# Measuring generative AI usage and cost | PointFive Research

Canonical: https://www.pointfive.co/research/finops-for-ai-the-tokenomics-frontier

How token-level usage data helps organizations understand the cost of generative AI workloads.

[All research](https://www.pointfive.co/research) 

Yael Cinamon  FinOps Analyst  January 22, 2026  2 min read

## The Blind Spot in AI Spending

Organizations deploying Generative AI have inadvertently created a "Black Box" in their cloud bills. Services like Azure OpenAI, Amazon Bedrock, and Google Vertex AI typically appear as single line items, leaving teams unable to determine which deployments or models drive spending. You cannot manage what you cannot measure.

### Aggregated Spend: The Top-Down Approach

Traditional cloud billing systems show totals without context. A $50,000 Azure OpenAI charge provides no insight into whether costs stem from high-value customer features or wasteful experimental processes in development environments.

### Tokenomics: The Bottom-Up Approach

The alternative measures cost per million tokens, individual units of model input and output. This transforms cloud expenses into Cost of Goods Sold metrics, enabling meaningful scalability insights.

## PointFive's Allocation Magic

PointFive addresses this through its Data Fabric model, which ingests telemetry across AI and cloud infrastructure. The platform decomposes aggregated costs into granular, deployment-level insights.

## Case Study: Eliminating the Availability Premium

A development deployment used Provisioned Throughput (PTU), designed for high-traffic production environments, despite showing only 0.6% average utilization in non-production settings.

Result:  Shifting to on-demand pricing achieved a  99% cost reduction  for that resource.

## Case Study: Suboptimal Model Selection

A customer ran an outdated reasoning model despite newer, more efficient alternatives being available. PointFive demonstrated the newer model's lower cost-per-request through visual breakdowns of input token costs and cached inputs.

Result:  An  86% savings  by updating to a more efficient model version.

## Key Optimization Workflows

The platform enables several approaches:

- Automated Attribution:  Mapping tokens to specific business units without manual tagging

- Capacity Right-Sizing:  Identifying unused provisioned commitments

- Performance-to-Price Optimization:  Benchmarking model versions across vendors

- Operational Intelligence:  Surfacing usage patterns for infrastructure scaling

## The Bottom Line

AI efficiency requires moving beyond total spending toward tokenomics-based management. When your dashboard reflects your actual data assets, "Allocation Magic" becomes your most powerful tool for scaling AI sustainably.

---
Source: the public page above. Product screenshots and illustrative interfaces are examples, not live customer data.