---
title: "How to read CloudMonitor's AI cost reports"
canonical: "https://cloudmonitor.ai/docs/how-to/how-interpret-ai-token-cost-reports/"
description: "Walks through the five KPIs on the AI cost dashboard — total spend, cost per developer, model mix, cache hit rate, batch percentage — and where to drill in."
---

This guide walks through the AI cost dashboard CloudMonitor exposes once at least one AI provider connector is configured (GitHub Copilot, OpenAI, Anthropic Claude, or Cursor). The dashboard appears in two places: the **AI cost** page in your Power BI workspace, and the **AI providers** tab of the Admin App.

## Five KPIs at the top

The header strip shows the same five numbers in both surfaces.

### 1. Total AI spend

Sum of dollar cost across every connected provider for the selected period (default: month-to-date). Click the figure to drill into provider → account → workspace → model.

### 2. Cost per developer

Total spend divided by **distinct active developers** during the period. CloudMonitor counts a developer once even if they appear in multiple providers — say Copilot for IDE completions, Claude Code for terminal work, and Cursor for editor chat. The figure normalizes for the team size you have today, not the team size on the first of the month.

### 3. Model mix

Share of dollar spend by model tier: **frontier**, **mid**, **entry**. Frontier means Opus, GPT-5, and Gemini Ultra-tier models. Entry means Haiku, GPT-5 Nano, and Gemini Flash. Rule of thumb: every percentage point of frontier spend you shift to entry without quality loss is roughly 10× cheaper.

### 4. Cache hit rate

Share of input tokens served from prompt cache (Anthropic and OpenAI both expose this). Anthropic charges cache reads at 0.1× the standard input rate; OpenAI's cached input is similarly heavily discounted. A cache hit rate above 30% is healthy on workflows with long system prompts.

### 5. Batch percentage

Share of spend run through batch APIs. Both OpenAI and Anthropic offer 50% off on batch. If you run asynchronous work — back-fills, evals, document summarization — and batch percentage is under 10%, you're leaving money on the table.

## Drill-downs

Below the header strip, the dashboard has four tabs.

- **Spend** — provider, account, workspace, model, day. Date-range slicer at the top.
- **Developers** — per-developer cost, model mix, acceptance rate (Copilot and Cursor). Top spenders and unused seats sit on the same page.
- **Efficiency** — cache hit rate, batch percentage, and **context-cliff exposure** (requests over 200 K tokens that trigger 2× input pricing on most providers).
- **Forecast** — month-end projection vs budget per cost center.

## Anomaly alerts

CloudMonitor opens an alert when daily spend exceeds the 30-day mean by **3 sigma** for any `(provider, account, model)` combination. Alerts land in the Microsoft Teams channel you wired up at installation. Acknowledge from the Admin App **Anomalies** tab — acknowledged anomalies stop notifying but stay in the audit log.

## Allocation tagging

To attribute AI spend to a cost center, tag the connector account at setup or open **Settings** → **AI providers** → choose the connector → **Allocation**. Tags propagate to the cost center column on the Spend tab and feed the Allocation Coverage KPI on the Quantify domain dashboard.

For finer-grained per-PR or per-feature attribution, see the LiteLLM proxy integration (currently in private preview).

## What's next

- [How to connect GitHub Copilot](/docs/how-to/how-connect-github-copilot/)
- [How to connect OpenAI](/docs/how-to/how-connect-openai/)
- [How to connect Anthropic Claude](/docs/how-to/how-connect-anthropic-claude/)
- [How to connect Cursor](/docs/how-to/how-connect-cursor/)