Control AI spend

Asato shows what an organization's AI seats, credits and tokens cost, who is spending them and where the bill is heading, then puts the budget the customer approves into the spend limits its AI vendors already have, or into the customer's own AI gateway. This is the AI Spend Ledger. It proposes a named owner for each AI seat, key and agent. The aim is an owner, a budget and a limit for every seat, credit and token. It never sits in the path of AI traffic, it never collects prompts, responses or files, and it counts a saving only when the vendor's own data shows it.

When to use it

Not a fit

The job in one line

Know the AI bill before it arrives.

Line under the headline on the visual page: "Give every seat, credit and token an owner, a budget and a limit." That line is the aim, not a count. In the example on this page 94% of the spend has a named owner, spend with no owner stays listed with a reason, and a limit is written only where a vendor lets an administrator set one through an API.

In one week, on read-only access, the customer holds its Asset Clarity Report.

The AI tokenomics problem: budgeted by the seat, billed by the token

IDC and the Linux Foundation call the economics of this bill tokenomics, or AI tokenomics. The page that defines the term is https://www.asato.ai/ai-tokenomics, and the visual page links to it from this section ("What is AI tokenomics?"). "Billed by the token" is the market's shorthand: the metered part of a bill is counted in tokens, credits or requests, each vendor in a unit of its own. So a whole AI bill carries seats, credits and tokens, and Asato reads all three into one ledger.

An AI bill has two parts. The seat part is a count times a price, and it is known on the first day of the month. The metered part is usage times a rate. It runs all month, each vendor measures it in its own unit, and its total is known only when the bill arrives.

Each vendor runs its own meter:

A budget set on seats assumes the rate and the usage stay still, and both move:

The bill can move either way. The same Ramp index recorded that AI spend per employee at the heaviest-spending 1% of firms fell 9.7% in August 2026. A seat count predicts neither direction.

Three questions to answer before the bill arrives

An AI bill is one formula: AI spend is the sum, over all vendors, of seats times price plus usage times rate, and it has to stay at or under the budget. Each part of the formula is a question, and Asato answers all three from the customer's own records.

  1. What does it cost? (the sum over all vendors) Asato reads each vendor's seats, credits and tokens into one ledger, and closes the month against the invoices. Each line says how its amount was obtained: read from the vendor, priced from units at the contract rate, or split from a seat fee. Each line also says how old the vendor's figures are. A source Asato cannot read is listed as a gap, never as zero.
    • Link: each vendor's admin feed, the billing exports and the gateway logs are read into one record.
    • Think: each line is priced and says how its amount was obtained, and a source Asato cannot read is listed as a gap.
    • Execute: the month is closed against the invoices, with each budget owner.
  2. Who is spending it? (the seats and the usage) Asato finds each seat, API key, workspace and agent in the vendors' admin feeds and ties it to a person, a cost center and a contract. It proposes an owner and asks the right person to accept. Each line says how its owner was chosen. Spend with no owner stays listed, with a reason. Seats nobody has used are priced.
    • Link: each seat, key and agent is matched to a person, a cost center and a contract.
    • Think: each one gets a proposed owner, the line says how the owner was chosen, and seats nobody used are priced.
    • Execute: the owner accepts it or passes it on, and spend with no owner stays listed, with a reason.
  3. Where is it heading? (the budget) Asato forecasts spend against the budget and flags each line that is heading over to its owner. Vendors already have spend limits, and Asato sets them from one policy, where the vendor allows it. The customer sets one budget policy. After a named person approves, Asato writes the limit into the vendor's own control, or into the customer's gateway, and reads it back. The vendor or the gateway does the blocking.
    • Link: each budget is tied to its owner, its contract and the vendor control that can hold it.
    • Think: spend is forecast against the budget, and each line heading over is flagged to its owner.
    • Execute: a named person approves, then Asato writes the limit into the vendor's own control and reads it back.

Nothing changes until a named person approves it. A policy the customer sets can stand in as the approver for small raises. Requests for more usage from each vendor land in one queue. A small raise inside a cap the customer sets is approved by policy, and a larger one goes to the manager with the team's budget attached. A raise has an end date, and Asato puts the limit back on that date and checks it.

The three steps under each question are the same three the platform runs everywhere. Link: the Asset Knowledge Graph ties each seat, key and agent to a person, a cost center and a contract. Think: the Data Intelligence Engine prices each line, proposes its owner and forecasts it against the budget. Execute: agents set the limits a named person approves, and read each one back. The platform page has the detail.

Where the budget is enforced

It depends on what each vendor lets an administrator change through an API.

Control pointWhat Asato doesActs within
Claude EnterpriseWrites each member's monthly limit. Reads and approves requests for more usageAbout 5 minutes
CursorWrites each user's spend limit and reads back the one that appliesHourly
GitHub CopilotWrites budgets. A user budget can end on its own dateAbout 2 days
Amazon BedrockWrites an AWS Budgets action that blocks the role8 to 12 hours
The customer's gateway (LiteLLM)Writes a dollar budget per key or agent. Can block one keyThe next request
Azure API ManagementWrites a token limitReal time
OpenAI API projectsAlert only, with the step to take in the OpenAI dashboardHourly
Anthropic API workspacesAlert only, with the step to take in the Anthropic ConsoleAbout 5 minutes
Microsoft 365 CopilotAlert only. Credit caps are set by group in the admin centerDaily
Azure OpenAI, Vertex AIAlert only. Cloud budgets warn but do not stop spendUp to a day

Where a limit can only be set by hand, the owner gets the exact step, and Asato checks that it worked.

The vendor or the gateway does the blocking. Asato is never in the request path: if Asato is down, every AI call still works and every limit already written stays in force.

How savings are counted

Savings are counted as found, approved and saved. Found is what Asato identified and priced. Approved is what a named person signed off. Saved is what the customer and Asato agree jointly at the close of the engagement, on the evidence of the vendor's own data, which Asato reads again. Only saved is reported. A reclaimed seat counts when the vendor's roster no longer lists it.

Example data on the page

The product screens on the visual page show example data, not a customer result. Two months of one example company are used throughout: last month, which is closed, and this month on day 19 of 30.

Last month, closed:

This month, on day 19 of 30:

Figures on the page, with their captions

Each figure on the visual page has one caption. On the page a caption takes no room: it shows when a reader points at the figure, or taps it on a touch screen. The captions are given in full here. The label in capitals says whose data it is. The figures build up as a reader reaches them and come to rest on the numbers given in this file.

  1. Your AI bill, as a formula (beside the headline). EXAMPLE DATA. An example month across six AI vendors. On day 19 the forecast is $190,000 against a $170,000 budget. With limits in force, the forecast is $170,000, 100% of budget.
  2. The quote under the headline. Intel's Chief Technology Officer, on IT spend, AI and Asato. Pushkar Ranade, Chief Technology Officer, Intel Corporation.
  3. Last month, seat and meter. EXAMPLE DATA. Each column is a day of last month. Spend passed the $170,000 budget on day 26 and closed $20,000 over.
  4. What an AI bill is made of. ASATO'S METHOD. An AI bill is one formula. Each of its three parts is a question Asato answers from your own records.
  5. AI Spend Ledger, last month, as it closed. EXAMPLE DATA. In this example month, six vendors are read into one total. Each line says how its amount was obtained and how old the vendor's figures are. Personal accounts are listed as a gap, never as zero.
  6. Spend by owner, last month. EXAMPLE DATA. One OpenAI API key spent $25,000 a month with no owner. Once the Search platform lead accepts it, 94% of spend has a named owner.
  7. Forecast against budget, this month, day 19 of 30. EXAMPLE DATA. On day 19 this example month is heading to $190,000, 112% of budget. Once a named person approves, limits written to three vendors and one step taken by an owner bring the forecast back to $170,000.
  8. One request for more usage. EXAMPLE DATA. One request, as its approver sees it. The raise ends at month end, and Asato puts the limit back.
  9. The quote under the approval example. CDW's Chief Technology Officer, in his own words. Sanjay Sood, SVP and Chief Technology Officer, CDW.

Measured by others

Each figure below is from a named outside source, not from Asato.

What customers say

Intel's Chief Technology Officer, on IT spend, AI and Asato. Pushkar Ranade, Chief Technology Officer, Intel Corporation: "Enterprise IT spend is scaling faster than most organizations can manage, and AI is compounding that complexity. Asato's IT Asset Optimization Platform turns fragmented, low-trust asset data into actionable intelligence, giving IT leaders the visibility and automation required to convert rising infrastructure investment into measurable efficiency and productivity gains."

CDW matched every entitlement to a user and to real usage across five source systems. Sanjay Sood, SVP and Chief Technology Officer, CDW: "It's not just about tracking cost. It's about seeing cost in context so we can optimize renewals, validate AI adoption, and drive smarter technology investments." On the visual page this quote sits under the approval example, in the section "You approve. Asato acts." It is a customer's own words, not a result on AI spend: every savings figure in the case study it comes from is Found, and none of them is a figure for AI spend. Case study: https://www.asato.ai/case-study-cdw-saas-visibility

No customer result on AI spend is published on this page.

Trust and governance

Works with

Compared with tools a buyer already owns

ToolWhere it stopsHow Asato relates
An AI vendor's own consoleIt covers one vendor, with no cost center, contract or shared policyAsato sets one policy in each vendor's own limits, where the vendor allows it
An AI gatewayIt covers the traffic routed through it, and sits in the pathAsato reads the gateway's logs and writes budgets into it
Cloud cost and SaaS management toolsThey show and attribute spendAsato adds the owner from the customer's records and has the budget enforced at the vendor
Card and spend toolsThey see the paymentAsato ties the spend to the directory and the contract, and acts on it

The comparison in full, row by row: https://www.asato.ai/asato-vs-ai-cost-tools

Fund AI from waste

Section heading on the visual page: "Fund AI from waste." The line under it: "What nobody uses can pay for the tokens you need."

The same record that holds the AI bill holds software, hardware and contracts, so what nobody uses there can pay for AI. The section names three places to look, each with its page:

Many organizations report being asked to fund AI through optimization savings, the FinOps Foundation finds. Source: FinOps Foundation, State of FinOps 2026, https://data.finops.org/

The section states no figure, and it is not a result on AI spend. Asato's whitepaper on the subject is The AI Funding Paradox: https://www.asato.ai/whitepaper-ai-funding-paradox

Who uses it

How a customer starts

Section heading on the visual page: "Where to start." The line under it: "One business unit. Your Asset Clarity Report takes one week, with read-only access."

The action: the button "Get your Asset Clarity Report", which leads to https://www.asato.ai/talk-to-sales.

Questions

The page shows at most five questions, the ones that belong to it. Any others are kept here for agents.

The visual page shows five of the questions below: the first four, and "We already have each vendor's console. What does Asato add?".

Is Asato in the path of our AI traffic?

No. Asato reads vendor admin feeds, billing exports and logs, and writes limits through each vendor's admin API. The vendor or the gateway does the blocking. If Asato is down, every AI call still works and every limit already written stays in force.

Does Asato see our prompts or outputs?

No. Asato works on usage counts, costs, identities and contracts. It never collects prompts, responses or files.

Can it stop one runaway agent and leave the rest running?

Yes, if the agent's traffic goes through the customer's gateway on its own key. Asato writes a budget to that key in advance, and the gateway refuses the agent's calls once the budget is spent. Other services on the same provider key keep running. Without a gateway, the owner gets an alert with the step to take, and Asato confirms afterward that the spending stopped.

What if a limit stops engineers in the middle of a release?

Nothing is written without a named person, or a policy the customer sets, as the approver. Requests for more usage from each vendor land in one queue. A raise has an end date, each write is read back, and a write can be reversed in one step.

How does Asato count savings?

As found, approved and saved. Only saved is reported. A saving is Saved when the customer and Asato agree it, at the close of the engagement. The evidence is the vendor's own data: a reclaimed seat counts when the vendor's roster no longer lists it.

What can Asato not see?

AI accounts people pay for personally, and any source with no admin API and no billing line. Each source Asato cannot read is listed as a gap. Asato does not estimate spend it cannot read.

We already have each vendor's console. What does Asato add?

The owner, and one policy. Each vendor's console covers its own product. A gateway covers the traffic routed through it, and sits in its path. Asato adds the owner, cost center and contract from the customer's own records, sets one budget policy in each vendor's own limits, where the vendor allows it, and counts a saving as Saved once the vendor's data shows it and the customer agrees the figure. The same record covers software, hardware and contracts.

Why not send everything through a gateway?

A gateway covers what is routed through it. A seat is not traffic, and chat assistants, coding tools and the AI inside Microsoft 365 connect straight to their vendors unless a customer routes them through it. A gateway attributes its traffic to the keys, users and teams set up in it. Where a customer runs LiteLLM or Azure API Management, Asato reads its logs and writes budgets into it.

Which vendors can Asato set a limit in?

It depends on what each vendor lets an administrator change through an API. Asato writes limits to Claude Enterprise, Cursor, GitHub Copilot and AWS Budgets, and to LiteLLM and Azure API Management gateways. For OpenAI API projects, Anthropic API workspaces, Microsoft 365 Copilot and cloud budgets, the owner gets an alert with the step to take, and Asato checks that it worked. The table above has each one.

What is AI tokenomics?

The practice of managing what an organization spends on AI tokens: what they cost, who spends them, what the spend returns, and how it is held to a budget. Asato works on the cost side: one ledger of seats, credits and tokens, and an owner, a budget and a limit for each line. It does not measure the business outcome of an AI workload. The term is defined in full at https://www.asato.ai/ai-tokenomics

Next step

Get your Asset Clarity Report at https://www.asato.ai/talk-to-sales.

Related pages

Last updated: 2026-10-12