AI in the help desk
how many tickets really get handled automatically?
Gartner says 80 percent - we check it against real deployments. No marketing wrapper, just the numbers.
"AI will solve 80 percent of your tickets without human involvement." I have heard this line at a conference, in a vendor pitch deck and in several industry articles back in 2024. The number comes from a Gartner report and is real, but it carries conditions that are rarely quoted: it applies to selected interaction types in organizations that have invested in automation for at least 2 years. In year 1 of AI deployment, the realistic figure is 20-40 percent.
Note: the 80 percent Gartner forecast applies to 2029, to agentic AI in B2C customer service, not to the internal IT help desk. For internal IT environments, automation rates are realistically lower and depend on knowledge base maturity and integrations.
4 levels of AI in the help desk - what each one actually does
Level 1 - Automatic classification and routing (Tier 0)
AI reads the content of a ticket and assigns: category (hardware/software/access), priority, responsible technician. The user still waits for a human response, but the ticket reaches the right person faster. Real value: shortening resolution time by 20-30 percent without automating the resolution itself. Works well already with 100-200 historical tickets.
Level 2 - Knowledge base suggestions (Tier 1)
AI suggests knowledge base articles to the user before the ticket reaches a technician. If the user resolves the problem themselves, the ticket is closed without technician involvement. Real value: 15-30 percent of tickets deflected. Condition: a solid knowledge base - without it, AI has nothing to suggest.
Level 3 - Virtual agent / chatbot (Tier 2)
AI runs a conversation with the user, gathers information about the problem and tries to resolve it through prebuilt workflows: password reset, account unlock, granting application access. Real value: 25-50 percent of tickets in a given category solved automatically. Condition: integration with systems (Active Directory, business applications, CMDB).
Level 4 - Generative AI assisting the technician
AI does not solve the ticket instead of the technician - it assists the technician. It suggests solutions based on history, analyzes symptoms, generates a first draft answer ready for editing. Real value: 20-40 percent shorter resolution time per technician.
How many tickets AI really solves - indicative ranges
The ranges below are indicative estimates of automation potential per ticket category, not a guarantee. Actual numbers depend on knowledge base maturity, integration scope and company specifics.
Categories with the highest automation potential
| Ticket category | Automation potential | Condition |
|---|---|---|
| Password reset | 70-90% | AD integration, MFA configured |
| AD account unlock | 60-80% | AD integration |
| Application access request | 40-60% | Application catalog and approval workflow |
| IT questions (where is X, how do I do Y) | 30-50% | Good knowledge base, minimum 50 articles |
| New employee onboarding | 50-70% | Defined workflow, HR integration |
Categories with low automation
| Ticket category | Automation potential | Why it is low |
|---|---|---|
| Physical hardware failure | 5-15% | Requires physical intervention |
| Issue specific to a business application | 10-20% | Too much variability, business context |
| Server or infrastructure outage | 5-10% | Requires an expert, the risk is too high |
Realistic overall numbers for a 100-300 employee company with a solid knowledge base, AD integration and 12 months of AI maturity: 20-40 percent of all tickets without technician intervention. The 80 percent from the Gartner report is the maximum ceiling in optimal conditions, reachable after more than 2 years of system maturity.
How to calculate ROI - sample math
The numbers below are an illustrative example, not the result of any specific deployment. They show the calculation method - plug in your own data to get a realistic picture for your company.
Example - starting point (no AI): assume a 200-employee company, 600 tickets per month, 45-minute average handling time, 4 IT technicians.
Example - after AI deployment in year 1 (assume a cautious 25 percent deflection rate):
- ~150 tickets resolved automatically, ~450 going to technicians
- If a technician works faster with AI suggestions, say average handling time of 35 minutes instead of 45
- Total: 450 x 35 minutes = 262 hours (instead of 450 hours)
- Difference: ~188 hours per month - multiply by your own hourly rate to estimate the saving
Each of these variables (ticket count, handling time, hourly rate, actual deflection rate) depends on the specific company - which is why we do not give a single "annual savings" figure. A credible calculation only emerges after analysis of real ticket data.
Cost of AI rollout in SDP: Zia AI is available in higher SDP editions (Professional/Enterprise) - no separate licensing costs beyond the edition itself; the scope of features depends on the version and plan. On top of that, there is a one-off cost of configuration and knowledge base build (category mapping, chatbot launch, first knowledge base articles) - its size depends on project scope.
For more details on Zia AI and its capabilities, see the article AI in ITSM 2026: how AI is changing the IT help desk.
Zia AI in ManageEngine ServiceDesk Plus
Zia is the name of the ManageEngine AI assistant. The Professional and Enterprise editions of SDP include specific features:
- Automatic classification: Zia analyzes ticket content and assigns category, priority and technician based on historical patterns. Accuracy grows with the number of correctly categorized historical tickets - the richer and cleaner the history, the better the model fit.
- Knowledge base suggestions: Before the user submits a ticket, Zia shows them 3-5 articles that may solve the problem.
- Virtual agent (chatbot): Conversational interface on the self-service portal. Guides the user through flows: "Is the problem X or Y? Try step 1... Didn't help? I'm creating a ticket for the technician."
- Zia Assistant for technicians: The technician opens a ticket and Zia shows similar tickets resolved in the past, a suggested solution and the time remaining to SLA breach.
- Sentiment analysis: Zia rates the tone of submissions and flags tickets from frustrated users - the technician knows to respond faster.
5 conditions for AI to actually work
- A solid knowledge base: A minimum of 30-50 articles covering the most frequent ticket categories. Before rolling out AI: spend 2-3 weeks building the knowledge base from historical tickets.
- Active Directory integration: Password reset and account unlock (the most automatable categories) require AD integration. Without it, the chatbot can record a request but cannot reset the password on its own.
- Historical ticket data: AI classification learns from history - the more tickets with correct categories, the better. If you are moving from email, importing historical data is critical.
- A user who trusts the system: Deflection only works if users actually use the self-service portal. Portal adoption above 70 percent is a precondition for AI effectiveness.
- Time to mature: Plan 3-6 months for setup and learning, 6-12 months for full maturity. If you are still rolling out the help desk, see ManageEngine for a 10-100 employee company, where we also describe what needs to be in place before enabling AI.
FAQ
Will AI replace my IT technicians?
No. AI shortens the handling time of simple tickets and lets technicians focus on harder problems. The typical effect: the same number of technicians serves a larger user base, or the same user base in less time. Replacing technicians is not realistic for 100-500 employee companies within the next 3 years.
How long does an AI rollout in ServiceDesk Plus take?
Zia configuration: 2-3 weeks. Knowledge base build (first version): 2-3 weeks. Chatbot launch and testing: 1 week. Total: 5-7 weeks. Full effects after 3-6 months of usage. Zia is available in SDP Professional and Enterprise - if you already have those editions, the cost is only configuration.
Does AI work as well in Polish as in English?
Zia implemented in SDP works in the language of the tickets - if the knowledge base is in Polish and tickets are in Polish, AI learns on Polish text. The quality is slightly lower than for English (less training data within Zoho), but the gap does not disqualify the solution. The Polish configurations we have seen worked effectively.
What matters more: a good knowledge base or AD integration?
If your company has a lot of password resets and account unlocks (typical for companies without self-service), AD integration will deliver faster ROI. If you have lots of "how do I do X in system Y" questions, the knowledge base is the priority. Most often it is worth starting both in parallel - they usually take similar time.
How many tickets would AI solve in your company?
In 45 minutes we will analyze your ticket categories and tell you honestly: what percentage can be automated and how the ROI looks for your specifics.
Book a free ROI analysis