AI in IT Operations — Where the Real Productivity Lands
ITSM and IT operations are document-heavy, repetitive, and high-volume — well-matched to AI augmentation. The deployments that ship share recognisable shape; the ones that stall share recognisable failure modes.
IT service management and IT operations have been a target for AI augmentation for years. ITSM is document-heavy, repetitive, and high-volume — well-matched to current AI capabilities. The earlier waves of AIOps and chatbots produced limited results; the current wave with LLMs is producing more. The deployments that ship share recognisable patterns.
This piece is a practitioner view of AI in IT operations in 2025 — where the real productivity is landing, where the marketing exceeds reality, and what makes the difference.
What ITSM and IT ops actually involve
A working IT services organisation handles:
- Incident management — service is broken; restore it
- Problem management — figure out why incidents happen; prevent recurrence
- Change management — planned modifications to production systems
- Request fulfilment — routine user requests (access, equipment, software)
- Knowledge management — runbooks, procedures, documentation
- Service desk operations — frontline support for users
- Capacity and availability management — keeping enough resources, keeping services up
- Asset and configuration management — knowing what exists, how it relates
Each of these is potentially affected by AI. The actual impact varies.
Where AI is landing
Incident triage and routing
The most consistent production use case. Incoming incidents are classified by AI — affected system, category, severity, suggested team. The classification feeds the workflow.
The impact: mean time to assignment drops; the assigned team is more often right the first time; rework reduces.
The pattern works because the classification is bounded; the feedback signal is fast (the assignee accepts or reassigns); the cost of an occasional misclassification is low.
Service desk first-line
AI agents handle the routine: password resets, access requests, status inquiries. Users get faster resolution; agents handle fewer ticket categories.
The deflection rate varies (30-60% for well-built systems). The cost case is real where the volume is high. The pattern doesn't replace agents; it shifts what agents handle.
Knowledge base augmentation
AI surfaces relevant knowledge base articles for incoming tickets. The agent sees them alongside the ticket; the user sees them in self-service interfaces.
The impact is in time-to-resolution. Articles that were hard to find become easy to surface.
Runbook drafting
Operations runbooks for new systems get drafted by AI from system documentation. The team curates and refines. Documentation that wouldn't have been written gets written.
Post-incident drafting
After an incident, AI drafts the post-mortem from the timeline, the tickets, the chat history. The team refines. Documentation that was often deferred ships.
Log and trace analysis
AI summarises log spikes, identifies patterns, surfaces likely causes. Engineers investigate the candidates rather than searching from scratch.
Change risk assessment
AI assesses planned changes against similar past changes, surfacing risk factors. Change advisory boards use this as input.
Where AI underperforms in IT ops
Autonomous incident resolution
The aspiration of AI resolving incidents end-to-end. In practice, the reliability for autonomous resolution is too low; incidents that misresolve cause secondary incidents.
The realistic pattern: AI proposes; engineers execute. The boundary stays explicit.
Root cause analysis from first principles
Identifying the underlying cause of complex incidents requires investigative reasoning across many systems. AI helps with hypotheses; the investigation is human.
Complex change planning
For non-trivial changes, the planning requires reasoning about system interactions AI doesn't fully have. AI suggests; engineers plan.
Capacity forecasting at scale
Predicting future capacity needs requires reasoning about business drivers, seasonality, and growth patterns. AI helps; the forecast remains a human decision.
The integration with existing ITSM platforms
Enterprises typically run ServiceNow, BMC, Jira Service Management, or similar. The integration is where most of the work goes:
Bots inside the platform
Many ITSM platforms now have native AI assistance. Adoption is uneven; the capabilities are improving.
Custom integrations
Where the platform's native AI is inadequate or unavailable, custom integrations through APIs. The platform remains the system of record; the AI is a service alongside.
Slack and Teams integrations
A growing pattern: the user interacts through their messaging platform, the AI bridges to ITSM. Friction reduces; adoption rises.
Self-service portals
User-facing self-service with AI assistance. The portal looks like a chat; the underlying integration with ITSM creates tickets, fulfils requests, surfaces status.
What makes deployments successful
Strong knowledge base
The same pattern as in customer support. AI is an amplifier; the knowledge base is what gets amplified. Teams that invest in the knowledge base see disproportionate returns.
Clear escalation paths
When AI can't help, the path to a human is fast and friction-free. This determines satisfaction more than AI's success rate.
Integration with existing tools
The AI fits the team's existing workflow. It doesn't add a new tool to learn; it makes the existing tools faster.
Measurement on outcomes
Time to assignment, time to resolution, deflection rate, ticket re-open rate. Specific metrics with targets.
Human oversight on consequential actions
AI proposes restarts, changes, access grants. Humans approve and execute (except for the most routine actions where the audit trail justifies automation).
What we keep seeing
Patterns in enterprise IT operations AI engagements:
The first quick win is incident routing. Lower-risk, high-volume, well-defined. Most engagements start here.
Service desk augmentation is real value. Productivity gains for desk agents are measurable.
Knowledge management improves materially. The catalogue gets better; institutional knowledge gets captured.
Autonomous resolution stays aspirational. The reliability isn't yet at the level required for autonomous fixes.
Vendor sprawl emerges. Multiple AI tools for different ITSM functions. Consolidation through the ITSM platform's native capability is common.
The IT operations team's role evolves. Less time on routine; more on harder problems, on improving the system, on supporting users at higher levels.
What we recommend
For enterprise IT operations teams adopting AI in 2025:
- Start with incident triage and routing. Lowest risk, fastest payoff.
- Invest in the knowledge base. The AI's value is bounded by its content.
- Integrate with the existing ITSM platform. Don't create a parallel surface.
- Build escalation paths carefully. User experience depends on what happens when AI fails.
- Measure outcomes, not initiatives. Time to resolution, deflection, satisfaction.
- Keep humans in the loop for consequential actions. The autonomy aspirations are not yet aligned with reality.
- Engage the operations team as collaborators, not just users. Their adoption determines the success.
AI in IT operations in 2025 produces meaningful productivity gains where it is deployed deliberately. The teams that integrate with existing workflows and invest in the foundations capture real value. The teams that chase autonomous resolution or replace existing systems wholesale produce expensive disappointments. The discipline that produces value is the same as in other AI domains; the domain-specific patterns are recognisable now.
RELATED READING
More from the field.
Service practices the article draws on, related programmes, and other pieces on adjacent topics.
Service practices
Related pieces
From AI Pilot to Production — The Playbook That Bridges the Gap
Every enterprise has AI pilots. Far fewer have AI in production. The bridge between the two is more about organisational discipline than technical capability. A practitioner playbook.
LLMOps Maturity — A Practitioner's Maturity Model
Most enterprises are operating LLM workloads on engineering intuition alone. A maturity model helps locate where you are, what to invest in next, and what the next stage actually requires.
LLM Cost Discipline — Engineering Practices That Keep Bills Predictable
Most teams discover LLM cost through the bill. By then, the cost shape is set and hard to change. The engineering practices that keep costs predictable are not exotic, but they have to be in place from the start.
Discuss this work
Bring an enterprise programme.
If anything in this piece resonates with what you're building, talk to us. Senior practitioners engage directly on architecture and delivery.
Work with the practitioners
Bring an enterprise programme.
Architecture audit, new delivery, modernisation, or in-flight rescue — Intellectual engages directly on enterprise programmes with senior practitioners.