The expected AI resolution rate for L1 tickets depends on your ticket mix, integrations and knowledge base quality, so set it from your own baseline. Measure it as L1 tickets closed without agent action that stay closed for seven days.
Published figures vary widely because vendors define "resolution" differently. A number from your own data is more useful. The method below gets you there in a few days of ticket analysis.
What is the expected AI ticket resolution rate for L1 tickets?
Your realistic rate is the share of L1 tickets that fall into categories AI can close end to end, multiplied by how often it succeeds in each. Estimate both before buying anything.
Step 1. Estimate the ceiling. Export three to six months of L1 tickets. Group them by category and count those with a repeatable fix the AI could carry out or explain.
Step 2. Check each category for what blocks automation.
| Driver | Raises the rate | Lowers the rate |
|---|---|---|
| Ticket mix | Many password, access, status and how-to tickets | Hardware faults, one-off errors, specialist apps |
| Integrations | AI can act in identity, endpoint and ITSM systems | AI can only answer with text |
| Knowledge base | Current articles with owners | Stale or missing articles |
| Identity verification | Automated, strong verification | Manual checks required |
| Channel | Portal or chat where users describe the issue | Phone and vague emails |
| Hand-off design | Clean escalation keeps trust high | Dead ends push users back to phone |
Step 3. Pilot and measure. Run one or two categories for four to six weeks and measure the real success rate. Extend category by category.
How do you measure AI resolution rate honestly?
Define terms before the pilot. Vendors and internal teams often mix different numbers under one label.
| Measure | Definition | Why it matters |
|---|---|---|
| Auto-resolution rate | L1 tickets closed with no agent action that stay closed for 7 days, divided by L1 tickets in the same categories | The core outcome |
| Pre-ticket deflection | Questions answered before a ticket was created | Real value, but a separate count |
| Assisted resolution | Tickets the AI routed or drafted, closed by an agent | Saves agent time; counted apart |
| Reopen rate | Auto-closed tickets reopened | Catches false success |
| Satisfaction | Short survey on AI-closed tickets | Shows whether users trust it |
| Agent minutes saved | Handling time before and after, per category | Converts results into cost |
Which L1 tickets can AI resolve?
Tickets with a clear fix and a system the AI can act on. Everything else is better served by assisted resolution.
- Password resets and account unlocks, after automated identity verification.
- MFA re-enrolment and VPN access restoration, under the same verification.
- Standard software requests through endpoint management (Intune, Jamf and similar) once a manager approves.
- Access to approved groups through the identity provider, with an approval step.
- Status questions about outages, maintenance and ticket progress, answered from the ITSM platform.
- How-to questions answered from a current knowledge article, often before a ticket exists.
A conversational front end for these requests is what our AI employee assistant provides, connected to the identity provider and the ITSM platform.
How does AI ticket classification and routing work?
The model reads each new ticket and proposes category, priority and resolver group; high-confidence tickets are routed automatically and the rest go to a dispatcher with suggestions.
- Read subject, description and structured fields such as device, department and location.
- Classify into your category tree, priority tier and resolver group.
- Apply a confidence threshold. Above it, route automatically. Below it, show the dispatcher the top suggestions for a one-click choice.
- Learn from corrections. Re-routes and dispatcher changes become labelled examples for the next evaluation and tuning cycle.
Measure routing accuracy on a held-out set of your own tickets before and after each change.
How do you automate Jira ticket processing with AI?
Connect a model to Jira Service Management through its REST API or automation webhooks, and write suggestions back into the ticket. Start with suggestions agents accept, then automate categories where accuracy is proven.
- Trigger: a new request fires a webhook or automation rule.
- Read: the integration fetches summary, description, reporter and linked assets.
- Decide: the model suggests request type, priority, team and a draft reply from the Confluence or knowledge base articles.
- Write back: suggestions go into fields and an internal comment; the reply stays a draft.
- Act (later): for proven categories, the integration transitions the ticket, triggers the password reset or access workflow and closes it.
- Log: every suggestion, acceptance and correction is stored for accuracy reports.
Atlassian also sells its own AI for Jira Service Management: Rovo, which it describes as "AI-powered support – from instant self-service, to powerful agentic ticket resolution with Rovo Service". Atlassian says Rovo is available on Standard, Premium and Enterprise Cloud plans. ServiceNow also has built-in AI features. A custom integration makes sense when you need your own model choice, data residency, or logic across several systems. For multi-step flows across ITSM, identity and endpoint tools, see AI workflow automation.
How does AI keep the IT knowledge base current?
AI drafts, monitors and flags; people approve.
- Drafts from resolved tickets. Symptom, diagnosis and fix are pulled from the ticket thread into an article draft for an expert to review.
- Staleness checks. Articles that are viewed but fail to resolve the issue get flagged for revision.
- Gap log. Questions with no matching article are grouped weekly and assigned to the right resolver team.
The AI can only resolve what the knowledge base covers, so article quality caps the L1 resolution rate.
How do you start an AI helpdesk pilot?
Pick two ticket categories, define the measures and run a short pilot with a clear decision at the end.
- Week 1: ticket export, category analysis, ceiling estimate, baseline handling time.
- Weeks 2–3: integration with the ITSM platform in suggestion mode; knowledge articles for the chosen categories reviewed.
- Weeks 4–6: auto-resolution switched on for the two categories, with verification and reopen tracking.
- Decision: extend, adjust or stop, based on auto-resolution, reopen rate and agent minutes saved.
Where tickets contain personal information of Quebec employees and the AI vendor processes data outside Quebec, Law 25 requires a privacy impact assessment first.
Remolda runs this as a six-week AI Pilot Sprint, a fixed-price package at $9,800 CAD, or starts with a one-week readiness review at $490 CAD.