💹AI Analytics

Easy visualization summarized AI usage & cost across a number of different metrics

AI Analytics

Revenium's AI Analytics module gives teams complete visibility into the cost, performance, and behavior of every AI transaction—down to the token level—across all providers, models, agents, and use cases.

Our platform automatically captures standard fields returned by AI providers (e.g., OpenAI, Anthropic, etc.), and customers can optionally enrich transactions with custom metadata (e.g., product, customer, task type, agent) to enable fine-grained grouping and reporting.

This system is purpose-built to help developers, analysts, and budget owners understand how AI workloads are behaving and what they’re costing—in real time and over time.


Transaction-Level Visibility

Every AI call is captured with full fidelity, including cost, latency, token breakdowns, throughput, mediation latency, and more.

  • Input/output token counts and pricing

  • Completion timing and time-to-first-token

  • Custom identifiers: task type, model source, subscription, agent, credential, user ID

  • Response quality scores (optional)

  • Performance indicators: tokens/sec, streaming, retry behavior

Example of metadata available on every AI transaction in Revenium

AI Analytics Overview

You can group, filter, and compare costs and performance across:

  • Providers

  • Models

  • Agents

  • Products

  • Customers

  • Subscriptions

  • Task types

  • Credentials

The chart configurator makes it easy to build custom dashboards that answer questions like:

  • “Which product is costing us the most per request?”

  • “How does GPT-4 latency compare to Claude for summarization?”

  • “Which API keys or agents are responsible for recent cost spikes?”

  • “How is token usage trending across months or quarters?”


Summary Dashboard

Provides a top-level view across providers and models, including:

  • Total cost by provider

  • Average cost per request

  • Token throughput trends

  • Cost by model

  • Cost by API key

Ananlytics Summary Page

Charts help visualize trends—but alerts help you act on them in real time. Revenium supports configurable cost and performance alerts so you can detect anomalies before they turn into surprises.

Set alerts on metrics like total cost, cost per transaction, token throughput, and error rates—scoped to specific agents, models, customers, or products.

📘 Learn how to configure alerts →

Customer View

Analyze usage and cost breakdowns by customer, organization, or subscription. This is especially useful in multi-tenant environments or B2B SaaS integrations.

Customer Profitability & Cost Analytics

Product View

Breaks down average cost per transaction and overall profitability by product, helping teams track margin impacts from changes in model selection or user behavior.

  • Cost vs revenue per product

  • Average cost per transaction by product

Product Cost & Profitability Analytics

Agent View

Ideal for organizations using agents or microservices powered by LLMs. Tracks:

  • Total and per-agent cost over time

  • Specific agent cost trends that help you see if your agents are getting more or less efficient over time

  • Task performance: time per task, number completed


Task-Level Insights

Compare AI vendor performance and cost across different task types (e.g., summarization, classification, extraction).

  • Average time to complete per task

  • Cost per task by model or vendor allows you to see how costs & performance for specific tasks vary by provider and model


Custom Metadata = Custom Reporting

To unlock full value, customers can pass additional metadata fields on each transaction.

📘 Read the full usage metadata reference guide


Example Use Cases

  • Flag high-cost models in specific agents (avg cost > $0.30/request)

  • Track token consumption trends by customer, agent, or team

  • Benchmark throughput and latency across AI providers

  • Identify low-margin or unprofitable tasks

  • Immediately identify regressions in cost per transaction or cost per task after new code deployments updates


Last updated

Was this helpful?