a scoped pilot, measure auto-resolution rate and time saved against baseline, then expand scope once the model and workflow are proven.

Built For Enterprise Decision-Makers

Operational outcomes

These aren't slideware projections — Smart Desk and AI Incident Co-Pilot are working, live-demo accelerators with measured prototype benchmarks, designed to integrate with ServiceNow-based service delivery workflows. The 96% → 99.8% SLA figure comes from an enterprise delivery engagement.

Why This Works

GenAI for ITSM is a crowded market right now, but much of what's on offer is built by AI specialists who've never run a service desk, or by ITSM specialists with no hands-on AI implementation experience. Smart Desk, AI Incident Co-Pilot, and the Enterprise AI Knowledge Assistant were all built by someone who has done both — which tends to be the difference between a working production tool and another pilot that quietly disappears after the demo.

Frequently Asked Questions