# LLM Cost Audit Service

Intelligence Per Watt audits production LLM spend for teams using OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI, Cursor, Claude Code, Codex, LangChain, or an LLM gateway.

The goal is simple: find the fastest credible reduction in cost per successful AI workflow without asking for secrets, production credentials, raw prompts, raw outputs, customer data, or production write access.

## Who This Is For

This is most useful when at least one is true:

- Your AI bill is above $2,000 per month and growing.
- Claude, OpenAI, Gemini, or Bedrock spend is rising faster than revenue.
- You run agents, RAG, support automation, document extraction, voice, coding agents, or eval jobs in production.
- Finance or leadership wants a plain-English answer on AI gross margin.
- Engineering suspects waste but needs a ranked, dollar-backed plan.

## What Gets Reviewed

The diagnostic can start from sanitized exports or screenshots:

- spend by provider and model
- request volume by feature or workflow
- input, output, cached, and reasoning token patterns
- retry and failure rates
- batchable versus interactive work
- long-context or repeated retrieval payloads
- eval and test-loop cost
- cost per customer, account, workflow, or successful task when available

## Common Findings

Most early savings come from a few patterns:

- frontier models handling narrow or verifiable work
- repeated system prompts, tool schemas, or RAG context
- missing cache boundaries for stable prefixes
- retry loops and tool-call amplification
- eval sweeps that run too often or on unchanged paths
- background work running synchronously instead of batch
- unpriced heavy users creating negative gross margin

## Offers

- **$99 cost leak triage:** concise written triage from sanitized usage numbers.
- **$299 24-hour LLM bill roast:** written diagnosis of visible cost leaks, routing candidates, cache/batch opportunities, and whether a deeper audit is worth buying.
- **$1,500 founding 72-hour audit:** fixed-scope audit for qualified case-study slots.
- **$2,500 standard 72-hour audit:** finance-readable baseline and engineering-readable patch plan.

Start here:

https://www.intelligenceperwatt.com/roast-my-llm-bill

Sample output:

https://www.intelligenceperwatt.com/sample-bill-roast.md

https://www.intelligenceperwatt.com/sample-audit-report.md

## Data Policy

Do not send API keys, credentials, raw prompts, raw outputs, customer data, account IDs, or unrelated personal data. Sanitized provider screenshots and aggregate exports are enough for the first pass.