AI unit economics is the cloud cost of producing one unit of AI value: one resolved ticket, one generated document, one active user served for a month. It turns an undifferentiated AI bill into a per outcome number the board can hold against the revenue or saving that outcome creates. This matters because a rising AI total can be entirely healthy when cost per outcome is falling and volume is climbing, and a flat total can hide waste. Give the board the unit metric and it can tell good growth from runaway spend, which a headline number never reveals.
Here is how to define the unit, instrument the data, and present AI cost in terms a board acts on rather than worries about.
Why a total AI bill is the wrong board metric
Total spend answers one question, how big is the bill, and that is rarely the question that matters. If the AI assistant resolved twice the volume of support tickets this quarter, a higher bill is a sign of adoption, not waste. If volume was flat and the bill doubled, something is wrong. The total cannot distinguish the two cases, so it sends the board into either false alarm or false comfort.
Unit economics fixes this by dividing cost by the outcome it produced. The board then sees the trend that matters: is each outcome getting cheaper to serve, and is the value per outcome above its cost. That is a margin conversation, which boards are built to have.
How do you define the AI unit?
Pick the outcome closest to value the business already measures. The right unit is specific to the product.
- Cost per resolved interaction. For an AI support assistant, the cloud cost of fully resolving one ticket, set against the human cost it displaces.
- Cost per generated artefact. For a content or code feature, the cost to produce one document, summary, or pull request.
- Cost per active user per month. For an embedded copilot, the cost to serve one engaged user, set against that user's subscription contribution.
Each unit pairs a cost with a value the business understands, which is what makes it a board metric rather than an engineering one.
How do you instrument it?
The cost side comes from tagging AI workloads so billing attributes every token, GPU hour, and provisioned throughput charge to the feature that incurred it. On AWS that means the Cost and Usage Report carries the AI tags; the equivalent discipline applies to Azure OpenAI usage and to GCP. The outcome side comes from the product analytics that already count tickets, documents, or active users. Join the two and you have cost per outcome over time.
Watch the quiet drivers that move the unit cost without anyone deciding to: retrieval pipelines that re embed unchanged documents, agents that loop more than expected, and a default to a frontier model where a smaller one would serve. These inflate cost per outcome silently, which is why the metric needs a regular review cadence.
A worked example
A European SaaS company reported AI spend to its board as a single rising line, and the board kept asking whether to cap it. Reframing the spend as cost per resolved support interaction changed the conversation entirely. The unit cost was falling quarter on quarter as the team routed simple intents to a smaller model and cached repeated lookups, while volume grew as adoption spread. The board could see each resolved ticket cost a fraction of the human alternative and was getting cheaper, so the question shifted from whether to cap spend to how fast to expand the feature. Figures are verified against billing data and anonymised.
Frequently asked questions
What is AI unit economics?
Why is total AI spend the wrong board metric?
How do you lower cost per AI outcome?
Put board ready AI metrics in place with us
We help enterprises instrument AI cost per outcome across AWS, Azure, GCP, and OCI, so the board governs AI as a margin question rather than a spending fear. Our guarantee: we reduce your cloud spend or we reimburse our service fee, on either a Fixed Fee or a no risk Gainshare basis. We send more of this thinking through The Cloud Spend Navigator.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.