“What does each AI workload cost — and what did it return?”
Every token type and price, every way AI spend reaches you, the meter, allocation, unit economics, the agents doing the spending and the value they returned — the FinOps Foundation’s guidance for AI, run in the same book as cloud, software, people, budgets and strategy. Every token is priced at the rate in force on the day it was used and assigned to an owner. Allocate to your own consumers and missions, and set your own bar for proof. When model prices, the framework or the billing standard change, old usage is never re-priced. And FinOps, TBM and AI regulation see the same agents and the same spend.
Runs in
Covers
FinOps for AI · token and model spend
The pressure
AI spend comes through cloud providers, model vendors, SaaS seats and agent platforms, each billed differently. Prices change by model and by month. Input, output and cached tokens are priced apart, so volume and cost tell different stories. Agents run with no budget or register. And ROI is estimated from surveys of time saved — in a different spreadsheet from the people costs it is meant to replace.
What it looks like today
Full coverage
AI spend leaves the vendor dashboards. Each part of FinOps for AI is metered, owned and valued in the same book as cloud, people, the budget and the strategy.
Mapped to FinOps for AI in the FinOps Framework (March 2026)
What a model call actually costs.
Every channel, in one book.
Spend you can trust before you allocate it.
Who the spend serves.
The measures FinOps for AI asks for.
Who spends, and within what limits.
Where AI spend changes decisions.
What changes
Token bills by vendor
Every token priced at what it cost that day, with an owner
One blended price per model
Input, output and cached tokens priced apart, with an effective-dated price history
Agents nobody has a list of
An agent register with budgets, run results and every crossed limit flagged
ROI estimated from time saved
Proven return on evidence-ready spend, with AI and people costs in one chain
Volume is not cost
Input, output and cached tokens are priced differently, so the tokens you use most are rarely the ones you pay most for. Dotwork shows token volume beside cost for every model, team and agent, priced at the rate in force on the day of use, with cache reads counted at their own rate. Beside it sits each agent’s reach and its limits — and any run that crosses a recorded limit is flagged for its owner.
Where other tools stop
LLM observability tools count tokens per trace. Cloud cost tools see one line per vendor. Neither prices the tokens against the budget, the owner and the work they were meant to do.
Flexed to how you run
Every company buys AI differently — through APIs, SaaS seats and agent platforms — and allocates and judges it differently too.
Your consumers and missions
Allocate AI spend to the teams, products, capabilities and agent missions your company already uses — not a vendor’s list of projects.
Every vendor, one meter
Hyperscaler, model vendor, SaaS seat or agent platform: each is priced its own way and lands on the same meter, so the comparison is like for like.
Your bar for proof
Decide how much evidence and confidence a value claim needs before it counts as proven. Claims below the bar are shown, not counted.
Start with the meter
Begin by getting the spend right and owned, then add the agent register, unit economics and value as the program grows.
As AI pricing and guidance move
Token economics has no single standard yet. The FinOps Foundation reshaped where AI sits in its framework this year, FOCUS plans native AI billing columns, and the Linux Foundation has announced a Tokenomics Foundation to set open standards — while model prices change every few months. Whichever moves, the change is proposed against your categories, measures and prices for approval, and past periods keep the prices and shape they were reported under.
AI moved from a framework scope to a technology category, FinOps for AI
AI spend labeled as its own category, in the same book as cloud, SaaS and the data center
Scopes redefined as business segments — an AI initiative can be its own scope, overlapping cloud
AI initiatives become views over the same spend, counted once in the total
Named measures for AI, including cost per inference, cost per API call and token consumption
Read from the meter by model, team and agent
Executive Strategy Alignment added as a capability
AI spend read beside the strategy and the portfolio it serves
Planned for FOCUS 1.5: model identity and input and output tokens as native billing columns
Tokens arrive today through each provider’s own usage fields; when the columns land, the mapping is proposed and history kept
Every framework, one model
AI spend touches FinOps, TBM, asset management, AI regulation and the goals it is meant to move. In most companies each has its own number. Here there is one meter, and every framework reads from it.
Cloud accountability and billing data
AI is a FinOps technology category; token spend sits in the same book as the cloud bill it often arrives on.
Explore FinOps & FOCUSIT spend taxonomy
TBM’s AI solution type is filled from this meter rather than estimated.
Explore TBMEU AI Act · NIST AI RMF · ISO 42001
The agents in the register are the AI systems the EU AI Act, NIST AI RMF and ISO 42001 ask you to inventory.
Explore AI RegulationISO/IEC 19770 · licenses, contracts and renewals
AI seats, licenses and model contracts sit beside the token spend they overlap.
Explore IT Asset ManagementGoals tied to funded work
AI spend traces to the key results it was meant to move, so return is measured, not claimed.
Explore OKRsActivity-based costing · rolling forecasts · Beyond Budgeting
The people cost an agent is compared against is priced from finance’s rate cards.
Explore Cost & Planning MethodsAI-native tokenomics
The meter is checked on every load, every token is priced and assigned, and every agent is watched. Owners approve what changes, and every change is logged. Dot keeps linked spend, observed outcomes and verified financial benefit clearly apart — it never claims a return the evidence does not support.
Finds repeated invocations, calls that produced nothing, usage with no price and rows where tokens times rate does not equal cost.
Prices every token at the rate in force on the day it was used, with an effective-dated history of each model’s input, output and cache rates.
Assigns metered spend to teams, people, applications, capabilities and offerings.
Watches each agent’s spend against budget, run failure rate, cost trend, confidence against threshold and runs above their transaction limit.
Runs every evidence gate on AI and human spend, computes proven return on the spend that clears them and turns every gap into a named to-do.
Compares the people time an agent’s work avoided with what the run and its human review cost — failed runs avoid nothing.
When the framework, FOCUS or a provider’s usage fields change, works out which categories, scopes and measures move and drafts the remaps.
Answers questions about AI spend, agents and value, citing the records behind every figure.
Where it runs in Dotwork
Dotwork does not ask you to adopt a new method. The framework becomes a view over the same model these solutions already read and write.
How it compares
AI cost trackers and LLM observability tools count tokens by trace, team and model. Spreadsheets and BI join them to the rest of the budget by hand. Tokenomics on Dotwork meters, owns and governs the spend, then traces it through the work to the outcomes it moved.
| Capability | AI cost & LLM observability tools | Spreadsheets & BI | Tokenomics on Dotwork |
|---|---|---|---|
| Token spend by model, team and agent | Partly | ||
| Input, output and cached tokens priced apart | Partly | ||
| Priced at the rate in force, with price history | Partly | ||
| Meter integrity checks — duplicates, calls billed for nothing | Partly | ||
| Agent register with budgets and run results | Partly | ||
| AI, cloud, SaaS and people spend in one book | Partly | ||
| Return from measured value, not estimated time saved | Partly | ||
| Every part of FinOps for AI mapped to a home | Partly | ||
| Price, framework and billing changes with past periods kept | |||
| Wired to budgets, the portfolio and strategy |
Who it is for
“What does our AI cost, and what has it provably returned?
“Which agents and use cases are worth scaling, and which should we stop?
“Is the AI meter right, and who owns each dollar?
“Which model gives us the quality we need at the lowest cost per task?
“Which AI contracts and seats are we actually using?
“What does my team’s AI use cost, and what did it deliver?
Before you buy
How to get a number instead of a pitch.
1
Good answer · It stays at the price in force when the tokens were used.
Bad answer · Everything is re-priced at today’s rate.
2
Good answer · Each at its own rate, with volume and cost shown side by side.
Bad answer · One blended rate per model.
3
Good answer · Proven return on the spend that clears every evidence gate, and a to-do for the rest.
Bad answer · An estimate from a time-saved survey.
More in Finance & Cost
IT spend taxonomy
ExploreCloud accountability and billing data
ExploreASC 350-40 · ASU 2025-06 · IAS 38
ExploreISO/IEC 19770 · licenses, contracts and renewals
ExploreActivity-based costing · rolling forecasts · Beyond Budgeting
ExploreFinance & Cost
The standards finance and IT finance use to say what technology costs and who uses it.
The model underneath
Steer, Fund & Optimize and Govern & Deliver all read and write the same Enterprise Operating Model. The first module you license builds the model. Every module after it adds context instead of another silo — and Dotwork AI reads all of it, stopping where the decision is yours.
See the Enterprise Operating Model© 2025-2026 Dotwork, Inc. All rights reserved.