Domain 2 · Quantify Business Value
Prove your AI spend is worth it.
AI ROI is the capability that answers the board's question: what did the AI budget return? CloudMonitor measures the efficiency of every token and seat from the data it already ingests, then lets you divide that cost by a business unit you define, so AI value stops being a story and becomes a number.
The problem
Everyone can see the AI invoice. Few can defend it.
98% of FinOps teams now manage AI spend, up from 31% in 2024 (State of FinOps 2026, 1,192 organizations). Yet most teams can show what AI cost and little about what it returned. McKinsey finds only about 6% of organizations attribute more than 5% of operating profit to AI, and fewer than one in five track well-defined AI KPIs.
ROI lives in two systems.
Token and seat cost sits with the provider; the outcome it produced sits in a ticketing tool, a repo, or a CRM. Joining them by hand is a quarterly slide, not a daily report.
Cost per token answers nothing.
A falling cost per token can still mean a rising, wasteful bill. The figure that matters is value per token: how much useful output each dollar bought.
The waste is invisible.
Frontier models on routine jobs, cold caches, un-batched work, and dormant seats are recoverable spend no provider invoice breaks out.
How CloudMonitor answers
Efficiency we measure. Value you define.
The efficiency side comes straight from the token and seat data CloudMonitor already ingests. The value side is yours: wire in the unit your business reports on, and the ratio falls out against a cost group finance already trusts.
Output per dollar.
Rank teams and developers by output tokens per dollar rather than raw spend, so the efficient user leads the board, not the heaviest.
Model-mix savings.
Share of spend by frontier, mid, and entry tier. A job moved off a frontier model with no loss of quality is roughly ten times cheaper.
Cache and batch capture.
Prompt-cache hit rate and batch-tier share, with the dollars left on the table when input could have been cached or run off-peak.
Seat efficiency.
Cost per active developer, plus the paid seats under 10% acceptance or with no activity in 30 days. Clear targets for reclaim.
Bring your denominator.
Feed in the unit you already trust through CSV, REST, or webhook: a resolved ticket, a merged PR, a generated document. CloudMonitor divides AI cost by it daily.
Tied to cost groups.
Every AI ROI figure anchors to a cost group, so engineering sees the same number finance reports upstairs.
The unit-cost model
The token bill is the tip of the iceberg.
A real AI unit cost is three layers deep. CloudMonitor accounts for all three, so the ROI you report is the loaded one, not just the model invoice.
Inference.
The headline cost: input, output, and cached tokens by model and provider. The part every dashboard already shows.
The harness.
Retrieval, vector stores, orchestration retries, and agentic loops that re-call the model. On agent workloads this can rival or exceed the inference line itself.
Total unit cost.
Inference plus harness, allocated to a cost group and divided by your denominator. The number a CFO can actually defend.
Outcomes
A figure leadership can act on.
Per token
Value measured as output per dollar
Your unit
Cost per outcome on a denominator you define
Daily
Same cadence as the bill
AI ROI in CloudMonitor
The cost side of the ratio, ranked and allocated.
Output per dollar
AI cost ranked by what it returns.
Token spend lands in the same analysis as the rest of Azure, grouped by provider and model, so the costly habit stands out before the invoice does and an efficiency ratio is one denominator away.
Recoverable waste
The dollars you can hand back.
Model-mix routing, idle GPUs, and dormant seats surface in the recommendation queue with the savings attached, so improving AI ROI starts with spend you can recover this month.
Related Capabilities
More in the Quantify Business Value Domain.
Unit Economics
Cost per customer, per transaction, per release for the rest of your estate, on the same denominator pattern.
FinOps for AI
The full AI pillar: connectors, model-mix and seat efficiency, and Azure commitment economics for AI.
All Quantify Business Value Capabilities
See every Capability in this Domain side by side.
Source: this page interprets the Quantify Business Value domain published by the FinOps Foundation, licensed under CC BY 4.0. The wording, examples, and product mapping on this page are CloudMonitor’s own.
Turn AI spend into a number you can defend.
The beta connects your AI providers and maps your value denominator, so cost per outcome shows up in minutes.
