Choosing the Right FinOps AI on Azure: Lessons from AWS FinOps Agent
AWS offers AWS FinOps Agent, a service that combines cloud cost monitoring, anomaly investigation, and optimization recommendations. As of August 17, 2026, it is available in Preview. Azure offers several options for the same goals, including FinOps Hub, Microsoft Cost Management, Azure Copilot, and Azure Advisor.
This investigation began when I learned that Microsoft had published a guide to configuring AI agents for FinOps hubs. After reading it, I wondered whether I could build something similar to AWS FinOps Agent on Azure and whether implementing it on this site would be worthwhile.
This article goes beyond a simple product comparison. It looks at which options fit systems of different sizes and what a practical starting architecture would look like for this site.
Table of Contents
- The short answer
- Why I started this investigation
- AWS FinOps Agent is a dedicated managed service
- Azure combines several capabilities
- How AWS FinOps Agent and a FinOps Hub AI agent differ
- The right choice depends on system scale
- A practical starting architecture for this site
- Define what the AI can do
- Summary
The short answer
AWS FinOps Agent and an Azure FinOps Hub AI agent take a similar approach to natural-language cost analysis, but they are delivered at different levels of abstraction.
- AWS FinOps Agent is a dedicated managed service that combines anomaly investigation, conversational analysis, recurring reports, optimization recommendations, and Jira and Slack integrations.
- A FinOps Hub AI agent consists of prebuilt instructions and connection guidance for analyzing FOCUS-formatted data stored in FinOps Hub through Azure Data Explorer and an AI client.
- Azure Copilot provides an interactive entry point for analyzing Cost Management data on demand.
- The Cost Management API provides the data needed to build small custom automations.
Azure does have a purpose-built option. Azure Copilot Optimization Agent explains recommendations for optimizing cost and carbon emissions and can generate Azure CLI and PowerShell scripts. As of August 17, 2026, however, it is in Preview and does not provide anomaly investigation, recurring reports, and external ticketing integration in a single product as AWS FinOps Agent does.
For a single site like this one, a lightweight approach is a better fit than an always-on FinOps Hub platform. Start with interactive analysis in Azure Copilot, then add the Cost Management API and Budget alerts when needed.
Why I started this investigation
What stood out to me in the AWS FinOps Agent announcement was that a cloud provider was offering a FinOps-specific agent as a managed service.
When I looked for something similar on Azure, the first option I found was an architecture for querying FinOps Hub data from GitHub Copilot or Copilot Studio. That suggested it would be possible to build a FinOps agent on Azure as well.
I investigated these questions in order:
- Are AWS FinOps Agent and a FinOps Hub AI agent the same thing?
- Can I build something smaller with the Cost Management API and skip FinOps Hub?
- How much interactive analysis can Azure Copilot already handle on its own?
- Is this worth implementing and testing on this site?
The answer depends less on which option has the best features and more on who performs the analysis, how many clouds or subscriptions they manage, and how often they need to analyze them.
AWS FinOps Agent is a dedicated managed service
The official AWS FinOps Agent documentation lists these primary capabilities:
| Feature | Description |
|---|---|
| Cost inquiry | Ask natural-language questions about Cost Explorer data |
| Anomaly investigation | Investigate causes in response to Cost Anomaly Detection events |
| Recurring reports | Generate daily, weekly, and monthly reports in HTML, PDF, and PPT formats |
| Optimization | Consolidate recommendations from Cost Optimization Hub and Compute Optimizer |
| External integration | Send investigation results and optimization opportunities to Jira or Slack |
| Context and memory | Continuously use organization-specific documents and user preferences |
For anomaly investigations, the service examines not only cost changes but also API activity recorded in AWS CloudTrail. Its role extends to correlating periods of increased cost with infrastructure changes and compiling the findings into an investigation report.
On the agent creation screen, you configure IAM roles for the agent and web application, along with connections to Jira and Slack. The service also generates the user-facing web application.
As of August 17, 2026, AWS FinOps Agent is in Preview. Its specifications and availability may change.
Azure combines several capabilities
Rather than handling every FinOps task in one service, Azure lets you combine capabilities based on your needs.
| Option | Strength | Best suited for |
|---|---|---|
| Azure Copilot | Interactively review Cost Management summaries, breakdowns, comparisons, forecasts, and savings recommendations | On-demand analysis by a person |
| Azure Copilot Optimization Agent | Explain Advisor recommendations and generate alternatives and implementation scripts | Turning optimization opportunities into concrete actions |
| Cost Management API | Programmatically retrieve Query, Forecast, and Cost Details data | Small recurring jobs or custom agents |
| Cost Management Export | Periodically export cost data, including FOCUS data, to Storage | Long-term retention and external analysis |
| FinOps Hub | Ingest and normalize FOCUS data from multiple scopes for advanced analysis | Large-scale, multi-organization, or multi-cloud analysis |
| Budget/Anomaly alerts | Notify you of budget overruns or unusual cost changes | Triggering continuous monitoring |
You can ask Azure Copilot questions such as, “Summarize costs for the past six months,” “Break down last month’s costs by service and region,” and “Forecast the next three months.” It uses the scope most recently opened in Cost Management as the conversation context and asks you to select a scope when that context is ambiguous.
That is useful enough for routine checks on this site. However, conversations cannot continue for more than 24 hours, and some lists are limited to the top five results. Ongoing operations, such as automatically generating a monthly report or investigating the Activity Log and opening an issue when an anomaly occurs, require a separate mechanism.
Using FinOps Hub
FinOps Hub ingests Cost Management exports, processes them with Storage and Data Factory, and provides a foundation for analysis in Azure Data Explorer or Microsoft Fabric. When you use Azure Data Explorer, it creates a Hub database and an Ingestion database.
To connect an AI client, first configure FinOps Hub and scope ingestion. The AI client can then query the Hub database through Azure MCP Server or Kusto Query MCP Server. The provided instructions include the FOCUS schema and KQL query patterns.
According to Microsoft’s estimates as of August 17, 2026, a configuration that includes Azure Data Explorer starts at $120 per month, with approximately $10 in additional storage and processing costs for every $1 million in monitored spend. The example using Microsoft Fabric F2 starts at $300 per month. Actual costs vary by agreement, region, data volume, Power BI licensing, and other factors, so check the official cost estimate before deployment.
Using the Cost Management API
If you only need to manage a single subscription or resource group, you can query the API directly without deploying FinOps Hub.
- Use the Query API to retrieve actual costs by service, resource group, resource, or tag
- Use the Forecast API to predict end-of-month and future costs
- Use Azure Advisor to retrieve savings recommendations and estimated savings
- Use Resource Graph to inspect current SKUs and configurations
- Use Activity Log to review deployments and configuration changes made before a cost increase
Use an SDK where one is available and supplement it with the REST API for capabilities whose SDK support is still catching up. This approach is easier to operate in a small environment because the agent can be limited to a small set of read-only tools.
How AWS FinOps Agent and a FinOps Hub AI agent differ
The two approaches aim for similar outcomes, but they differ in how complete the delivered solution is.
| Comparison | AWS FinOps Agent | FinOps Hub AI agent |
|---|---|---|
| Delivery model | Dedicated managed FinOps service | Architecture pattern connecting FinOps Hub to an AI client |
| Data sources | Cost Explorer, Cost Anomaly Detection, Optimization Hub, Compute Optimizer, and CloudTrail | FOCUS and pricing data ingested into FinOps Hub |
| Conversational analysis | Dedicated web application | GitHub Copilot, Copilot Studio, or another client |
| Anomaly investigation | Automatically triggered by events | Analysis with KQL; automatic triggering requires separate configuration |
| Recurring reports | Generates HTML, PDF, and PPT files | Requires separate queries and workflows |
| Task integration | Jira and Slack | Copilot Studio, Teams, Logic Apps, or a custom ITSM integration |
| Optimization recommendations | Integrates Optimization Hub and Compute Optimizer | Combines FOCUS analysis, Azure Advisor, and other capabilities |
| Organizational context | Context files and memory | Instructions and knowledge files that you configure |
AWS packages the FinOps practitioner’s workflow into one product. Azure provides Cost Management, Advisor, Copilot, and FinOps Hub as components that you can combine as needed.
The AWS option is therefore easier to get started with, while the Azure approach is easier to adapt to an existing data platform, identity model, and workflow.
The right choice depends on system scale
The choice depends more on the complexity of cost management than on the AI’s capabilities.
| System profile | Start with | Why |
|---|---|---|
| Single site, single subscription | Cost Analysis and Azure Copilot | Review breakdowns, comparisons, and forecasts without additional infrastructure |
| A few subscriptions that need recurring notifications | Budget alerts and the Cost Management API | Build small, event-driven automations |
| Multiple subscriptions, long-term history, and shared reports | FOCUS exports and Fabric/ADX | Retain data and analyze it using common metrics |
| Multiple tenants, multiple clouds, and departmental chargeback | FinOps Hub | Normalize data to FOCUS across scopes and providers |
| Conversational access for a large organization | FinOps Hub and Copilot Studio | Publish governed data to Teams or Microsoft 365 Copilot |
FinOps Hub is especially valuable when you need to consolidate data from multiple subscriptions or tenants, including AWS or GCP data, and continuously analyze allocations or commitment discounts by department, product, or customer. Azure Data Explorer also has an advantage when you repeatedly run complex queries over several years of data.
By contrast, if you check the cost of a single site only a few times a month, the fixed cost and operational overhead of FinOps Hub can become a new cost driver. Base the decision on the absolute amount you can save and the hours of work involved each month, not just the percentage reduction.
A practical starting architecture for this site
This site uses Azure Static Web Apps, Azure Functions, Storage, Azure Front Door, Application Insights, and Log Analytics, but the environment centers on a single site. Departmental chargeback, Reservations, Savings Plans, and integrated multi-cloud analysis are not current requirements.
For now, it makes sense to skip FinOps Hub and proceed in this order:
- Use Azure Copilot to interactively review costs for the past six months, services with increased costs, forecasts, and Advisor recommendations
- Set a budget at the resource group level, and consider alert thresholds of 90%, 100%, and 110% for actual costs and 110% for forecast costs
- If ongoing investigation becomes necessary, call the Cost Management API from a timer or alert trigger
- Combine cost data, current resource details, and Activity Log entries, then record possible causes and recommended actions in a GitHub issue
- Reevaluate FOCUS exports and FinOps Hub when the number of subscriptions or the need for long-term analysis grows
The proposed architecture looks like this:
You can route a Budget alert to a Logic Apps workflow through an action group. As of August 17, 2026, Cost Anomaly alerts rely primarily on email notifications, so Microsoft’s official guidance for automation is to have Logic Apps monitor a dedicated mailbox.
The investigation process combines the Cost Management Query API with Azure Advisor, Resource Graph, and Activity Log. This narrows a possible cause from merely “Log Analytics cost increased month over month” to “which meter increased, and which diagnostic setting was added immediately beforehand.”
This architecture has not yet been implemented on this site. It is a design proposal based on the findings from this investigation.
Define what the AI can do
In cost optimization, automatically making changes based only on potential savings can undermine availability, security, and recovery requirements.
For example, shortening log retention, removing an Always Ready instance, and changing storage redundancy affect retention requirements, cold starts, and disaster recovery, respectively. A safer approach is to limit the agent to this boundary:
| Let the AI handle | Leave to people |
|---|---|
| Read cost, forecast, resource, and change history data | Decide which budgets and risks to accept |
| Identify services and meters with increased costs | Decide whether reduced availability or security is acceptable |
| Explain Advisor recommendations | Decide whether to adopt a recommendation |
| Create issues with possible causes, evidence, and recommended actions | Decide whether to change Bicep or Azure resources |
Assign the agent’s managed identity only the read permissions it needs, such as Cost Management Reader, Reader, and Monitoring Reader. Do not allow it to stop resources, change SKUs, or delete anything. Instead, have it open a GitHub issue so a person can review the proposed changes.
Include supporting evidence in every response. Recording the target scope, aggregation period, cost type, amount of the increase, resource ID, and timestamps from the Activity Log makes it possible to verify the result rather than relying on the AI-generated explanation alone.
Summary
AWS FinOps Agent and an Azure FinOps Hub AI agent both use AI to analyze cloud costs, but they are delivered differently.
AWS FinOps Agent is a dedicated service that combines anomaly investigation, recurring reports, optimization, and Jira and Slack integrations. On Azure, you combine Azure Copilot, Optimization Agent, the Cost Management API, Advisor, and FinOps Hub based on your goals and scale.
For a single site, simply asking Azure Copilot questions about Cost Management can already provide substantial insight. When you need automation, add Budget alerts and the Cost Management API. Consider FinOps Hub when you need multiple subscriptions, long-term history, departmental chargeback, or multi-cloud analysis.
For this site, starting with Azure Copilot and, if necessary, a lightweight read-only agent that opens GitHub issues is a better fit for the site’s scale and cost than deploying FinOps Hub solely for evaluation.
References
- AWS FinOps Agent
- What is AWS FinOps Agent?
- FinOps hubs
- Configure and use AI agents for FinOps hubs
- Analyze, estimate, and optimize cloud costs using Azure Copilot
- Azure Copilot Optimization Agent
- Microsoft Cost Management REST API
- Create and manage Cost Management exports
- Identify anomalies and unexpected changes in cost