Artificial Intelligence is extremely valuable for the Service Desk. It can significantly reduce the number of tickets requiring human intervention. But this reduction, on its own, is not yet proof of success.
A user can close a conversation with a virtual agent without having resolved anything. They may try again shortly after, open a new ticket, send an email, or call support directly. In the AI dashboard, however, the first interaction might continue to appear as “deflected.”
This is where the ticket deflection rate risks becoming a seductive but incomplete metric… or, worse, a misleading one.
Ticket deflection rate: what are we really measuring?
The ticket deflection rate is generally used to indicate the share of requests that does not require agent intervention thanks to self-service, knowledge base, automation, or AI assistants.
But definitions are not always identical. Some sources speak of deflection only when the request is actually resolved before a ticket is created. Others note that, in market practice, many measurements simply consider conversations concluded without human intervention.
Beyond definitions, the key point to keep in mind is that these are two very different things. If the user finds a procedure, solves the problem, and does not return to ask for assistance, we have achieved a real result. If they abandon the chatbot because they cannot find the answer, we have resolved nothing: we have only prevented, at least temporarily, the work from entering the Service Desk queue. This is why the ticket deflection rate should be treated as an initial efficiency metric, not as standalone proof of resolution.
From deflection to resolution without human intervention
The shift in perspective is simple and unsparing: do not measure the absence of a ticket, but the outcome of the request.
A “resolution without human intervention” should indicate a case in which the user obtained what they needed through AI, automation, or self-service and did not subsequently have to turn to an agent for the same issue.
Be careful, though! The ticket deflection rate should not be eliminated: it should be placed within a more complete context. Below, we frame it through 5 decisive metrics.
1. Effective containment
The first metric to consider is effective containment: the percentage of requests handled entirely through AI or self-service that are genuinely concluded without subsequent human intervention.
A terminated conversation should not automatically be considered contained. It must be verified that it was not followed, within a defined time window, by a reopening, a new ticket on the same topic, or a switch to another channel.
2. Re-contact on the same issue within a defined window
Let us return to what we have just brought to light: a request that appears resolved today but reappears tomorrow is not a true success of automation.
The re-contact rate measures how many cases classified as resolved by AI generate, within a defined time window, a new request or new contact on the same intent. When a previous ticket exists, reopening can also be included.
3. Duplicate ticket rate
Work can return to the Service Desk in a less obvious form: duplicates. A user who does not receive an answer may subsequently open a ticket or, if they had already opened one, create another on the same issue using different words.
Recognising these duplicates requires intent correlation, user identity, the service involved, and temporal proximity.
N.B: Similar reports coming from different users should instead be correlated to verify whether they indicate a common incident or a broader problem.
4. Work shifted to other channels
This is one of the most dangerous blind spots. The user abandons the virtual agent and calls. Or switches to email, Teams, or the portal. From the perspective of the individual channel, the ticket has been “deflected.” From the organisation’s perspective, the work still exists. Sometimes it has even increased, because the user must explain the problem again.
This is why measurement must always be omnichannel.
5. Correct and timely escalations
AI should not be measured solely on its ability to avoid escalations. In some cases, a good escalation is exactly the result we want.
When the confidence level is insufficient, information is missing, or the request concerns a sensitive activity, the AI should pass the case to an agent before producing a potentially incorrect response.
Confidence thresholds must therefore be calibrated by risk and request type, even though they should not automatically be interpreted as a probability of correctness. The useful metric is not “as few escalations as possible,” but “the right escalation at the right time.” In short, we are talking about a balance between automation and human control, which becomes even more crucial with agentic AI.
Measuring the value of AI with connected data
At this point it is clear that, to seriously evaluate a vendor or an internal project, the ticket deflection rate should never be read in isolation. It must be put in relation to other metrics: effective containment, the close recurrence of the same issues, the number of duplicate tickets, the shift to other channels, the quality of escalations and not just their quantity. In other words, it is not enough to measure how many requests do not reach the Service Desk. It is necessary to understand what happens to those requests after the interaction with AI.
There are no universal benchmarks valid for all organisations. Every Service Desk has different volumes, complexity levels, request mixes, channels, and risk profiles. For this reason, the most useful approach is to start from one’s own baseline and compare homogeneous results by service, intent, channel, and risk level.
It is equally important to ask the vendor how a “resolved” request is defined, what conditions trigger this classification, and whether the data can be independently verified.
There is also a further decisive point: data must not remain locked inside the chatbot. If self-service interactions are analysed as a separate world, the risk is that of building apparently positive metrics that are disconnected from users’ actual behaviour.
To obtain a reliable scorecard, interactions must be able to be correlated with tickets, users, channels, workflows, and subsequent outcomes. Only in this way does it become possible to distinguish between a contact that has been genuinely absorbed and one that has simply disappeared from the AI dashboard only to re-emerge shortly after in another form.
And it is precisely this ability to connect data that transforms a surface-level metric into a concrete measure of the value produced.
How EasyVista helps read AI outcomes in the Service Desk context
The goal is not to collect as much data as possible, but to create continuity between what happens in self-service, what enters the Service Desk, and what subsequently occurs in the workflow. This is where an integrated ITSM platform becomes decisive.
In this sense, the EasyVista ecosystem allows working on multiple complementary levels.
- EV Service Manager centralises incidents, requests, workflows, and operational data within the same environment. This makes it simpler to observe the complete lifecycle of a request and verify, for example, whether an interaction apparently resolved by AI subsequently generates a new ticket, an escalation, or a follow-up activity.
- EV Self Help works on the self-service and knowledge side, combining content, conversational support, and usage analytics. Its value, in this context, lies in the ability to read not only how many interactions are intercepted, but also how users engage with content and at which points the self-service journey works or breaks down.
- To this is added EV Pulse AI, which brings artificial intelligence capabilities directly into service management workflows.
The combination of these tools therefore makes it possible to shift attention from the simple deflection figure towards a more complete view: which part of the work is effectively eliminated, which is shifted, which re-emerges, and where human intervention continues to be necessary.
Conclusions
The ticket deflection rate remains a useful indicator: it shows how much demand did not immediately require the work of an agent. But it does not say, on its own, whether that demand was resolved.
To understand this, one must look at what happens afterwards: re-contacts, reopenings, duplicates, channel switches, and escalations. It is there that the difference emerges between an AI that reduces the visible number of tickets and an AI that truly reduces the Service Desk’s workload.
For CIOs and IT managers, this distinction is also the key to measuring the real benefit of AI and building a credible business case. ROI then emerges by comparing these benefits with the costs required to generate them.
Ultimately, the best metric is the one that withstands the simplest question: was the user’s problem truly resolved, without someone else having to deal with it afterwards?
FAQ
What is the ticket deflection rate?
The ticket deflection rate measures the share of requests that does not require the direct intervention of an agent thanks to self-service, automation, knowledge base, or AI.
What is the difference between ticket deflection and resolution without human intervention?
Deflection indicates that the request did not reach an agent. Resolution without human intervention requires something more: the problem must be genuinely resolved and must not re-emerge through reopenings, duplicates, or other channels within a defined time window.
What metrics should be used to evaluate AI in the Service Desk?
In addition to the ticket deflection rate, it is useful to monitor effective containment, re-contacts on the same issue within a defined window, duplicate tickets, the shift of work to other channels, and the quality of escalations based on confidence thresholds.
A guide to AI in ITSM
Discover how to integrate artificial intelligence into your ITSM, redesign your processes, and take your company’s efficiency to the next level.
A guide to AI in ITSM
Discover how to integrate artificial intelligence into your ITSM, redesign your processes, and take your company’s efficiency to the next level.