Tag Archives: api management for ai

Azure API Management Token Metric Policy: AI Cost Observability and Cross-Charging

Posted on by
6

Part 4 of 7 in the “APIM for AI Workloads” series

The Azure API Management token metric policy turns AI cost data from a finance problem into an engineering one. In Part 3, we covered enforcement: how to set consumption boundaries per consumer. This post covers the complementary piece: how to measure that consumption. More importantly, it shows how to make it visible to the right people and use it to drive internal cross-charging and FinOps dashboards.

At my current company, one of the first questions the architecture board asked was straightforward: which teams are consuming what, and what does it cost? Without instrumentation at the gateway layer, that question is genuinely unanswerable. The token metric policy is how you answer it.

Azure API Management Token Metric Policy: How It Works

The policy sits in the outbound section of your APIM pipeline. After the AI backend returns a response, APIM reads the token usage fields from the response body. These include prompt tokens, completion tokens, and total tokens. APIM then emits them as custom metrics to Application Insights under a namespace you define.

Crucially, the policy emits metrics after the response arrives. It uses actual token counts from the API response rather than estimates. As a result, the data is accurate rather than approximated. It also means the metric emission adds no latency to the request path: the response is returned to the caller immediately, and the metric is emitted asynchronously.

Azure API Management token metric policy observability pipeline emitting token counts to Application Insights for cross-charging
Diagram 1: Token metric policy observability pipeline. Token counts from the AI backend response flow through the APIM metrics layer to Application Insights, broken down by dimensions for cross-charging and cost allocation.

The generic variant, llm-emit-token-metric, works identically for non-Azure backends. Both policies share the same dimension model, so the configuration patterns below apply regardless of which AI provider sits behind APIM.

Choosing Dimensions for Azure API Management Token Metric Policy

Dimensions are the labels attached to each metric event. They explain how to slice and aggregate token consumption data in Application Insights. Choosing the right dimensions is the most important configuration decision for making the data useful for cross-charging.

Azure API Management token metric policy dimension strategies for cross-charging using Subscription ID User ID and API ID
Diagram 2: Three-dimensional strategies for cross-charging and showback. Subscription ID maps to teams and cost centers, User ID enables per-user billing in multi-tenant apps, and API ID breaks down cost by AI workload or feature.

The three primary dimension options are:

Subscription ID. The most common choice for internal enterprise deployments. Each APIM subscription maps to a team, product, or cost center, so filtering Application Insights metrics by Subscription ID gives you direct per-team token consumption. This pairs naturally with the subscription key authentication pattern from Part 2 and the per-subscription counter-key from Part 3.

User ID. Sourced from the JWT subject claim or a custom header, User ID enables per-user consumption reporting. This is the right dimension for multi-tenant SaaS applications where individual end users have their own token budgets, or where you need to identify heavy consumers within a shared subscription.

API ID. Identifies which APIM API product generated the consumption. Useful when a single subscription uses multiple AI-backed APIs: one for a conversational agent, one for content generation, and one for document summarization. API ID lets you break down cost by use case rather than just by subscriber.

In practice, combining all three dimensions gives you the most flexibility. A single metric event tagged with Subscription ID, User ID, and API ID can answer questions at every level: how much did the platform spend in total, how much did Team A spend, how much did User X consume, and which AI feature is the most expensive to run.

Querying Token Metrics in Application Insights

Once the policy is emitting metrics, you query them in Application Insights using the custom metrics namespace you configured. The metrics appear under the namespace name you set in the policy (for example, “AzureOpenAI” or “MyLLM”), with separate metric events for prompt tokens and completion tokens.

A practical starting point is a KQL query that aggregates the total number of tokens by Subscription ID over the past 30 days. From there, you can add filters by API ID to isolate specific workloads, or pivot by User ID to identify the highest consumers within a team.

For FinOps dashboards, the most useful view is a stacked time-series chart of total token consumption broken down by subscription, updated daily. This gives finance and engineering a shared view of AI spend trends without exporting data from Azure Monitor to a separate BI tool. Azure Workbooks can host this directly in the Azure portal, making it accessible to non-technical stakeholders.

From Observability to Cross-Charging

Observability is the prerequisite for cross-charging. However, they are not the same thing. Observability tells you what happened. Cross-charging, by contrast, is the organizational process of allocating those costs to the right budget owners.

The token metric policy gives you the raw data. To turn that into a cross-charge, you need two additional steps. First, agree on a price per token with your finance team — usually derived from the Azure cost per 1,000 tokens for your model and region. Second, automate a monthly report that multiplies token consumption by the subscription price.

This does not need to be complex. For example, a Logic App or Azure Function that queries Application Insights on the first of each month works well for most organizations starting out. It aggregates tokens by subscription, multiplies by the agreed rate, and emails a cost summary to each team lead. The Application Insights REST API makes this straightforward to automate.

Finally, the most important advice: have this conversation with finance and product teams before AI consumption scales. Retroactive cross-charging is significantly harder to establish than an upfront model with clear methodology and tooling.

What’s Next in This Azure API Management for AI Series

Part 5 covers load balancing and circuit breaking: how to distribute traffic across PTU and PAYG backends, configure backend pools, and set up circuit breaker rules for automatic failover when a primary endpoint becomes unavailable.

Azure API Management Token Limit Policy: Controlling AI Token Consumption Per Consumer

Posted on by
5

Part 3 of 7 in the “APIM for AI Workloads” series

The Azure API Management token limit policy is one of the most direct cost control levers you have for AI workloads. In Part 1 of this series, I argued that token consumption is invisible without the right instrumentation. The token limit policy is the enforcement side of that equation: once you know how many tokens consumers are using, you set boundaries so that no single consumer can exhaust your model capacity or run up an unexpected bill.

This post covers how the policy works, which counter-key strategy to choose for your workload, how to size your tokens-per-minute (TPM) limits, and the difference between the Azure OpenAI-specific policy and the generic LLM variant for non-Microsoft backends.

Azure API Management Token Limit Policy: How It Works

The azure-openai-token-limit policy sits in the inbound section of your APIM policy pipeline. Before any request reaches the AI backend, APIM checks a sliding window counter keyed to the value you specify. If the caller is within their TPM budget, the request passes through. If they’ve exceeded it, APIM returns a 429 Too Many Requests response with a Retry-After header, and the backend never sees the request.

This is important: the throttling happens at the gateway, not at the Azure OpenAI endpoint. That means you’re not paying for rejected requests, and your model deployment is protected from saturation by a single runaway consumer.

Azure API Management token limit policy funnel throttling AI requests with 429 response and Retry-After header
Diagram 1: The token limit policy acts as a funnel in the APIM inbound pipeline. Requests within the TPM budget pass through to the AI backend. Requests exceeding the limit receive a 429 status code with a Retry-After header before the backend is even reached.

The policy has two variants. The azure-openai-token-limit policy is purpose-built for Azure OpenAI and Microsoft Foundry endpoints, and uses the actual token counts returned in the API response. The llm-token-limit policy is the generic variant for any LLM backend, including Mistral, Cohere, and others. Both share the same attribute model, so the configuration patterns below apply to either.

Choosing a counter-key for Azure API Management Token Limiting

The counter-key attribute is the most important decision in configuring the token limit policy. It determines the scope of the limit: who shares a TPM bucket, and who gets their own.

Azure API Management token limit policy counter-key strategies per subscription IP address and JWT claim with TPM sizing table
Diagram 2: Three counter-key strategies and TPM sizing guidance by workload type. The right scope depends on whether you are separating teams, protecting a public endpoint, or enforcing per-user limits in a multi-tenant application.

The three main strategies are:

Per subscription: @(context.Subscription.Id). This is the most common pattern for internal enterprise use. Each API product subscription gets its own TPM counter, which maps cleanly to a team, a product, or a cost center. Combined with the Token Metric policy covered in Part 4, this provides per-subscriber cost visibility and enforcement in a single configuration.

Per IP address: @(context.Request.IpAddress). Better suited to public-facing endpoints or developer portals where you don’t have a subscription model. It’s a blunt instrument — NAT and shared egress can mean multiple users share a counter — but it’s effective for abuse prevention and trial access scenarios.

Per JWT claim or custom header: @(context.Request.Headers.GetValueOrDefault(“x-user-id”,””)). The most flexible option. If your application passes a user identifier in a header or JWT claim, you can scope limits to the individual user. This is the right approach for multi-tenant applications where each end user should have their own token budget, independent of which subscription they’re calling through.

Sizing Your TPM Limits

TPM limits are context-dependent, but a few principles apply across most workloads.

Start by profiling your actual token usage in a staging environment before setting production limits. The remaining-tokens-variable-name attribute exposes the remaining token budget as a policy variable, which you can log via the Token Metric policy to build a usage baseline before enforcing hard limits.

For the estimate-prompt-tokens attribute: set it to false in production. When set to true, APIM estimates prompt tokens before the response is returned, enabling earlier throttling but reducing accuracy. In practice, counting actual tokens from the response is more reliable and avoids throttling requests that would have been within budget.

A common mistake is setting a single global TPM limit too low, which throttles all consumers the moment a batch job runs on any team. The better pattern is tiered limits by API product: a Developer product with a low TPM ceiling, a Standard product for normal workloads, and an Unlimited product for production pipelines that need burst capacity.

Handling 429 Responses in Calling Applications

Any application calling an APIM-fronted AI endpoint needs to handle 429 responses gracefully. APIM returns a Retry-After header indicating how many seconds until the token window resets. Well-behaved clients respect this header and back off rather than retrying immediately.

For agentic workloads with multiple pipeline steps, a 429 response midway through can leave the agent in an inconsistent state. The recommended pattern is to expose the remaining-tokens-variable-name value in a response header so the calling application can monitor its own budget and slow down proactively, rather than waiting for a hard rejection.

The Azure OpenAI token limit policy documentation covers the full attribute reference, including tokens-per-minute, counter-key, estimate-prompt-tokens, and remaining-tokens-variable-name. The llm-token-limit variant has the same interface for non-Azure backends.

What’s Next in This Azure API Management for AI Series

Part 4 covers the Token Metric policy: how to emit token usage data to Application Insights broken down by consumer dimensions, and how to use that data for internal cross-charging and spend dashboards.

Azure API Management for AI: Why Your APIs Need a Gateway

Posted on by
5

Part 1 of 7 in the “APIM for AI Workloads” series

Over the past year, I’ve been doing a lot of work with integration services, including Azure API Management and, recently, also on AI adoption: evaluating models, designing agentic architectures, and figuring out how to govern AI consumption across the organization responsibly. One thing that keeps coming up in those conversations is a question that sounds almost too basic to ask: Who is keeping track of what we’re spending on tokens?

The answer, more often than not, is nobody.

That’s the problem this series is about. AI APIs are fundamentally different from the REST APIs we’ve been managing for the past decade, and the differences matter operationally. Before we dive into the mechanics of Azure API Management policies, load balancing, and semantic caching in subsequent posts, I want to make the case for a gateway layer in front of your AI services. Before we dive into the mechanics of Azure API Management for AI workloads, policies, load balancing, and semantic caching, I want to make the case for why you need a gateway layer in front of your AI services.

Tokens Are Not Requests

Traditional API management was built around a relatively simple model: count the requests, enforce rate limits, log the traffic, and call it done. One call in, one response out. The cost model was predictable.

AI APIs broke that model completely.

When you call an Azure OpenAI endpoint, you’re not paying per request. You’re paying per token. And a token count is invisible at the API gateway layer unless you specifically instrument for it. A single call from a conversational agent might consume 500 tokens. A call from a poorly-optimized batch process might consume 50,000. Both look the same at the HTTP level: one POST, one 200 OK.

This creates a blind spot that grows dangerously as AI adoption scales across an organization. Teams start building intelligent apps: conversational agents, personalized content generators, voice assistants, copilots, and each one is independently calling AI backend services. Nobody has a view across the whole estate of what’s being consumed, by whom, and at what cost.

The diagram below shows what this looks like in practice: multiple application types hitting multiple AI providers, with token-based pricing models sitting underneath.

Azure API Management control plane between intelligent apps and AI providers showing PTU and PAYG token billing
Diagram 1: Intelligent applications on the left, AI service providers on the right, with both PTU and PAYG billing models underneath. Without a control plane in the middle, you’re flying blind.

The Three Problems Azure API Management for AI Solves

Azure API Management acts as the centralized control plane between your intelligent applications and your AI backends. It addresses three distinct categories of problems.

Performance optimization: AI model endpoints have throughput limits. Azure OpenAI Provisioned Throughput Units (PTU) give you reserved capacity at a fixed price, but cap out at a hard ceiling. Pay-as-you-go (PAYG) endpoints scale elastically but incur higher per-token costs at high volumes. Without a gateway layer, individual applications can’t know whether PTU capacity is available or saturated. A gateway can make that routing decision automatically, serving from PTU when it has headroom, falling back to PAYG when it doesn’t. That’s a meaningful cost optimization with no changes required to the calling applications.

Cost control: Tokens consumed by one team are costs borne by another team’s budget if you’re centralizing AI spend, which most organizations will do, at least initially. Without per-consumer visibility into token usage, internal cross-charging and showback are impossible. APIM’s token metric policies make this tractable by emitting token consumption data broken down by dimensions such as User ID, Subscription ID, or API product, all of which feed into Application Insights for dashboarding and alerting.

Data security: Routing AI traffic through a managed gateway gives you a single enforcement point for authentication, authorization, and policy. You can validate JWT claims, require subscription keys from API consumers, use Managed Identity to authenticate to Azure OpenAI without exposing credentials, and ensure traffic never leaves your controlled perimeter. Without a gateway, every team builds its own auth story, or more commonly, skips it.

PTU vs. PAYG: Why the Billing Model Shapes Your Architecture

Before we go further, it’s worth spending a moment on the two Azure OpenAI billing models, because they have direct architectural implications.

Provisioned Throughput Units (PTU) give you reserved capacity on a model. You pay a fixed hourly rate regardless of how many tokens you actually consume. The benefits are predictable costs and guaranteed throughput. The risk is waste if your utilization is low, and hard throttling if you exceed the provisioned limit.

Pay-as-you-go (PAYG) charges per token consumed. No upfront commitment, no capacity ceiling, but costs scale linearly with usage and can surprise you if consumption spikes.

Most production AI deployments end up using both: PTU for baseline load, where utilization is predictable, and PAYG as an overflow layer. This makes a load balancer with circuit breaking essential, which we’ll cover in Part 5 of this series.

The same logic applies beyond Azure OpenAI. APIM now supports generic LLM backends via the llm-* policy family, which means you can manage traffic to Mistral, Cohere, LLaMA, and other providers through the same control plane. The diagram below shows this architecture: APIM in the center, with load balancing across PTU and PAYG instances, token metrics flowing to Application Insights, and the full provider landscape behind it.

Azure API Management AI control plane with token limit, token metric, load balancing, semantic caching and circuit breaker policies across PTU and PAYG backends
Azure API Management as the centralized AI control plane, with performance, cost, and security governance across multiple providers and billing models.

What This Looks Like in Practice

Let me make this concrete with a scenario I’ve seen play out multiple times.

An organization deploys its first Azure OpenAI service for a conversational agent. A few months later, a second team wants to use AI for content generation. Then a third team builds an internal copilot. Each team provisions its own Azure OpenAI resource, authenticates directly, and manages its own rate limiting. There’s no visibility into combined spend. No shared capacity optimization. No centralized audit trail.

This is the point where someone in finance asks a question that nobody can answer: “How much are we spending on AI, and which team is spending what?”

Centralizing AI traffic through APIM is how you get out of that situation before it becomes a problem. The policy-based approach means you can add governance without changing anything in the calling applications. They call the APIM endpoint, APIM handles the rest.

Azure API Management for AI Workloads: What’s Coming in This Series

The next six posts will go deep on the specific capabilities that make APIM a serious AI control plane:

Each post will include the relevant policy XML, real-world sizing guidance, and the architectural decisions behind the patterns.

If you’re building AI-powered applications at scale and you’re not yet routing that traffic through a gateway, the rest of this series is for you.