Measuring the ROI of AI in the IT service desk.

By COGScontrol Team · July 8, 2026

To measure AI service-desk ROI, compute the fully loaded cost per AI-resolved ticket and compare it to the loaded cost of an agent-resolved ticket — about $22 at level 1 on industry benchmarks. The denominator is genuinely resolved tickets, not deflected ones, because a deflection that re-opens or escalates can cost more than the agent ticket it avoided. The service desk is the cleanest function to measure precisely because its human baseline is so well established.

That established baseline is the advantage. Where most internal functions struggle to price the work AI displaces, IT support already knows it. MetricNet's benchmarking puts the average level-1 cost per ticket at about $22, with an escalated ticket reaching roughly $84 once desktop support is involved; HDI/MetricNet's 2021 benchmarks show per-ticket costs from roughly $6 to $40 across agent-assisted channels. Those figures date to 2021 — MetricNet publishes updated benchmarks annually, and the structure has held (level 1 in the low $20s, escalation running 3–4× higher) — so validate the current ranges against your own desk's actuals before a business case rests on them. The AI just has to beat that baseline — fully loaded, at equal quality.

What is the cost-per-ticket formula, on a known baseline?

Two terms decide whether the result is real. Fully loaded AI cost is not just the per-conversation model fee — it includes the ITSM platform's AI add-on, the knowledge retrieval layer, and the human handling of every escalation the AI generates. The model bill alone understates it: ServiceNow's CFO says AI reasoning is less than 10% of cost to serve — the other 90% sits in orchestration, governance and context. Genuinely resolved tickets are the ones the employee did not re-open and that did not escalate. The unit-cost mechanics are the same as in AI unit economics.

Why can deflection cost more than the ticket it avoided?

The escalation math makes the deflection trap concrete. A level-1 agent ticket costs about $22. An escalated ticket costs about $84. So a bot that “deflects” a ticket the employee then re-opens and that escalates to desktop support has not saved $22 — it has spent its own AI cost and driven an $84 ticket, for a worse outcome than sending the employee straight to level 1. Counting that as a deflection win is how service-desk AI ROI gets overstated. Resolutions are the denominator; escalation rate and re-open rate are the guardrails you watch beside them.

How do you map service-desk metrics to the right denominator?

What does beating the baseline look like in practice?

The numbers below are illustrative — a hypothetical company, not customer data. Call it Northgate: a 4,000-employee firm whose service desk logs 6,000 tickets a month, with an AI assistant whose fully loaded cost — the ITSM platform's AI add-on, model usage, knowledge retrieval, and the human handling of escalations it generates — is $12,000 a month.

The ITSM dashboard credits the bot with 2,400 tickets: a 40% deflection rate. Measured honestly, they split three ways: 1,800 genuinely resolved (no re-open, no escalation), 400 re-opened back to level 1, and 200 that escalated to desktop support.

• Cost per resolved ticket. $12,000 ÷ 1,800 = $6.67 — less than a third of the $22 level-1 baseline. The deflection version, $12,000 ÷ 2,400 = $5.00, flatters the number by 25%.
• ROI. The AI displaced 1,800 level-1 tickets at $22: ($39,600 − $12,000) ÷ $12,000 = 2.3×.
• The guardrail line. The 200 escalations cost about $84 each — $16,800 of desktop-support work, versus $4,400 had those tickets gone straight to level 1. That $12,400 of created cost never appears on a deflection dashboard, and it names exactly which ticket categories the bot should stop attempting.

The honest math still clears the baseline comfortably — that is the point of measuring. The defensible number survives review, and the escalation line turns a vague “the bot struggles sometimes” into a routing decision with a dollar figure attached.

Why do your ITSM and your cost tool each see only half?

Your ITSM platform reports ticket volume, deflection and escalation but not the fully loaded cost of the AI behind each resolution; your cloud-cost or FinOps tool reports the AI spend but has no notion of a resolved ticket to divide it by. Each holds one half of the ratio, and a manual join breaks as soon as a second model or channel ships — the spend-versus-value boundary covered in FinOps vs AI Value Management and across the categories of AI ROI tooling.

COGScontrol joins them. It attributes the fully loaded cost of the service-desk AI — tokens, platform fees, retrieval, escalation handling — into one ledger with an audit trail, divides by the resolved tickets your ITSM already records, holds escalation and re-open rates in the same view, and reconciles cost to provider invoices every 24 hours. It is the operating loop from how to measure AI ROI and internal AI ROI, pointed at the service desk. When the CIO asks whether the AI beat the $22 baseline, you answer with a fully loaded cost per resolved ticket and a trend, not a deflection rate. See the features or start free; pricing is a fixed subscription, never a percentage of your AI spend.