Skip to content

As of

How to measure the ROI of AI in IT service management

Measuring the ROI of AI in service management means taking a baseline before the AI agent starts, then counting six lines, including one that vendor calculators leave out.

Every service management vendor now ships an AI agent, and most ship an ROI calculator with it. You enter your ticket volume and a cost per ticket. The calculator multiplies the tickets the AI will handle by that cost, and a large number appears in the corner of the slide.

Then the slide meets a finance team. The usual first question is where that money shows up, and a ticket count can't answer it.

To be fair to the calculator, ticket volume is real data. It's easy to pull, everyone understands it, and a falling count does say something. The trouble is what it leaves out: whether anyone waited less, what your team did with the time, and what the platform costs to keep running. A defensible case measures those against a baseline you took before you started.

Why ticket counts make a weak ROI case for an AI agent

Picture a team that added an AI agent to the service desk in spring and is asked in the autumn to show the return before renewal. The calculator produces a figure the size of a third of the IT budget. Finance asks how many of those requests would have become tickets at all, where the analyst hours went, and what the platform cost to run beyond the license. The agent may have done well, and the team still has no good answer to any of the three.

A ticket the AI handled is a counterfactual. It assumes every request would otherwise have reached an analyst and cost the full average. Some would have. Others were the third message about the same outage, or a question someone would have answered in a channel.

The cost per ticket is shaky too. It's usually a published average, not your team's number, and it bundles the easy password reset with the week-long access investigation.

And the count says nothing about the people the service desk exists for. An employee who waits two days for Salesforce access doesn't care how the request was categorized. A finance team asking about return wants to see time and money that moved, on a record it can check.

What should an ROI case for AI in service management measure?

It should measure six lines, each against a baseline taken before the AI agent goes live:

  • Employee wait time: how long people wait from asking to having what they asked for, by request type.
  • Analyst hours moved to project work: the time your team spent on routine requests that now goes to the backlog.
  • License savings: seats reclaimed and renewals reduced, from usage evidence.
  • Audit preparation time: the hours it takes to assemble evidence for an access review or SOC 2 audit.
  • Tools consolidated: contracts you retire at renewal because something else now does the job.
  • Upkeep: the hours spent keeping the platform's copy of your organization current.

Wait time comes first because it's the line employees feel. Measure it from the request's first message to the moment the access was granted or the device fixed, not to the moment the ticket closed.

Analyst hours need care in how they're described. The point is what the same team now does with the time: the device refresh that kept slipping, the identity provider cleanup, the security project nobody had a week for. Count the hours and name where they went. A finance team trusts "we shipped the laptop refresh a quarter early" more than an hourly rate multiplied out.

License savings and tools consolidated are the easiest lines to defend, because they end in an invoice that got smaller. Audit preparation is the one teams forget to time, so time the next one before anything changes.

Take the baseline before the AI agent starts

Most of what you need already exists. Your service desk timestamps every request. Your last audit has a calendar of the weeks people spent pulling evidence. Finance has the renewal invoices and the partner's statements of work.

The part that doesn't exist yet is analyst time. Ask your team to log a few typical weeks by category, roughly, in half-hour blocks. Precision matters less than doing it before the change, because afterwards everyone remembers the old way as slower than it was.

Write the baseline down, date it, and give finance a copy. A number that finance held before the project started is worth more than any number you produce after.

The upkeep line nobody counts

Upkeep is the time spent telling the platform what changed in the organization. A team moves, so assignment groups change. A manager leaves, so approval chains in a dozen forms point at the wrong person. A new application arrives, so someone adds it to the catalog and the configuration database.

After a reorganization, a platform admin can spend most of a month rebuilding assignment groups, approval chains and the configuration records behind them, while the partner handling the larger changes invoices separately. None of that reaches a calculator that only counts what the agent did.

It rarely appears in an ROI case, and the reason is structural. On most service management platforms, that work is what implementation partners bill for, year after year, each time the organization changes shape. A vendor whose ecosystem runs on that work has little reason to ask you to total it.

To count it, list the last year's reorganizations, acquisitions and new applications, and ask your admins and your partner what each one cost in hours and invoices. Then do the arithmetic with your own figures: upkeep hours in a year, times your loaded hourly cost, plus partner fees. That total belongs on the cost side of the case for whatever you run today.

How STLabs changes the lines an AI agent's ROI rests on

STLabs builds the facts an agent works from with Axiom™, a live model of your people, devices, apps and access. Axiom™ reads from your HR system, identity provider, device management and applications, and keeps itself current from them. When your organization reorganizes, the change arrives through the integration. There's no configuration database to update by hand, so most of the upkeep line goes away.

Approval rules live in Policies & Approvals, written once and applied to every request they cover, so a manager leaving doesn't mean editing a dozen forms. Workflows are described in plain language and versioned.

The other lines become things you can read off the record. STLabs logs every AI decision, workflow run and configuration change, and the Analytics view tracks workflow runs quarter over quarter and who has been waiting longest. Access review reports (Growth plan and up) shorten audit preparation. SaaS Spend Management (Scale plan) finds unused licenses, reclaims inactive seats and flags duplicate apps, which feeds both the license and consolidation lines.

The cost side is simpler too. STLabs is priced per employee per year, with no meter on AI resolutions and implementation included, so the price doesn't rise as the agent does more. Teams go from integration to results in five days, which is what an AI-native platform with no records to populate makes possible; the guide to how long a service management implementation takes explains the difference.

How to test this on your own systems

Run the comparison on your own numbers rather than anyone's calculator.

Take the baseline for all six lines first: wait times by request type, a few weeks of analyst time logs, last year's license and partner invoices, the hours your last audit took, and a year of upkeep. Then run STLabs alongside the service desk you already run, starting with one high-volume request type such as software access. Ask each vendor what it charges while you run both.

After a quarter, measure the same six lines the same way. Hand finance both columns and let them check the arithmetic. If a line didn't move, the case should say so.

Close

A case built this way has six lines, a dated baseline and no cost-per-ticket figure. Finance can check every number in it against records they already hold.

The return on an AI agent is whatever your baseline can prove.

Frequently asked questions

How do you measure the ROI of AI in ITSM?

Take a baseline before the AI agent starts, then measure employee wait time, analyst hours moved to project work, license savings, audit preparation time, tools consolidated and platform upkeep. Compare the same lines after a quarter.

Are tickets handled by AI a good ITSM ROI metric?

They're a weak one on their own. A ticket count assumes every request would have reached an analyst at an average cost, and it ignores wait time, analyst hours and what the platform costs to run.

What hidden costs belong in an ITSM ROI case?

The hours spent keeping the platform's copy of your organization current: assignment groups, approval chains and configuration records after every reorganization, plus the partner fees for that work. Total them from last year's changes.

How soon can AI in ITSM show a return?

Lines like wait time and analyst hours can be measured within a quarter of going live. STLabs teams go from integration to results in five days. To see it against your baseline, book a demo.

See it on your tools.