The FinOps for AI maturity model: Where does your organization stand?
By now, most companies have some handle on cloud costs. They have tagging strategies, budget alerts, reserved instance policies, and monthly cost reviews. Then AI spend enters the picture and a lot of that discipline goes out the window.
AI costs move faster and they're harder to attribute. A developer spins up a Bedrock integration or an Azure OpenAI endpoint. A data team starts running models on SageMaker. A product team adds an AI feature. Before long, you have significant spend with no clear picture of where it's going or who's responsible for it.
And most organizations are further behind on managing it than they think.
The maturity model below is a way to honestly assess where your organization stands and what to focus on next.
Level 1: No visibility
This is where most teams start, and more organizations are here than would admit it.
At this level, AI spend shows up as a single line item in your cloud bill. You know roughly what you're spending, but you can't break it down by team, model, or application. There's no tagging in place, no budget alerts, and no process for reviewing AI costs. Teams are spinning up models and making API calls with no central oversight.
The risk here isn't just overspending. It's that you have no way to catch a problem until it shows up in the invoice. A new feature goes live, usage spikes, and you find out three weeks later when the bill lands.
Signs you're at Level 1
- AI spend is a single aggregated number in your billing dashboard.
- You have no tags for Bedrock, Azure OpenAI, SageMaker, or Azure Machine Learning resources.
- No one owns AI cost management.
- Cost reviews happen after the fact, if at all.
Level 2: Basic visibility
At this level, some cost data exists but it's usually messy and manual. Teams might be pulling billing exports into spreadsheets or using basic AWS Cost Explorer or Azure Cost Management views to get a rough breakdown. Some tagging is in place but it's inconsistent, which means some resources are tagged, others aren't, and the naming conventions vary by team.
The main characteristic of Level 2 is that you're reactive. You can investigate a cost spike after it happens, but you're not set up to catch one before it does. Finance knows there's an AI spend problem but engineering doesn't feel the pressure, and the two teams aren't working from the same data.
Signs you're at Level 2
- You can break down costs roughly but it takes manual effort.
- Tagging exists but isn't enforced or consistent.
- Budget alerts are either missing or set too high to be useful.
- Cost conversations happen monthly, not continuously.
Level 3: Active management
This is where cost visibility becomes operational. Spend is broken down by team, model, and application automatically. Budgets are set at a granular level, alerts fire before costs spiral, and someone actually owns the response when they do.
At this level, engineering and finance are working from the same data. Teams can see their own AI spend, which creates accountability. Optimization starts happening proactively, where teams make deliberate decisions about model selection, prompt caching, and whether workloads need to run in real time.
Unit economics also start to matter here. It's not just, "We spent $40,000 on Bedrock last month." It's, "Our cost per API call is $0.008 and it's been climbing for six weeks." That's the kind of signal that leads to action.
Signs you're at Level 3
- Spend is broken down by team, model, and application in a dashboard.
- Tagging is enforced and consistent.
- Alerts fire on anomalies and threshold breaches, not just at month end.
- Engineering teams are actively optimizing, not just spending.
Level 4: Optimized and governed
At this level, AI cost management is embedded into how engineering teams work. Cost is a consideration at the design stage, not an afterthought. Every workload has clear cost attribution, and unit economics are tracked continuously.
Teams at Level 4 are doing things like running model benchmarks before choosing which one to deploy, setting per-feature cost budgets, and treating cost per output as a product metric alongside latency and quality. AI spend is forecasted, not just reported, and those forecasts are accurate enough to inform business decisions.
This level takes time to reach and requires buy-in across engineering, finance, and leadership. But the organizations that get here tend to scale AI much more confidently—because they understand what it costs to do so.
Signs you're at Level 4
- Cost attribution is complete across every AI workload, regardless of cloud platform.
- Unit economics tracked per model, per feature, per team.
- AI spend is part of product planning, not just finance reviews.
- Optimization is continuous, not a one-off project.
How to move up a level
Level 1 to Level 2: Get the basics in place
Start with your AWS and Azure tags. Make sure Bedrock, Azure OpenAI, SageMaker, and Azure Machine Learning resources are tagged by team, application, and environment. Then connect a cost management tool so that data is actually usable. The goal is to stop working off a single aggregated number.
Level 2 to Level 3: Turn visibility into action
Visibility alone doesn't change behavior. Set budgets that are tight enough to matter, make sure alerts go to the people who can act on them, and assign clear ownership over AI cost management. When a spike happens, there should be a person and a process, not a scramble.
Level 3 to Level 4: Make cost part of how you build
This is a culture shift more than a tooling one. Cost needs to be part of engineering decisions at the design stage, not something that gets reviewed after the fact. Teams at this level treat cost per output as a product metric, the same way they treat latency or error rates.
Most organizations are at Level 1 or Level 2. The jump to Level 3 is more achievable than it looks, and it's where the biggest practical gains are.
How CloudSpend accelerates your FinOps for AI practice
CloudSpend is built for teams trying to move from Level 1 to Level 4 across AWS, Azure, and GCP. Connect your accounts and you get:
- A breakdown of your AI spend by model, team, and application, including for Bedrock, Azure OpenAI, SageMaker, and Azure Machine Learning
- Cost per million tokens tracked per model, so you can directly compare what different models are costing you across both clouds
- Threshold- and anomaly-based alerts across email, Slack, and other channels
- Full visibility into what you're spending, where, and why, so engineering and finance are working from the same data
If you're not sure where your organization sits, that's usually a sign you're at Level 1. The first step is getting a clear picture of what you're actually spending.
Sinjan covers cloud cost management and FinOps at ManageEngine. He writes about the financial realities of running cloud infrastructure, from optimizing cloud spend to fostering cost accountability across teams, helping organizations move from cloud cost chaos to confident, data-driven decisions.