The AI FinOps Maturity Model in 2026
Managing AI spend is near universal, but three in four organizations still cannot connect it to a measurable outcome. Here is what each stage of the FinOps maturity model looks like for AI, the signal that proves you are in it, and how to move up one stage in a quarter.
The FinOps maturity model rates capabilities, not teams, and that matters more for AI than it did for cloud. The FinOps Foundation is explicit that reaching Run in every capability is not the goal, so the useful question is which single capability is holding your AI practice back.
Managing AI spend is solved. Proving its value is not. 98% of the 1,192 practitioners in the FinOps Foundation’s 2026 survey manage AI spend, up from 31% two years earlier, while 43% name proving value or ROI as their top challenge.
Naming an owner is the cheapest move available. Organizations with defined ownership of AI economics are 3.7 times more likely to show value to the CFO, and 12% still have not defined one.
Visibility is not proof. 39% of organizations are not confident they can connect AI spend to a measurable outcome and another 34% only slightly, and 60% of those who rate their visibility as moderate still cannot show an outcome.
Each stage has one signal, and a trap. Share of spend attributed to a named owner proves stage two; one unit metric in the monthly review, presented by someone outside FinOps, proves stage three. The traps are attributing by API key when one key serves several features, and ownership that exists on an org chart and nowhere else.
How the FinOps maturity model changes when the spend is AI, the signal that proves each stage, and how to move up one level in 90 days.
Your AI bill tripled last quarter and finance wants to know which team spent it. If the honest answer is that you can see a total per provider but cannot split it by team, feature or customer, you are not behind. You are roughly where the market is.
Managing AI spend is now near universal. In the FinOps Foundation’s 2026 State of FinOps survey, 98% of the 1,192 practitioners who responded said they manage AI spend, up from 31% two years earlier. Proving the value of that spend is where it falls apart. The September 2026 State of Tokenomics survey of 472 organizations found proving value or ROI to be the most commonly named challenge at 43%, and 12% have still not defined who owns AI economics.
What is the FinOps maturity model?
The FinOps maturity model is the FinOps Foundation’s Crawl, Walk, Run framework for rating how far a practice has developed in any given capability. It rates capabilities rather than whole teams. The Foundation’s own guidance is that reaching Run in every capability is not the goal, and that teams should mature the capabilities that deliver the highest business value first.
| Stage | What it looks like |
| Crawl | Reporting and tooling are minimal. Basic KPIs and policies are defined but not followed by every major team. The focus is low hanging fruit. |
| Walk | The capability is understood and followed across the organization. Automation or process covers most requirements. Hard edge cases are identified, and some are knowingly ignored. |
| Run | Every team follows the capability. Difficult edge cases get addressed. Automation is the preferred approach and KPIs are set very high. |
Maturity improves with repetition, which is why the model describes a direction rather than a finish line. If you are building the practice itself rather than assessing it, FinOps 101 covers the structure, and the FinOps principles cover the operating rules underneath it.
Why the FinOps Maturity Model Breaks Down for AI Spend
The framework still holds for AI, but three properties of AI spend defeat the tooling and the cadence that got teams through cloud FinOps. Cost moves per call rather than per month, a large share of it never touches a cloud bill, and there is no commitment instrument to fall back on.
AI cost is per request, not per resource. A cloud instance costs the same whether it is busy or idle, which makes monthly review a reasonable cadence. A model call costs what its tokens cost, so a prompt change shipped on a Tuesday can reset your run rate before the invoice closes. Monthly review catches that four weeks late.
The spend is scattered across billing relationships. In the Tokenomics survey, 96% of respondents buy from a model provider directly, 87% buy tokens through a cloud, 64% pay for AI embedded in other tools, 31% rent GPUs and 29% run their own hardware. Half of respondents use four of those six channels. Only some of that lands on a cloud bill, so a practice that starts from the cloud bill starts with a partial picture.
There is no reserved instance for tokens. Commitment management, the lever that carried cloud FinOps through its middle years, has no equivalent here. The available levers are model choice, routing, caching and prompt design, all of which sit with engineering rather than procurement. That changes who has to be in the room.
The Four Stages of the AI FinOps Maturity Model
The AI FinOps maturity model moves through four stages: seeing the total, splitting it by owner, measuring it as a unit cost, and governing it with budgets and alerts. These sit inside the Foundation’s three, with stage one tracking Crawl, stage two tracking Walk, and stages three and four splitting Run into measurement and control. Each stage has one signal that proves you are actually there, which matters because every stage has a version that looks finished from the inside and is not.
Stage 1, See It
You know what you spent on AI last month, by provider. Invoices arrive, somebody totals them, and the number goes in a deck. This is where the Foundation’s Crawl guidance puts AI pilots and prototypes, with manual tracking and cost limits set up front.
The signal: you can state last month’s total for every AI provider without opening five consoles.
The trap: counting only what shows up on the cloud bill. Direct model provider invoices and AI embedded in developer tools are the two that go missing, and together they are most of the spend for many teams.
Stage 2, Split It
Every provider lands in one view, refreshed daily rather than monthly, and spend splits by team, feature, project or API key. This is the stage where a team can be shown last month’s number for their own feature and recognize it as theirs.
The signal: the share of AI spend attributed to a named owner. If that share is under 100%, the gap is your backlog. Attributing AI API costs covers the mechanics for teams whose keys and projects do not map cleanly to teams.
The trap: attributing by API key when one key serves several features. The allocation reconciles to the invoice and still tells nobody which product drove the change.
Stage 3, Measure It
Spend becomes a rate rather than a total. Cost per session, cost per agent run, cost per resolved ticket or cache hit rate get reported alongside the dollar figure, so growth and inefficiency stop looking identical on a chart. The Foundation’s Run guidance for AI names unit economics in leadership reporting as the marker here.
The signal: at least one AI unit metric appears in the monthly business review, presented by someone outside the FinOps team. Our FinOps KPIs guide covers which metrics survive that audience.
The trap: choosing a metric only the platform team understands. Cost per 1M tokens is easy to compute and means nothing to a CFO. Cost per customer conversation means something to everybody.
Stage 4, Govern It
AI spend has a named owner, budgets that are set before the quarter rather than explained after it, alerts that fire on a step change within a day, and model and caching decisions made from measured cost rather than preference. Run maturity here means costs stay proportionate to delivered value, with budgets stable by component.
The signal: a named owner, plus spikes caught within a day rather than at month end.
The trap: ownership that exists on an org chart and nowhere else. The test is whether that person can stop a spend decision, not whether their title mentions AI.
The AI FinOps Maturity Model Matrix
Read down the column that matches your weakest capability rather than your strongest, because a practice is only as mature as the capability holding it back. This is the part of the FinOps maturity model that tends to be flattering in summary and uncomfortable line by line.
| Capability | Stage 1, See It | Stage 2, Split It | Stage 3, Measure It | Stage 4, Govern It |
| Visibility | Monthly invoice per provider | All providers in one daily view | Daily, down to the request | Request level with live alerting |
| Attribution | None, or cloud only | By team, project or key | By feature and customer | Fully attributed, reconciled monthly |
| Unit metrics | None | Cost per 1M tokens | Cost per session and per agent run | Unit cost tied to a revenue or outcome metric |
| Forecasting | Last month plus a guess | Trend by provider | Modeled on usage drivers | Budget by component, variance explained |
| Guardrails | None | Alerts after the fact | Thresholds per team | Budgets, caps and routing set before spend |
| Ownership | Nobody named | A reluctant volunteer | Shared between eng and finance | Named owner with authority to stop spend |
Score Your Own Practice in 10 Questions
Answer yes or no. Each yes is one point, and the total maps to a stage.
0 to 2 points: Stage 1. 3 to 5: Stage 2. 6 to 8: Stage 3. 9 to 10: Stage 4. Questions 1 and 10 carry more weight in the FinOps maturity model than the count suggests, because ownership and pre-set budgets are the two things that cannot be retrofitted from a dashboard.
Where Most Companies Sit Today
Most organizations have ownership on paper and cannot yet prove value, which places the market between stage two and stage three of the AI FinOps maturity model. The Tokenomics survey put AI spend ownership with technology leadership at 35%, shared across functions at 26%, and undefined at 12%, with the balance spread across executive teams, AI leadership and finance. So 88% have named someone.
Naming someone matters more than it sounds. Organizations with defined ownership of AI economics were 3.7 times more likely to be able to show value to the CFO. Of everything the survey measured, it is the change that costs the least to make.
Measurement is where the market stalls. Asked how confident they were in connecting AI spend to a measurable outcome, 39% said not confident and 34% said slightly, so roughly three in four cannot make that connection today. Visibility alone does not fix it either: 60% of the respondents who were moderately confident in their spend visibility still could not show a measurable outcome.
One lever shows up with a measured payoff. 86% of respondents are evaluating or using a model router, and router users were 4 times more likely to be able to show AI value, which fits a tool that puts a price on each request at the moment it is made. For the wider picture of how cloud and AI practices are converging, our State of FinOps 2026 breakdown covers all seven findings.
How to Move Up One Stage in 90 Days
Pick the next stage of the FinOps maturity model rather than the last one, and run two or three concrete changes. A quarter is enough for one stage transition and not enough for two.
Stage 1 to Stage 2. Connect every AI provider, including the ones billed outside your cloud account, and get them into one view. Then give each API key or project a team name, and publish the first attributed breakdown even when coverage is only 60%, because the gaps are what generate the next set of tasks.
Stage 2 to Stage 3. Choose one unit metric that a non technical executive would understand, such as cost per customer conversation or per completed agent run. Compute it for the last three months so the first report has a trend rather than a single point, then put it in the monthly review and have the product or engineering owner present it.
Stage 3 to Stage 4. Name the owner in writing and give them authority over model and routing decisions. Set next quarter’s AI budget by component before the quarter opens, and set alerting tight enough that a step change reaches a human the same day. Then start pricing model and caching choices against the unit metric from stage three rather than against list rates.
How Economize Supports Each Stage of the FinOps Maturity Model
Economize is built for the attribution and unit metric work that stages two and three depend on, with AI spend sitting in the same account as the cloud bill rather than in a separate tool.
- AI Costs tracks OpenAI, Anthropic, Amazon Bedrock, Google Vertex AI and Azure AI Foundry in one dashboard. OpenAI and Anthropic connect with an API key, and the three cloud providers come with your existing cloud connection.
- Attribution for stage two. Break spend down by provider, model, agent, project, region or token type. The first sync loads 180 days of history, so your first attributed report arrives with a trend rather than a single month.
- Unit metrics for stage three. Cost per 1M tokens, per request, per session and per agent run, including what one full agent run costs, which no provider reports on its own.
- Request level logs. When a cost jumps, the usage logs show the model, status, latency, tokens in and out, and estimated cost per call. Prompts and responses stay hidden unless you enable capture for that provider.
- Reports scope any view to an AWS account, GCP project or Azure subscription, and any report can be sent to a dashboard, exported as CSV for the month-end pack, or turned into an alert in Slack, Teams or email.
- Ask answers cost questions against your own data in plain language, so the monthly number does not have to route through whoever owns the dashboard.
Setup takes about five minutes with no agents to deploy, and you can start free on a single cloud account.
The FinOps maturity model is the FinOps Foundation’s Crawl, Walk, Run framework for rating how developed a FinOps practice is in a given capability. Crawl means minimal tooling and basic KPIs, Walk means the capability is followed organization wide with most requirements automated, and Run means every team follows it with automation preferred. The Foundation is explicit that reaching Run everywhere is not the goal.
AI cost moves per request rather than per resource, so monthly review arrives weeks late. Much of the spend sits outside the cloud bill, since 96% of surveyed organizations buy from model providers directly alongside the 87% buying tokens through a cloud. And there is no commitment instrument for tokens, so the levers are model choice, routing and caching rather than reserved capacity.
Share of AI spend attributed to a named owner measures stage two. At least one unit metric in the monthly review, such as cost per session or per agent run, measures stage three. Time to detect a step change and forecast variance by component measure stage four.
Plan on one stage per quarter rather than one jump. Connecting providers and attributing spend is a few weeks of work, agreeing a unit metric that finance accepts takes a reporting cycle or two, and governance takes as long as it takes to give one person real authority over model decisions.
98% of the 1,192 FinOps practitioners in the FinOps Foundation’s 2026 survey manage AI spend, up from 63% the previous year and 31% two years earlier. Managing it is close to universal. Proving its value is not, with 43% naming that as their top challenge.
Connect your cloud accounts and AI providers to a platform that reports both against the same dimensions. Economize covers OpenAI, Anthropic, Bedrock, Vertex AI and Azure AI Foundry next to AWS, Azure and GCP spend, with cost per 1M tokens, per session and per agent run, free up to $100,000 a month in tracked spend.
Start With the Stage You Are Actually In
The FinOps maturity model rewards honesty more than ambition. With roughly three in four organizations unable to connect AI spend to a measurable outcome, the fastest quarter most teams can run is the one that closes the attribution gap rather than the one that promises governance.
Our pricing is published if you want to size it yourself, or book a demo and we will walk the four stages against your own accounts.
Simran Sardar
FinOps enthusiastProduct Manager at Economize with over 3 years of experience, focused on FinOps strategies and cloud cost optimization. Dedicated to helping organizations streamline cloud expenses and drive financial efficiency.
Maximize Cloud Efficiency and Optimize Costs
Get started free in our sandbox or book a personalized call with our experts
