CloudCostChefs

Ep18 is out. cloudcostchefs.com/podcast #CloudCostChefs #FinOps

The automation watched the meter, not the money: AWS's own cost alarms fired on phantom ESTIMATES — a $1 tripwire, a $233M phantom. Same week, Fable 5's $10/$50 card got arbitraged in days. The screen is not the truth.

Copilot Credits pool at the tenant level, bill in arrears like a consumption meter, and (once linked to the right billing account) decrement your Azure MACC. That's not a seat fee. Put it in the same review as your Azure meters.

Your M365 invoice just grew a cloud-spend-shaped line, and most finance teams haven't clocked it yet.

Most reviews of agent spend are licence audits wearing a FinOps hat. The fix is three lines. One: which agents ran, not which team owns the licence. Two: cost per completed task, not per seat. Three: what happened the week nobody watched it.

Everything on the counter is free and vendor-neutral, no sponsor deals behind any of it: cloudcostchefs.com/?utm_source=x&…

Saturday admin: if your AI cost tracking lives in one person's head and a spreadsheet nobody updates, that's the whole problem already. We keep the free FinOps tools and guides on the counter. No login wall, no sales call. 🍳

Ep17: the rate card is dead — read the meter. cloudcostchefs.com/podcast #CloudCostChefs #FinOps

IBM lost $67B in a day because customers hoarded RAM ahead of price hikes. AWS now schedules GPU repricing like a Fed meeting. And your AI bill moved into the fine print.

Twenty seats, one power user. Blend that into cost per user per month and the dashboard says everything is fine. Per-user averages are where AI spend anomalies go to hide. Curious what FinOps teams track instead of the blend. Drop yours.

Every chargeback model we've tried breaks on the same question: whose spend is this? Not price, tagging. If nobody can name the cost centre behind a seat or an agent run, the invoice stays unallocated no matter how granular the vendor's report gets. #FinOps

If your metered line is under 5% of the Copilot invoice (our working line, not a standard), do not build chargeback machinery for a rounding error. Run showback for two quarters and re-check monthly. The model should cost less than the waste it allocates.

Who's holding the leverage? Ep16 🍳 cloudcostchefs.com/podcast

Tesla capped employee AI spend at $200/week — exemption: the CEO's other AI company. Microsoft is swapping Anthropic out of Office. Token pricing just split in two: $1/M at the bottom, $50/M at the top.

We built the full model: The Copilot Chargeback Playbook. 10-tab workbook with formulas live in the cells, one-page policy finance signs, 12-question method checklist, 40-minute close runbook. $49. store.kesslernity.com/l/copilotcharg…

If your metered line is under 5% of the Copilot invoice (our working line, not a standard), do not build chargeback machinery for a rounding error. Run showback for two quarters and re-check monthly. The model should cost less than the waste it allocates.

The June 2026 Copilot billing change is an additive metered layer: $0.01 per Cowork credit (Microsoft list) on top of the flat $30 seat. Your seat price did not change. Your allocation model still broke: one invoice now needs two methods, and headcount math covers only one.