AI Spend Audit

Audit AI tool spend against actual usage and produce a reallocation memo. The number that matters is cost per active user, not cost per seat — and most orgs are simultaneously overspending on shelfware and underfunding their power users.

What you need from the user

Ask for three things:

1. Vendor list and seat counts. Every AI tool the org pays for. Include bundled licenses (Copilot inside M365, Gemini inside Workspace). If they don’t know the full list, ask them to check: the Microsoft admin center, any ChatGPT/Claude/Gemini enterprise dashboards, and the last three months of expense reports for personal AI tool reimbursements.
2. Usage data. Monthly active users per tool, or login frequency if that’s all they have. Most enterprise dashboards export this as CSV. If they can’t get usage data, ask for their best estimate of how many people actually open each tool weekly.
3. Monthly cost per tool. The invoice amount or per-seat price times seat count. Include API spend if any.

If they have a CSV or spreadsheet, ask them to paste or upload it. Work with whatever format they give you.

How to analyze the spend

Run through these checks in order:

Compute cost per active user. Total monthly cost divided by monthly active users, per tool. This is the number that matters, not cost per seat. A $30/seat tool with 20% usage costs $150 per active user. Write this number next to every tool.

Flag idle seats. Any tool where fewer than 40% of seats logged in last month is a problem. Below 20% is shelfware. Note these for the memo.

Find the lopsided distribution. In most orgs, the top 10-15% of users produce 80-90% of the value. If usage data has engagement tiers (messages sent, sessions per week), call out the power users vs. the long tail of near-zero usage.

Check for shadow AI. Ask if anyone is expensing personal AI subscriptions or putting API spend on a corporate card outside the approved budget. Shadow AI means the approved tool is worse than what people can get on their own. This is a procurement signal, not a compliance problem.

Benchmark the total. For knowledge-work orgs in 2026, total AI spend should land between $40 and $120 per employee per month, average closer to $80. Below $40, power users are likely underfunded. Above $120, idle seats are inflating the number. State where they fall and whether the shape underneath justifies it.

Check allocation shape. Good spend follows roughly 70/20/10: 70% on the tools power users actually open (typically Claude or ChatGPT), 20% on workspace AI for the broad middle (Copilot, Gemini in Workspace), 10% on API budgets for engineers and operators building internal tools. Compare their actual allocation to this benchmark.

Spot common traps. Flag if you see any of these:

• Bundled Copilot nobody opens (paid penetration across enterprise is under 4%, accuracy NPS negative for nine months running)
• Flat per-seat distribution where everyone gets the same tier regardless of usage
• Multi-year commit on a single vendor when the market is still moving
• Training budget counted as AI budget with no measurable usage impact
• Agent platform spend before any pilot has produced results

What to output

Produce a one-page reallocation memo with these sections:

Current state. Two to three sentences: total monthly AI spend, number of tools, cost per active user across the portfolio.

What to cut. Specific tools or seat blocks to eliminate, with dollar savings. Default to cutting the bottom 30% of seats on any tool with under 40% active usage.

What to consolidate. If they’re paying for overlapping tools (e.g., both Copilot and ChatGPT for the same population), recommend consolidating to the one with higher usage. Don’t consolidate to a single vendor for procurement convenience. Two to three tools is the right number for most orgs.

What to fund. Where the freed dollars should go. Typically: upgrade power users to higher tiers, add API budget for engineering, give the top 15% of users access to multiple tools instead of rationing.

The number for your CFO. Three sentences they can use in the next budget conversation. Format: “We’re spending $X/month on AI. $Y of that is producing no measurable usage. Reallocating to active users saves $Z/month or funds [specific capability].”

Keep the memo direct.