ai itsm

What is AI ITSM? A practical
guide for IT Teams

What is AI ITSM? A practical guide for IT Teams

Since the 1980s, IT support has run on the same loop: a request comes in, someone logs it, someone triages it, and it gets escalated to a human who actually fixes it. AI ITSM breaks that loop at the step that costs you the most, which is the wait for a person to pick the ticket up.

Agents

Workflows

Service Desk

Analytics

Agents

Workflows

Service Desk

Analytics

Gartner

80%

Of common customer service issues resolved by agentic AI without a human by 2029 (forecast)

Gartner

30%

Reduction in customer service operating costs Gartner expects to follow

NBER

14%

More issues resolved per hour by support agents with AI assistance

NBER

34%

Gain for novice and low-skilled agents in the same study

overview

What is AI ITSM?

AI ITSM (AI IT service management) uses large language models and AI agents to understand employee requests, answer them from your own knowledge and resolve common tickets. It does that by acting in the systems you already run, like your identity provider, MDM and ticketing tool. Humans still own the exceptions; the AI handles the volume.

The term gets stretched to cover everything from auto-categorization to fully autonomous agents, so be precise about what you're buying. If the tool suggests a knowledge base article, that's AI-assisted ITSM. If it verifies the user, triggers the MFA reset in Okta and closes the ticket in the thread, that's the thing this page is about.

overview

What is AI ITSM?

The shift

AI ITSM vs traditional ITSM

ITIL gave IT a shared vocabulary for incident, change and problem management, and nothing about AI makes those practices wrong. What has aged is the delivery model around them: the portal, the seven-field form, the category dropdown nobody picks correctly, and the queue where a simple access request waits hours for someone to read it.

Traditional ITSM

AI ITSM

Where requests start

A portal form with required fields

A message in Slack or Teams, in plain language

Ticket creation

The employee picks a category and hopes it's right

The AI reads the request, classifies it and routes it

Common requests

Wait in a queue for a human

Resolved in the thread, escalated when needed

Knowledge base

Written by hand and rarely updated

Drafted from solved tickets, reviewed by your team

Automation

Rules built and maintained by a specialist admin

Workflows with approvals handled in chat

Reporting

Ticket volume and resolution time

Deflection, knowledge gaps and every escalation

Your employees don't live in portals. They live in Slack and Microsoft Teams. Every time you send them somewhere else to ask for help you pay for it, in lost time and in requests that never get filed because people just DM whoever on IT they know.

The evolution

How we got here, fast

How we got
here, fast

The journey from manual ticketing to autonomous   resolution in five eras.

Manual ITSM

Humans triaged, assigned, resolved. ITIL formalized the practices.

1980s–2000s

2000s–2015

Workflow automation

ServiceNow, BMC, Cherwell. Drag-and-drop workflows, approval routing.

AIOps + ML assist

ML helps classify tickets, recommend KB, detect anomalies. AI supports agents.

2015–2022

2022–2024

Generative AI

LLM-powered summarisation, draft replies, KB generation. AI entered the conversation.

Agentic ITSM

Autonomous agents resolve tickets using multi-step reasoning and real tools.

2024–Present

Manual ITSM

1980s–2000s

Humans triaged, assigned, resolved. ITIL formalized the practices.

Workflow automation

2000s–2015

ServiceNow, BMC, Cherwell. Drag-and-drop workflows, approval routing.

AIOps + ML assist

2015–2022

ML helps classify tickets, recommend KB, detect anomalies. AI supports agents.

Generative AI

2022–2024

LLM-powered summarisation, draft replies, KB generation. AI entered the conversation.

Agentic ITSM

2024–present

Autonomous agents resolve tickets using multi-step reasoning and real tools.

Beyond chatbots

How does AI ITSM
actually work?

Most AI ITSM platforms run the same four steps on every request, and the quality of each step is where vendors really differ.

step 1

Understand the request

The model reads intent rather than keywords, so "can't get into Figma since my laptop got swapped" lands as an access issue tied to a device change.

step 1

Understand the request

The model reads intent rather than keywords, so "can't get into Figma since my laptop got swapped" lands as an access issue tied to a device change.

step 1

Understand the request

The model reads intent rather than keywords, so "can't get into Figma since my laptop got swapped" lands as an access issue tied to a device change.

step 2

Ground it in your knowledge

It pulls from your own sources (Confluence, Notion, Google Drive, past tickets) and cites them, which is what separates a useful answer from a confident guess.

step 2

Ground it in your knowledge

It pulls from your own sources (Confluence, Notion, Google Drive, past tickets) and cites them, which is what separates a useful answer from a confident guess.

step 2

Ground it in your knowledge

It pulls from your own sources (Confluence, Notion, Google Drive, past tickets) and cites them, which is what separates a useful answer from a confident guess.

step 3

Act inside guardrails

For requests it's allowed to resolve, it calls the tools you've connected: a password reset in your identity provider, an app request routed to the right approver, a fix pushed through your MDM.

step 3

Act inside guardrails

For requests it's allowed to resolve, it calls the tools you've connected: a password reset in your identity provider, an app request routed to the right approver, a fix pushed through your MDM.

step 3

Act inside guardrails

For requests it's allowed to resolve, it calls the tools you've connected: a password reset in your identity provider, an app request routed to the right approver, a fix pushed through your MDM.

step 4

Escalate with context, then learn

Anything ambiguous or high risk goes to a human with the full thread attached, and the human's fix feeds back into the knowledge base.

step 4

Escalate with context, then learn

Anything ambiguous or high risk goes to a human with the full thread attached, and the human's fix feeds back into the knowledge base.

step 4

Escalate with context, then learn

Anything ambiguous or high risk goes to a human with the full thread attached, and the human's fix feeds back into the knowledge base.

The piece most legacy tools miss is where the conversation lives. In a portal model, the back-and-forth gets flattened into ticket fields and thrown away. In an AI-native model the thread in Slack or Teams is the record, so the context from last quarter's VPN outage is still there when the same question comes back, and every resolved thread becomes material for the next answer.

Beyond chatbots

The AI ITSM use cases
that pay off first

Start with requests that are high volume, low ambiguity and reversible, because those are the ones an AI can resolve end to end without anyone losing sleep.

Reset passwords and MFA

Verifies the requester, triggers the reset in your identity provider and confirms in the thread.

Reset passwords and MFA

Verifies the requester, triggers the reset in your identity provider and confirms in the thread.

Software access requests

Routes to the right approver, triggers provisioning through your IdP and logs who approved what.

Software access requests

Routes to the right approver, triggers provisioning through your IdP and logs who approved what.

Triage and routing

Classifies, prioritizes and routes everything it can't resolve on its own.

Triage and routing

Classifies, prioritizes and routes everything it can't resolve on its own.

Device diagnostics

Runs first-line checks through your MDM before a human gets involved.

Device diagnostics

Runs first-line checks through your MDM before a human gets involved.

Knowledge base upkeep

Drafts articles from solved tickets for your team to review and publish.

Knowledge base upkeep

Drafts articles from solved tickets for your team to review and publish.

Onboarding and offboarding

Orchestrates the checklist across HR, your IdP and MDM so nothing gets skipped.

Onboarding and offboarding

Orchestrates the checklist across HR, your IdP and MDM so nothing gets skipped.

Approvals in chat

Approvers say yes or no in Slack or Teams with the request in front of them.

Approvals in chat

Approvers say yes or no in Slack or Teams with the request in front of them.

Knowledge answers

Answers how-to questions from your docs, with a source link, in seconds.

Knowledge answers

Answers how-to questions from your docs, with a source link, in seconds.

The best research on generative AI in support work comes from customer service rather than IT, so hold it loosely. An NBER study of 5,179 customer support agents found AI assistance raised issues resolved per hour by 14% on average. Novice and low-skilled agents gained 34%, while experienced agents barely moved. For an IT team that reads as a tier-1 story, because the biggest lift lands on the people working your most repetitive tickets.

Agentic AI

Where agentic AI fits in ITSM

Agentic AI is the part of AI ITSM that acts. A chatbot surfaces an article; an agent works out what needs to happen, does it across your systems inside the boundaries you set, and reports back in the thread.

Gartner predicts agentic AI will autonomously resolve 80% of common customer service issues without human intervention by 2029. That forecast is about customer service, not internal IT, so treat it as a direction of travel rather than a target for your queue.

The hard part isn't the reasoning, it's the boundaries, and our guide to agentic AI for ITSM goes deeper on how to set them.

Agentic AI

Where agentic AI fits in ITSM

The hard part isn't the reasoning, it's the boundaries, and our guide to agentic AI for ITSM goes deeper on how to set them.

Beyond chatbots

Three ways AI ITSM
goes wrong

Every automation you ship either earns trust or spends it, and an AI spends it faster when it gets something wrong in front of an employee.

pitfall 1

Acting where it should ask

A remote wipe can't be undone, and a privileged access change can turn into an incident. Those stay behind a human approval, every time.

pitfall 1

Acting where it should ask

A remote wipe can't be undone, and a privileged access change can turn into an incident. Those stay behind a human approval, every time.

pitfall 1

Acting where it should ask

A remote wipe can't be undone, and a privileged access change can turn into an incident. Those stay behind a human approval, every time.

pitfall 2

Answering from stale knowledge

If your docs are wrong, the AI will be wrong with confidence. Clean up the articles behind your most common requests before you switch anything on.

pitfall 2

Answering from stale knowledge

If your docs are wrong, the AI will be wrong with confidence. Clean up the articles behind your most common requests before you switch anything on.

pitfall 2

Answering from stale knowledge

If your docs are wrong, the AI will be wrong with confidence. Clean up the articles behind your most common requests before you switch anything on.

pitfall 3

Bolting AI onto a portal nobody uses

Adding AI to a portal employees already avoid doesn't change where they ask for help. Put it in Slack or Teams, where the requests actually start.

pitfall 3

Bolting AI onto a portal nobody uses

Adding AI to a portal employees already avoid doesn't change where they ask for help. Put it in Slack or Teams, where the requests actually start.

pitfall 3

Bolting AI onto a portal nobody uses

Adding AI to a portal employees already avoid doesn't change where they ask for help. Put it in Slack or Teams, where the requests actually start.

Build vs buy

Should you build or buy AI ITSM?

Most teams end up asking whether to add AI to the ITSM platform they already pay for or adopt one built around it. Both can work; the trade-off is how much of the build and the upkeep you want to own.

Build on your current ITSM

Buy an AI-native platform

Time to value

Months of build and testing before anything goes live

Faster, because the models and integrations already exist

Skills you need

Engineers who can build, test and run LLM workflows

Admins who know your processes and approval rules

Upkeep

Your team owns every model change, prompt and integration

The vendor ships the updates; you own the rules

Where employees ask

Usually the existing portal, with a chat layer added on

Slack or Teams from day one

Fit to your process

Shaped exactly around how you work today

You may adjust a few workflows to fit the platform

Best for

Teams with in-house ML engineers and unusual requirements

Teams that want resolved tickets this quarter without hiring for AI

If nobody on your team wants to own prompts, evaluations and integrations for the next few years, buying is usually the faster path to resolved tickets. Buying doesn't mean giving up control; you still decide what the AI does on its own and what needs an approval.

Buyer checklist

How should you evaluate
an AI ITSM tool?

How should you evaluate an AI ITSM tool?

Ask every vendor the same five questions. If you want those answers side by side, our comparison of the best AI-native ITSM platforms lays them out vendor by vendor.

Does it resolve requests end to end, or just suggest articles?

Does it work natively inside Slack or Teams, or send people to a portal?

Can you decide exactly what it does on its own and what needs approval?

Does every action leave an audit trail you could hand to an auditor?

Can it show deflection and the knowledge gaps it found, not just ticket counts?

Rollout

How do you roll out
AI ITSM?

Before phase one, pull last quarter's tickets and baseline the five request types that make up most of your volume. Tell your service desk early that the AI is taking the repetitive tier-1 work off them, because the rollout goes nowhere if they quietly route around it.

phase 1

Pilot the high-volume requests

Pick password and MFA resets, common access requests and how-to questions. Turn on automated resolution for those only, and read every thread until you trust the answers.

phase 1

Pilot the high-volume requests

Pick password and MFA resets, common access requests and how-to questions. Turn on automated resolution for those only, and read every thread until you trust the answers.

phase 1

Pilot the high-volume requests

Pick password and MFA resets, common access requests and how-to questions. Turn on automated resolution for those only, and read every thread until you trust the answers.

phase 2

Widen the coverage

Add multi-step approvals, onboarding and offboarding, and requests that cross IT, HR and Ops. Keep a human approval in front of anything privileged or destructive.

phase 2

Widen the coverage

Add multi-step approvals, onboarding and offboarding, and requests that cross IT, HR and Ops. Keep a human approval in front of anything privileged or destructive.

phase 2

Widen the coverage

Add multi-step approvals, onboarding and offboarding, and requests that cross IT, HR and Ops. Keep a human approval in front of anything privileged or destructive.

phase 3

Get ahead of the queue

Use what the AI has resolved to draft knowledge from solved tickets, flag gaps in your docs and spot repeat issues before they pile up in the queue.

phase 3

Get ahead of the queue

Use what the AI has resolved to draft knowledge from solved tickets, flag gaps in your docs and spot repeat issues before they pile up in the queue.

phase 3

Get ahead of the queue

Use what the AI has resolved to draft knowledge from solved tickets, flag gaps in your docs and spot repeat issues before they pile up in the queue.

Measurement

Which KPIs show it's working?

How should you evaluate an AI ITSM tool?

Baseline these for your top request types before you switch anything on, then compare month over month. Ticket counts on their own won't tell you much.

Automation rate: the share of requests resolved with no human touch.

Mean time to resolution, split by what the AI handled and what it escalated.

Deflection: questions answered from your docs before a ticket ever existed.

Escalation quality: how often a human reopens or redoes the AI's work.

Employee CSAT, asked in the thread right after the request closes.

Ecosystem

Tools that integrate
with AI ITSM

AI ITSM connects with the systems you already use, so there's nothing to rip and replace.

Team chat platforms

Identity providers

Ticketing platforms

Knowledgebases

Knowledge

bases

Device management tools

HRM systems, and more

Stop escalating what

you can resolve

The old loop of log, triage, escalate was built for a world where only a human could fix anything. If you want to see how we do it inside Slack and Teams, we're happy to walk you through Ravenna, or you can poke around the docs first.

Book a Demo

Stop escalating what you can resolve

Book a Demo

Built for enterprise,
secure by design

From request to resolution, give your teams full context, automated workflows, and AI assistance, all within Slack.

SOC2

HIPAA

GDPR

Loved by employees at:

FAQ

How does AI work in IT service management?

It reads requests in plain language and answers them from your own knowledge sources. Routine tickets get resolved through the tools you've connected, such as your identity provider or MDM. Anything ambiguous or risky goes to a human with the full context attached.

How does AI work in IT service management?

How can AI reduce manual workloads in ITSM?

It takes the repetitive tier-1 work off your team: password and MFA resets, access requests, onboarding steps and how-to questions. Even on tickets it can't resolve, automatic classification and routing removes a manual step from every request.

How can AI reduce manual workloads in ITSM?

What's the ROI of an AI-powered ITSM platform?

It depends on your ticket mix, so be wary of any vendor quoting one universal number. Baseline volume, resolution time and cost per ticket for your top five request types. Then track how many the AI resolves end to end, and what happens to resolution time for everything else.

What's the ROI of an AI-powered ITSM platform?

Can you use AI in ITSM without losing control?

Yes, as long as the platform lets you decide what it can do on its own. Keep low-risk, reversible actions automated, put an approval in front of anything privileged or destructive, and make sure every action is logged.

Can you use AI in ITSM without losing control?

Designed and built in Seattle, WA
— Powered by AI.

Products

Blog