Measuring the Value of AI in ITSM: A Guide for CTOs and CIOs

The scenario: in most large organizations, the first pilot project applying artificial intelligence to IT service management has already been completed. Results have been presented, the technology worked, and the team gathered support. Then, six or nine months later, comes the question that changes the tone of the conversation: how much did we spend and what did we get in return? This time it is the finance department asking. The question is legitimate and increasingly urgent.

Forrester’s forecasts for 2026 describe a year in which AI loses its shine and companies move to auditing the books: only 15% of decision-makers report an improvement in EBITDA attributable to AI over the past twelve months, and fewer than one in three can link the value of AI to changes in the income statement. Against this backdrop, a quarter of planned spending will be deferred to 2027 and CFOs will be systematically involved in purchasing decisions. On the technology leadership front, Forrester estimates that two-thirds of CIOs will need to justify their budget by linking technology spending to business value.

The crux of the matter, almost always, concerns the ability to measure spending on AI applied to IT service management — that is, AI ITSM spending — than with the quality of the technology adopted.

Why Is AI ITSM Spending Under Scrutiny?

Adoption itself is not the problem. According to the EasyVista report The Reality of ITSM in 2026, which gathers responses from 1,100 IT and security professionals, just 5% of organizations report not using AI in ITSM at all, while 37% rank workflow automation among their top three operational priorities. The same study finds, however, that satisfaction responses cluster around the option “met expectations” — a signal that AI is stabilizing operations rather than transforming them.

This is precisely where the difficulty of justifying AI ITSM spending to the CFO arises. An AI that makes daily activities more regular and predictable produces real benefits, but these are distributed across dozens of micro-operations and do not surface as a standalone budget line item. A recent MIT study dedicated to the generative AI gap found that 95% of enterprise AI pilot initiatives showed no measurable financial return — a figure that speaks as much to insufficient measurement as to missing results.

A second factor concerns starting conditions. A pilot project lives in a controlled environment, with a narrow application scope, limited integrations, and a dedicated team monitoring every phase of the project. These are precisely the conditions — advantageous but in some ways illusory — that disappear the moment the project moves into production at real scale, with real users and complete data.

The value observed in the lab does not automatically transfer to the broader environment, and this difference must be anticipated in the measurement model.

Value Comes from Use Cases, Data, and Integration

AI initiatives in ITSM that produce actual economic results share three characteristics: precisely defined use cases, a solid data foundation, and seamless integration into existing workflows. AI generates value when it supports everyday operational decisions and activities. When it is introduced as an end in itself, the return remains theoretical.

Data quality deserves particular attention, as it is the prerequisite for any implementation. An effective data management plan requires continuous maintenance: it is worth correcting errors in the data the system will use right away (duplicates, empty fields, incorrectly assigned categories, obsolete records), starting with the datasets most relevant to the chosen objective. Simple, shared rules are also needed — for example, permitted value ranges for numeric fields and uniform names for categories — so that the system recognizes information consistently. These are rules that must be maintained over time, because data changes along with the organization.

The ROI calculation is compromised from the start if there is no historical data with which to build a credible comparison. Gartner describes this phenomenon as a genuine AI value crisis, fueled by the difficulty of demonstrating profitable returns on initiatives launched with high expectations.

The Four Metrics That Make AI ITSM Spending Measurable

A useful framework for dialogue with the finance function rests on four indicators, all already present in service management systems.

Deflection Rate

Measures the share of requests resolved without operator intervention, through a self-service portal, knowledge base, or conversational assistants. Each genuinely deflected contact can generate an avoided cost, net of the costs of the automated channel and any re-contacts. It should always be read alongside the reopening rate, because a deflected ticket and a merely deferred ticket have different impacts on end-user satisfaction.

MTTR (Mean Time to Repair/Resolve)

Average resolution time has a direct impact on two aspects that are easily recognizable even outside IT: the working hours lost by users during an outage and the continuity of customer-facing services. It is an indicator that should be broken down by phase — detection, assignment, diagnosis, resolution — because AI acts on specific segments and the aggregate figure dilutes its effect. Alongside MTTR, it is worth tracking the reduction in alert noise and ticket routing accuracy.

Operator Productivity Tickets handled per operator, average handling time, and share of time devoted to value-added activities. This indicator captures the most common benefit of automation: the ability to absorb growing volumes without increasing headcount.

Cost per Ticket

This is the metric that closes the loop. Industry benchmarks show significant differences across support levels, and resolving a request at the first level costs a fraction of what the same resolution costs after an escalation. Every shift downward in the support pyramid generates an immediately calculable saving.

Defining the Baseline Before Switching the System On

Starting values must be recorded before the system is activated. A baseline reconstructed after the fact is based on rough estimates and data retrieved in a fragmented manner, and it is easily challenged during a review. Setting it in advance — at least three months before activation — is what makes the subsequent comparison credible.

For each metric, three elements are needed. The first is the value recorded before activation. The second is the target set at six and twelve months. The third is the economic rationale linking the two figures — that is, which mechanism the improvement should come from and how much it would be worth in monetary terms. Vague expressions such as “leaner processes” or “better service quality” do not survive scrutiny by management control, because they indicate neither the source of the saving nor its amount.

A gradual and controlled rollout — a phased deployment — also aids measurement. During the observation phase, the system generates recommendations without intervening in the process, which allows these suggestions to be compared with the actual decisions made by operators. The subsequent phase, within a controlled scope, produces the first defensible numbers.

Monitoring Over Time with a Defined Cadence

Monitoring works when it becomes an organizational routine. A monthly review of the four indicators, with a formal checkpoint — for example, at ninety days — allows course corrections while they are still possible. Alongside results, the actual cost of AI ITSM spending must be tracked, including licenses, token consumption, integration hours, training, and governance. In the EasyVista report The Reality of ITSM in 2026, usage costs appear among the main barriers to AI adoption, cited by 13% of respondents. The concern is well-founded: return on investment is obtained by comparing benefits against expenditure incurred. Without precise cost tracking — meaning if the denominator remains unknown — the calculation remains incomplete even when operational results are positive.

Translating Operational Data into Financial Language

The decisive step consists of converting operational metrics into figures that the CFO habitually uses. Few organizations manage this. According to Forrester, roughly one in three companies today can link their technology services to the business activities they support. To bridge this gap, Forrester recommends using IT finance tools, including technology business management, which associates overall technology spending with the business areas that benefit from it, and FinOps practices, which attribute variable costs such as those of the cloud.

The Four Cost Items to Present to the CFO

In the case of ITSM, the four metrics described above convert into an equal number of financial line items.

Avoided cost, obtained by multiplying deflected tickets by the unit cost of handling, net of the costs of the automated channel.

Freed capacity: these hours can be valued using the company’s hourly operator cost, including salary, contributions, and indirect costs, but they do not automatically correspond to actual savings. They become a direct economic benefit when they make it possible, for example, reduce overtime, reliance on outside vendors, or hiring needs.

Recovered business-side productivity, calculated on the reduction in MTTR multiplied by the users affected.

Avoided risk, quantified on missed SLA breaches and prevented incidents.

These items must be accompanied by the total expenditure for the period, with a realistic amortization horizon. An investment that breaks even at eighteen months with solid evidence is more defensible than a promise of immediate return with no supporting data.

Measurement and the Project Must Be Designed Together

Defining the parameters is an activity that must be built into the project from its very inception. Organizations that, before activating AI, establish starting values, define precisely which services and ticket categories are included in the calculation, and agree in advance with the finance function on how results will be converted into monetary terms, arrive at the review with verifiable data and secure approval to continue. In a year in which a quarter of AI spending risks being deferred and CIOs are asked to justify every budget line in terms of business outcomes, being able to demonstrate the return on AI ITSM spending becomes the condition for continuing to invest.

FAQ

1. How long does it take before the return on an AI project in ITSM can be measured?
   It is advisable to have a sufficiently representative baseline period, often in the order of three months and longer where seasonality is present.
   
2. Which metrics does the CFO actually care about?
   Avoided cost, freed capacity expressed in person-hours, recovered business-side productivity, and avoided risk. These are the four items that translate deflection, MTTR, operator productivity, and cost per ticket into financial figures.
   
3. How can you avoid crediting AI with improvements generated by other initiatives?
   By defining a clear scope of services, ticket categories, and teams included in the measurement, and by using the observation phase of the rollout to compare the system’s recommendations with the decisions made by operators.
   
4. Why do many successful pilot projects fail to generate value in production?
   Because the pilot operates in a controlled environment, with limited data, few integrations, and constant supervision. In production, volumes, case variety, and complexity all grow, and without a measurement model designed for real scale, the benefits remain invisible.