For CFOs and controllers: recompute your metered AI invoices against the contract rate card and your own usage logs, then draft the credit request. 12 prompts, no recovery firm on contingency.
Never used Claude? It is free and takes 30 seconds to open. Copy the instruction below, paste it into Claude, and it reads this page and walks you through everything, one question at a time.
A short instruction plus a link to this page lands on your clipboard. First copy asks for your email once. That unlocks every button across the whole library.
Free account, no card, 30 seconds. This tab stays open so you can come back.
Open claude.ai ↗Claude reads this page, asks which computer you are on, then guides you step by step until it works. If anything errors, tell Claude what you see, and it fixes it with you.
If you can browse the web, open and read this page in full first, it has the complete guide and every prompt you will run (the vault is under the-vault anchor): https://consultance.ai/library/ai-vendor-invoice-audit#the-vault . If you cannot open links, tell me and I will paste the page in, do not guess the prompts. You are my setup guide for the AI Vendor Invoice Audit Kit. Your job is to get me from zero to a finished recompute of one real AI vendor invoice, one step at a time. House style: calm, practical, one step at a time. Never dump the whole process at me. Never assume I know an acronym, define it the first time you use it. Do not tell me something is "not possible" because it is not a coding task, this is not a Terminal install and nothing here needs a command line. **This is NOT a Terminal install.** There is nothing to clone, install, or configure. Everything happens inside the Claude web app or desktop app, in a Project, using files I already have. If I ask you for install commands, tell me plainly that there are none and move me to the file gathering step instead. Ask me ONE question first, then wait for my answer before saying anything else: **"Where do your AI vendor invoices and usage records live right now?"** Offer these options and let me pick: (A) I download PDF invoices from the vendor's billing console and that is about it (B) I have invoices plus a usage or activity export, usually CSV, from the vendor's console (C) It runs through a cloud provider (AWS Bedrock, Azure OpenAI, Google Vertex) so it is inside the cloud bill (D) A FinOps or spend tool sits in the middle (Vantage, CloudZero, Cloudability, Finout, or similar) (E) Honestly, I am not sure, the bill just gets paid After I answer, walk me through setup in this order, one step per message, waiting for me each time: **Step 1. Make the Project.** In Claude, click Projects in the left sidebar, then Create Project. Name it "AI Vendor Invoice Audit". Explain in one line that a Project keeps files and instructions together across chats, and that on a Team or Enterprise plan the files stay inside my own company tenant. Nothing gets uploaded to consultance.ai, stored by us, or seen by us. Success state: I can see an empty Project with a knowledge panel on the right. **Step 2. Gather three things.** Tell me exactly what to fetch, tailored to the answer I gave you: - The invoice for one period. One month. Not the year. - The contract, order form, or if I am on standard published pricing, say so plainly, we will use list price and label it as an assumption. - A usage export covering the same period, at the finest grain the vendor gives me. For whichever option I picked, name the specific place to click. For (A), tell me where the vendor console usually exposes usage exports and that if there genuinely is none, we run a reduced scope audit and I should know that up front. For (C), point me at the cloud provider's cost and usage report rather than the summary bill, because the summary bill cannot be recomputed. For (D), tell me to export from the tool AND get the raw vendor invoice, because the tool's view is already an interpretation. If I come back and say a file does not exist, do not stall. Tell me what the audit can still do without it and what it cannot, then continue. **Step 3. Load the files.** Drag them into the Project knowledge panel. If a file is over the size limit, tell me to filter the usage export to the single period first rather than splitting it randomly. Success state: three files listed in the knowledge panel. **Step 4. Run prompt 01.** Paste prompt 01 from the vault, the onboarding router. It will ask me four blocks: audit scope, where my data lives, engagement context, and the output bar. Tell me to answer honestly rather than aspirationally, especially about whether I hold the contract. Success state: Claude reads my four blocks back to me and waits. Then run the first session drill with me, in this order, not all at once: **Drill 1, prompt 02.** Build the billing truth table from the contract. What good output looks like: one row per chargeable unit with a clause reference against each rate, plus a list of terms the contract is SILENT on. If every rate comes back with no clause reference, my contract probably is not in the Project or it is a summary rather than the order form. Check that before continuing. **Drill 2, prompt 03.** Normalize the invoice. What good output looks like: the line items foot to the invoice total. If they do not tie, stop there, that is a real finding on its own and worth an email before anything else. **Drill 3, prompt 04.** The recompute. This is the one that matters. What good output looks like: a comparison table with billed against recomputed, and crucially a stated match rate, the percent of billed dollars traceable to my own usage records. Tell me plainly that a low match rate is not a failure of the prompt, it is a measurement of my own visibility, and it is the single most useful number I will get this week. **Review check before I trust any of it:** every variance line should name the invoice line and the contract clause behind it. Anything without both is an observation, not a claim. And I do not send anything to a vendor until prompt 08, the independent re derivation, returns the word RECONCILED. Tell me that twice, once now and once when I get to it. Anti patterns for you to avoid: - Do NOT tell me this is "not possible" without a code integration. It runs on files and a chat window. - Do NOT invent what my contract says. If a rate is not in the document, it gets labeled ASSUMPTION. - Do NOT skip ahead to the credit request letter because it looks like the fun part. It is worthless without the recompute behind it. - Do NOT congratulate me on a variance before prompt 08 has reconciled it. Bonus path, never required: the vendor's own pricing page is useful for sanity checking list rates, and FinOps Foundation publishes free material on cloud cost allocation practice. Neither is needed to run the vault. Do not send me off to read them mid setup. Start now with the one question about where my invoices and usage records live. Nothing else in your first message.
Grab the whole pack as one file, or tap any prompt below to copy it on its own. Placeholders that look like {{THIS}} get swapped for your own numbers — and if you ran Step 1, Claude fills them in for you.
For CFOs and controllers: recompute your metered AI invoices against the contract rate card and your own usage logs, then draft the credit request
The setup rail and every prompt above are free and stay free. The cost is your time, and the risk of wiring it wrong on live data.
Back to the prompts ↑This kit is about 70% of the build & Consultance wires the last 30% into production: your billing and usage feeds pulled automatically so the recompute runs every month instead of once, exception alerts on model rate drift and failed run charges, an audit trail your auditor accepts on every recomputation, and multi entity allocation across a group or a portfolio. Reply "wire it" for a 30-minute slot.
Book a build call →Free to use on your own invoices inside your own Claude tenant. First pass billing analysis and decision support, not legal advice, not an audit opinion, and not an assertion that any named vendor has misbilled you. Claude cannot see the vendor's meter, it checks your contract against your own logs. A named human owns the credit request and the signature.
Want this wired into your stack instead of running it yourself? That is our AI deal desk and finance automation service.
The AI tools, launches, and shifts that actually matter, in plain English. New library drops the moment they land.
More like this
For CFOs and PE deal teams: run six Big 4 grade deliverables yourself with 12 prompts, including the quality of earnings pass and the LP reporting pack, plus the partner questions the model skips.
For PE deal teams, CFOs, and IC members: 11 Claude prompts that run a second read on a seller's quality of earnings, test every add-back against the source, catch the recurring cost dressed up as one time, and rebuild EBITDA the way the buyer should see it before it clears into the price.
For deal teams and corp dev building the buy-side merger model: 12 prompts that run accretion or dilution, the funding mix, goodwill, synergies, and the breakeven price where the deal stops adding to EPS, in an afternoon instead of a two-week associate build.
manual now, installed when you're ready
The free version above. It runs this read on a file you paste in, in your own Claude. Yours to keep.
The same read, wired into your cfo and back office stack, running the second a report lands. Nothing to upload, no new login.
See the installed versionThe person drowning in this every week is the one who'll actually want it.
AI Vendor Invoice Audit Kit is a finance and data build in the consultance.ai AI Build Library. For CFOs and controllers: recompute your metered AI invoices against the contract rate card and your own usage logs, then draft the credit request. 12 prompts, no recovery firm on contingency. It fits CFOs, controllers, FP&A leads, internal audit, and PE portfolio finance teams carrying metered AI and LLM spend on Anthropic, OpenAI, AWS Bedrock, Azure OpenAI or Google Vertex, who sign the invoice every month without recomputing it. Setup difficulty is Medium, with 5 plain-English steps.
For CFOs and controllers: recompute your metered AI invoices against the contract rate card and your own usage logs, then draft the credit request. 12 prompts, no recovery firm on contingency.
It fits CFOs, controllers, FP&A leads, internal audit, and PE portfolio finance teams carrying metered AI and LLM spend on Anthropic, OpenAI, AWS Bedrock, Azure OpenAI or Google Vertex, who sign the invoice every month without recomputing it.
Medium to set up — one guided setup instruction covering 5 plain-English steps, plus 12 ready-to-run prompts on the resource page.
This kit is about 70% of the build & Consultance wires the last 30% into production: your billing and usage feeds pulled automatically so the recompute runs every month instead of once, exception alerts on model rate drift and failed run charges, an audit trail your auditor accepts on every recomputation, and multi entity allocation across a group or a portfolio. Reply "wire it" for a 30-minute slot.
Free to use on your own invoices inside your own Claude tenant. First pass billing analysis and decision support, not legal advice, not an audit opinion, and not an assertion that any named vendor has misbilled you. Claude cannot see the vendor's meter, it checks your contract against your own logs. A named human owns the credit request and the signature.
AI Vendor Invoice Audit Kit is the starting point. On a free AI audit we map where it fits your stack and what consultance.ai would build around it.
This build comes from our AI consulting and AI implementation practice — see the full AI in finance guide and how we work with CFO teams.