AI Waste Report

Day 21 of 31 · 6 workflows across OpenAI, Anthropic and Google

Spend to date

$1,384

Forecast this month

$2,043

Estimated monthly waste

$695

34% of spend · recoverable

AI Waste Score

61/100D

Lower is better

Top 3 ways to recover $377/mo

1Output length

$175/mo

Constrain output length in "Code review bot"

Average output is 2,600 tokens where ~1,900 would satisfy the task. Add an explicit output contract (max length / JSON schema) to cut the excess.

easymedium confidence
2Caching

$101/mo

Cache the repeated context in "Code review bot"

~2,500 tokens of identical system prompt / instructions are re-sent on every one of 3,200 calls. Split stable from volatile context and enable Anthropic prompt caching to pay ~10% for the repeated part.

mediumhigh confidence
3Model routing

$101/mo

Route "Support ticket triage" from GPT-5 to GPT-5 mini

This is a low-risk, high-volume task on a frontier model. Move ~80% of calls to GPT-5 mini and keep GPT-5 as a fallback for the hardest cases.

easyhigh confidence

Spend by workflow

Code review botClaude Opus 4.8$1,200 · 59%
Docs Q&A (RAG)Gemini 2.5 Pro$308 · 15%
Support ticket triageGPT-5$229 · 11%
Invoice data extractionClaude Sonnet 4.6$200 · 10%
Internal agent experimentsGPT-5$80 · 4%
Marketing copy generationGPT-5$26 · 1%

Spend by model

Claude Opus 4.8Anthropic$1,200 · 59%
GPT-5OpenAI$335 · 16%
Gemini 2.5 ProGoogle$308 · 15%
Claude Sonnet 4.6Anthropic$200 · 10%

Spend by provider

Anthropic$1,400 · 69%
OpenAI$335 · 16%
Google$308 · 15%

All 15 findings

RecommendationTypeDifficultySavings / mo
Constrain output length in "Code review bot"Output lengtheasy$175
Cache the repeated context in "Code review bot"Cachingmedium$101
Route "Support ticket triage" from GPT-5 to GPT-5 miniModel routingeasy$101
Constrain output length in "Support ticket triage"Output lengtheasy$58
Constrain output length in "Docs Q&A (RAG)"Output lengtheasy$56
Cache the repeated context in "Support ticket triage"Cachingmedium$44
Cache the repeated context in "Docs Q&A (RAG)"Cachingmedium$35
Constrain output length in "Invoice data extraction"Output lengtheasy$28
Cut the 14% retry rate in "Invoice data extraction"Retriesmedium$25
Constrain output length in "Internal agent experiments"Output lengtheasy$20
Cache the repeated context in "Invoice data extraction"Cachingmedium$18
Route "Marketing copy generation" from GPT-5 to GPT-5 miniModel routingeasy$12
Constrain output length in "Marketing copy generation"Output lengtheasy$10
Cut the 12% retry rate in "Internal agent experiments"Retriesmedium$9
Cache the repeated context in "Internal agent experiments"Cachingmedium$4

This is sample data. See it for your own spend.

Paste or upload a month of usage and get this report for your real workflows in under a minute.

Prices illustrative. Never paste API keys — this report needs usage numbers, not secrets.