Five files for engineering, platform and FinOps teams running LLM features in production. You open them, put your own numbers in, and use what comes out. $49, licensed per team.
Project 12 months of LLM spend from your real workload: requests, token sizes, cache hit rates, batch share, growth. Compare 24 current models across 8 providers side by side, get AI cost per user and cost as a share of revenue. 140 formulas, every one recalculates from your inputs, nothing hardcoded.
Two templates for settling which team pays for what. One sets out a four-level allocation hierarchy with targets. The other is a tagging standard: the mandatory keys, where the tags physically live on Bedrock, Azure OpenAI and Vertex, and the enforcement rules, because an unenforced tagging standard stops describing reality within a quarter.
Starter queries for getting AI spend out of CUR 2.0 and Athena, Azure Cost Management exports and Log Analytics, and the BigQuery billing export: by model, by team, input against output tokens, and a daily trend for alerting. Billing schemas differ between accounts and versions, so each query notes which columns to check against yours. There's also an anomaly detection checklist organised around the common failure modes: retry storms, context bloat, silent routing to an expensive model, agent loops, leaked keys, forgotten batch jobs, and vendor price changes.
Turns per-team token usage into allocated cost: direct spend by model, shared platform cost spread pro-rata, and a summary page for finance. The input numbers come straight from the monitoring queries.
25 statements across five domains, scored 0 to 4, banded Crawl/Walk/Run, with a chart. Takes about twenty minutes and gives you a baseline plus a view of what to fix first.
Two tools that run on your actual vendor exports (OpenAI, Anthropic, xAI, DeepSeek, Mistral, Perplexity, or anything via the editable generic profile). The converter turns usage and cost exports into a dataset structured to the FinOps Open Cost and Usage Specification (FOCUS™) at the revision you choose, 1.0 through 1.4, so it can join the FOCUS data you already get natively from AWS, Azure and GCP; one row per charge component, ListCost computed from this site's verified prices, official FOCUS Validator referenced for checking output. The what-if rerater takes the same export and prices your real token volumes across all 24 models, showing what the identical usage would have cost elsewhere and the delta against what you actually paid, with the assumptions printed (list prices, cache treatment, tokenizer variance) rather than hidden. Standard library Python, no dependencies. Independent of, and not certified or endorsed by, the FOCUS project.
In the State of FinOps 2026 survey, 98% of organisations said they now manage AI spend in some form, AI cost management came out as the most sought-after skill, and granular AI spend monitoring was the most requested capability. Most teams are working all of that out from scratch. The kit is so you don't have to start from a blank spreadsheet.
Platform and engineering leads shipping LLM features, FinOps practitioners who've been handed AI costs as a new remit, and CTOs at scale-ups who want the numbers before the board asks for them. It isn't aimed at GPU cluster procurement or training-run economics at research-lab scale; it covers inference-led production workloads on managed APIs.
Prices in the kit were checked against vendor pages on 12 August 2026 and will drift, so every pricing table is editable and dated. Cloud billing schemas vary by account and version, and the queries note what to check against yours. The 200+ spreadsheet formulas are machine-verified to recalculate cleanly. Independently produced with AI assistance and human direction. No vendor affiliations, no affiliate links.