The metered line has landed on your Copilot invoice, and finance wants a name next to it. Your per-seat split has no answer, because a credit pool is not a seat count, and "evenly" is not a method. We built the model that is.First, the framing, because it gets mangled in most orgs: the June 2026 change is an additive metered layer. Cowork credits and agent runs bill at $0.01 per credit, on top of the flat $30 per seat (both Microsoft list, as of June 2026). The seat price did not change. What changed is that one invoice now needs two allocation methods, and headcount logic covers only one of them.The playbook builds the second method end to end, in one afternoon, in three steps. Run the 12-question checklist and land on a defensible allocation method in 20 minutes. Populate the workbook with your real invoice lines until the Reconciliation tab ties to the invoice within $1. Put the one-page policy in front of your finance partner and get it signed. From month two on, the close is a 40-minute runbook.The primary worked example is Tallgrass Foods, an invented mid-size manufacturer (every org name, headcount and credit volume in it is invented for illustration, and labeled that way throughout): an $80,000 invoice, $54,000 of flat seats plus $26,000 of credits across four business units, arithmetic shown line by line. The teaching point sits in the workbook's comparison view: split that credit pool by seats instead of consumption and Supply Chain gets subsidized by about $2,474 a month. That is the argument for consumption math, stated in dollars.Six files. Every bullet under "What's inside" names the actual file in the download.Who it's for: FinOps practitioners and cost analysts at M365 enterprises who already run tag-based chargeback for Azure or AWS and can read a FOCUS or CUR export without help. Nothing in here explains what chargeback, showback or tagging is. The Copilot-specific billing mechanics get the full treatment, because that part is genuinely new: what a credit is, where the meters land on the invoice, what maps to a named user and what belongs to a shared agent.Who it's not for: this covers the M365 Copilot invoice only. If you need Azure OpenAI or Foundry token chargeback, the meters and exports are different, and this playbook will not stretch to them. If your problem is seat cost (license optimization, reclaiming unused seats, the ROI argument), that is The Real Cost of Copilot ($29), not this. And if you want software (dashboards, automated pipelines, a connector), there is none here. The workbook is the tool, deliberately.How it was made: drafted with AI agents inside a gated pipeline, then checked by machines that have no feelings to spare. Every file passed a deterministic slop scanner and a build checker. The workbook is produced by a script that runs 57 self-checks, asserting formula strings cell by cell and recomputing the full allocation independently in Python: variance $0.00. Every price carries an "as of" date. A human reviewed the result, and each gate's output is logged, dated, in the product's STATUS file.What's inside- The playbook itself, a 22-page typeset PDF: nine sections running from how to read the metered invoice lines, through the four allocation methods that survive finance review, to two worked examples at $80,000 and $3,800 a month (both invented illustrations, arithmetic shown line by line). - The same playbook as an EPUB, for the commute before the finance meeting.- The 10-tab allocation workbook, with the formulas live in the cells (`credits_used * 0.01`, never a pasted result). It ships pre-populated with the Tallgrass worked example, so it runs as a model out of the box; swap your own invoice into the three input tabs and every allocation tab recomputes. The Reconciliation tab ties every allocated dollar back to the invoice and flags any variance over $1, and the Metered Allocation tab carries the headcount-distortion comparison: what each BU would pay under a seat split versus consumption.- The one-page policy template, blank plus the Tallgrass version filled in as an example, built around the four fields finance actually checks, with a 15-minute sign-off conversation script. - The 12-question decision checklist: yes/no questions scored toward one of the four methods, with the 5% showback gate up front (a working assumption, not a standard, and marked as such). Twenty minutes to a written method decision finance can hold you to. - The monthly close runbook: six steps, each with a time budget and a done-when condition, 40 minutes total against the 45-minute target, plus rules for five edge cases, from a runaway shared agent to a list price moving mid-quarter.FAQQ1: What exactly do I get?Six files in one download: the playbook as a 22-page typeset PDF and as an EPUB, the 10-tab Excel workbook pre-populated with the worked example, the one-page policy template (blank plus a filled example), the 12-question decision checklist and the monthly close runbook as Markdown. No videos, no login, no drip sequence.Q2: Will it work in my setup?You need three inputs: the M365 invoice, the Copilot usage export from the admin center and a seat roster by business unit. The workbook uses plain Excel formulas: no macros, no external connections. It works at any BU count. If your export cannot attribute every credit to a user, that is expected; the shared-pool rule and the seat-weighted fallback exist for exactly that gap, and where the admin-center export's attribution ends, the playbook says so, with a date, and labels the assumption.Q3: How long does it take?One afternoon for the build: 20 minutes on the method checklist, the rest populating the workbook and filling in the policy. From month two, the close runs to a time-budgeted runbook: 40 minutes total. And if your metered line is under 5% of the invoice, the checklist's first gate will tell you to run showback for two quarters instead, which is faster still.Q4: Is this just AI-generated fluff?It was drafted with AI agents, and we would rather tell you how than pretend otherwise. The draft went through a gated pipeline: a deterministic slop scanner on every prose file, a build checker on every deliverable, and a workbook build script that runs 57 self-checks, asserting formula strings cell by cell and recomputing the full allocation independently in Python (variance: $0.00). Every price carries an "as of" date. The worked-example orgs are invented and labeled as invented everywhere they appear. A human reviewed the result, and each gate's output is logged in the product's STATUS file. Fluff does not survive that pipeline, and arithmetic errors do not either.Q5: What about refunds and licensing?One purchase covers internal use at your org: run the model, share the BU statements and the signed policy inside the company, adapt everything to your setup. Do not redistribute the files themselves. Full terms are at kesslernity.com/license. It is a Gumroad purchase, so refunds go through Gumroad; if the workbook will not reconcile against your invoice, tell us anyway, because that is a bug we want to hear about.