// Pillar 03 · AI & ITSM Accelerators
AI & ITSM Accelerators: see it working before you commit
Reusable, working accelerators built from real service-delivery problems. Each one follows the same structure — problem, approach, technology, outcome, evidence — and is labelled honestly for what it is.
// The accelerators
Four accelerators, labelled honestly
Status labels used across this site: Production deployment (live in a client environment), Pilot (scoped trial with a client), Accelerator · live prototype (working build used to demonstrate the approach), and Reference architecture (a design model, not a deployment). None of the accelerators below is presented as a client production deployment.
SmartDesk — AI IT support triage
Accelerator · Live prototype
- Problem
- L1 and L2 service desks spend hours each day classifying, prioritising and routing tickets by hand.
- Approach
- A five-stage pipeline: ticket intake → AI analysis → smart routing → automatic notification → logging and dashboard. L1 / L2 / escalate routing with pre-filled agent briefings; AI-drafted user replies; Slack alerts for critical incidents.
- Technology
- Claude API, n8n, ServiceNow, Slack, email automation. Designed to integrate with existing ITSM tools.
- Outcome
- In the prototype, roughly 38–42% of tickets were auto-resolved, ~2.5 hours per day of manual triage saved, and triage time cut from about 8 minutes to about 90 seconds.
- Evidence
- Interactive prototype and written case study. Figures are prototype benchmarks, not a named-client result.
AI Incident Co-Pilot — intelligent incident response
Accelerator · Live prototype
- Problem
- During a P1, responders must correlate observability signals and draft stakeholder updates at the same time — and both slow resolution.
- Approach
- Real-time AI analysis of the live incident: likely causes, impact assessment and step-by-step resolution guidance, plus auto-drafted stakeholder communication. Every recommendation and draft is reviewed by a human before action.
- Technology
- LLM-based analysis with retrieval over incident knowledge; designed to integrate with Dynatrace, Splunk and ITSM toolchains.
- Outcome
- 20% MTTR reduction (prototype benchmark), with faster, more consistent incident communication.
- Evidence
- Live prototype and written case study. Treat the figure as an accelerator benchmark, not a client claim.
Enterprise AI Assistant — conversational AI for IT operations
Accelerator · Live demo
- Problem
- Operations teams lose time searching runbooks, knowledge bases and operational documentation for answers.
- Approach
- A retrieval-augmented conversational assistant that answers questions in natural language, grounded in your IT knowledge base, runbooks and operational documents.
- Technology
- Claude API, RAG with vector search, Python, Gradio, Hugging Face Spaces. Modular, so domain-specific knowledge bases can be swapped in.
- Outcome
- A working natural-language query interface over IT knowledge. No performance metric is claimed for this accelerator.
- Evidence
- Public live demo on Hugging Face Spaces.
AI Resilience by Design — interactive reference architecture
Reference architecture
- Problem
- AI becomes a single point of failure: if it is unavailable, wrong, replaced or more autonomous than intended, operations suffer.
- Approach
- A ten-layer reference architecture with AI-on, AI-degraded and AI-off operating modes, risk-tiered human approval, model portability and an explicit non-AI path for every business-critical capability.
- Technology
- AI gateway, model routing, retrieval, policy as code, AI observability. The interactive model is built in React and TypeScript.
- Outcome
- 8 failure simulations, a 14-dimension resilience scorecard and a CIO dashboard. All figures in the model are illustrative.
- Evidence
- Interactive model and overview page.