How to read vendor AI claims on ITSM pages: resolution rates, setup times and agent names

AI claims are now the most prominent numbers on ITSM product pages. They are useful as signals of what a vendor is aiming for, and risky as forecasts of what your team will see. This post explains how we treat them on this site and how to read them yourself. Every example below comes from a vendor page read on 27 September or 11 October 2026.

What kinds of claims appear?

Three kinds come up repeatedly:

• Resolution rates. SysAid states its AI can "automatically resolve more than 60% of incidents before they ever become tickets". Freshworks states that Freddy AI agents "instantly handle up to 80% of routine queries", in a general service context rather than for Freshservice specifically.
• Setup times. SysAid states a help desk can be "up and running in 30 days or less". monday.com states that most teams can go live in days.
• Capability descriptions. monday service says its Intake & Triage Agent "classifies and routes tickets instantly". ServiceNow ITSM describes "autonomous AI specialists". Ivanti describes agentic AI that works to "detect and remediate issues before end users report them". Zendesk offers "AI monitoring across 100% of interactions".

How do we treat them on this site?

We quote vendor figures in labelled callouts marked as vendor statements, with a link to the page. We never use them as our findings. Our fact grid records what is documented, such as which agents exist, what they do and which plan includes them, not the percentages attached.

What questions does each number raise?

Read every figure for three things:

• The unit. "Incidents", "queries", "tickets" and "conversations" are different denominators. "Up to" sets a ceiling, not an average.
• The scope. Is the figure about IT service management, customer service or the vendor's AI in general? Freshworks' 80% figure is stated in a general service context.
• The conditions. Resolution rates depend on the knowledge base, the request mix and how much the AI is allowed to do. A desk with mostly password resets will see different numbers from one with mostly hardware faults.

What do capability descriptions leave out?

Agent names and verbs such as "resolves" or "remediates" rarely say how a result is checked, what happens when the AI is wrong, or which plan includes the feature. Packaging details are often on the pricing page instead: monday service, for example, puts AI Ticket Triage on every plan but limits it to one board on Standard and three on Pro. Halo (formerly HaloITSM) includes Halo AI in its single price. Read the pricing page alongside the product page.

How do you test a claim yourself?

Turn the claim into a measurement on your own data during a trial:

• Pick the three request types with the most volume and the clearest answers.
• Load the knowledge the AI needs, and nothing more, so the test reflects your real starting point.
• Run a week of real or replayed tickets and count three outcomes: resolved correctly, escalated correctly, and handled wrongly.
• Ask the vendor how its own figure was measured, and compare definitions.

The share handled wrongly matters as much as the share resolved: a confident wrong answer costs more than an escalation.

What should a buyer take from the numbers?

Treat each figure as the vendor's description of what is possible under its own conditions. Use it to decide what to test, not what to expect. Then use your own trial results, and your team's appetite for maintaining knowledge, to answer the AI question in the ITSM fit finder.

What would a better claim look like?

The most useful vendor figures name their denominator, their time period and the kind of requests involved, and link to how they were measured. None of the figures quoted above does all of that on the page where it appears. That is common across the category rather than a fault of one vendor, and it is a fair question to put to any sales team: what exactly was counted, over what period, and for which customers?