Turn Microsoft 365 Cost Management exports into an interactive Cowork and Work IQ consumption report, with service breakdowns, spending-limit analysis, forecasts, and evidence-backed recommendations.

Cowork & Work IQ Consumption Advisor

Command: /analyze-cowork-consumption · Platform: Cowork

Copilot Cowork and the Work IQ API are billed in Copilot Credits. The Microsoft 365 admin center shows what was consumed - but tab by tab, as snapshots, with no join between who, which policy, which group, how many tasks and what it cost. This skill takes the five Cowork / Work IQ related exports an admin already has and turns them into one interactive report an executive can read in two minutes.

Why

Copilot Cowork and the Work IQ API are billed in Copilot Credits through usage-based billing. The Microsoft 365 admin center shows the numbers but only as snapshots, tab by tab, with no organisational context. This skill reads the exports an admin already has, enriches consuming users with department, job title, country and manager from Microsoft Graph, and produces one report an executive can read in two minutes. Group, policy and user exports are independent aggregates (the admin center does not export user-to-policy membership), so they are shown side by side rather than joined.

What you get

Headline KPIs - credits used, prepaid vs pay-as-you-go share, estimated cost, active users, credits per active user, credits per Cowork task, month-end or annualised forecast
Service breakdown - Cowork, Cowork apps, Work IQ API; prepaid vs pay-as-you-go ring
Spending-limit analysis - every policy with limit, usage rate, billing method; unlimited, near-limit and idle policies flagged
Departments and managers - every consuming user is enriched from Microsoft Graph (department, job title, country, direct manager) so leadership sees spend by department and an accountability view per manager - credits, share, users, avg per user, credits per task, near-limit count, top user
Groups and users - concentration (top 10 / top 20 %), limit tiers, over- and near-limit watchlist, dormant and unlicensed consumers, sortable and filterable tables
Evidence-backed recommendations - each with the numbers behind it and a concrete action
Data-quality notes - snapshot mismatches, overlapping groups, users over 100 % of a changed limit

Output: a self-contained consumption-report.html (no CDN, printable, safe to email), a consumption-summary.md for chat or Teams, and consumption-analysis.json with every figure.

Inputs

Exports from Microsoft 365 admin center > Copilot (CSV, any file name - detected by columns):

Cowork > Usage > Cowork usage details
Cost Management > Configuration > Export (Policies)
Cost Management > Consumption > Users
Cost Management > Consumption > Groups
Cost Management > Consumption > Agents and services

Only Users or Agents and services is strictly required; every extra export unlocks a section. Department and manager data is collected by the skill itself through Microsoft Graph (/users?$expand=manager, read-only). An Entra user export or a simple org CSV works as a fallback. Synthetic sample exports ship in assets/sample-exports/ (five CSVs plus a Graph user JSON with managers) so you can see the report before touching a tenant.

How to use

Drop the exports into the conversation and say “Build a consumption report from these files” (or /analyze-cowork-consumption). Optional: currency and contracted rate (--currency EUR --rate 0.0092), --anonymize for wide distribution (pseudonymises every output), --period ytd if you exported with a year-to-date filter (default is the current billing month).

Quick start outside Cowork:

python scripts/analyze_consumption.py --input <exports folder> --org <graph json folder> --out ./report

Good to know

Reporting only. It never changes spending policies, limits or billing methods.
Read-only Graph. Directory enrichment uses User.Read.All GETs only; unresolved users are reported as unknown, never guessed.
Computed, not guessed. All figures come from the bundled Python script (standard library only).
Invoice is truth. Costs are list-rate estimates; the Azure subscription in the billing method carries the real bill.
Privacy aware. --anonymize replaces names and UPNs with keyed pseudonyms (random per-run secret, so they cannot be reversed from a directory listing) and redacts input paths in every output; watchlists are spend-control, not performance ranking.

Complements the Copilot Adoption Dashboard skill (licence adoption from MAC usage reports) and the Copilot ROI team’s Consumption Central / Chargeback Power BI templates on Analytics Hub - same exports, no Power BI required.