This skill builds a Microsoft 365 Copilot adoption dashboard by feeding four MAC report exports through the Copilot Adoption Analysis framework (Framework. mentioned is custom. All KPIs used are custom developed and not using any public framework approach such as Microsoft Adoption toolkit or M365 Copilot Adoption Guide. Entire framework is within explained within references) and injecting the computed results into a bundled single-file HTML template. The output is one self-contained .html file (no libraries, no network) branded with the customer's name and logo, carrying only real data — every template placeholder, sample number, and template label is replaced. Every KPI tile and panel carries an i definition button, the segment donut, Top surfaces and Agent usage share charts reveal the underlying user counts on hover, and a Recommended next steps section below the charts turns the measured values into the framework's enablement plays — split into KPI-triggered plays and ongoing training/awareness focus. The framework logic (pillars, thresholds, KPI formulas, and the data contract) lives in references/adoption-framework.md; the computation and injection are done by scripts/build_dashboard.py.

When to Use

The user wants their four Copilot MAC exports turned into a populated dashboard.
The user runs the create-copilot-adoption-dashboard command.
The user asks to populate or brand the adoption dashboard template with real data.

When NOT to Use

A generic HTML page, report, or single chart with no MAC data — use the html or render-ui skill instead.
Reading numbers from a Power BI report or semantic model — use the powerbi skill instead.
Summarizing a meeting or email thread — use meeting-intel or the Outlook tools instead.
Evaluating individual users' performance — decline; this skill reports aggregate adoption only.

Quick Start

User: /create-copilot-adoption-dashboard
1. Explain the process (below) and list the four reports needed.
2. Ask for company name (required) + logo (optional); find the 4 CSVs in input/.
3. Inspect each CSV's headers and map it to a report (--usage / --agents-user /
   --agents-agent / --chat).
4. Run scripts/build_dashboard.py to compute the metrics and write the HTML to working/.
5. Review the KPI summary, publish to output/ with CopyArtifact, confirm with Glob.

Core Instructions

Phase 1 — Explain the process (always do this first)

Tell the user, in a few plain lines, how this works:

This builds your Copilot adoption dashboard from four exports you pull from the Microsoft 365 admin center Copilot reports: (1) Copilot usage (per licensed user), (2) Agents – per user, (3) Agents – per agent, and (4) Copilot Chat (includes unlicensed users). I'll compute license activation, the habitual-user and agent-adoption rates, user segments (Champion → Never), top app surfaces, per-agent usage share, and the unlicensed-Chat conversion funnel, then drop them into a branded, single-file dashboard. The reporting window is taken from the Report Period column in your export (the admin center default is 28 days), so the dashboard always states the period your data actually covers. I just need your company name (required), an optional logo, and the four CSV files. Note: Adoption Framework for KPIs is custom, as explained in the reference documentation.

Phase 2 — Collect inputs

Company name + logo: if not already given, ask with AskUserQuestion (company name is required; logo is optional — a PNG/JPG file). Do not ask for anything a lookup can answer.
The four CSVs: find them with Glob on input/**/*.csv. If fewer than four are present, tell the user which report types are missing and proceed with the ones supplied (missing reports degrade to a pending panel — never invented data).

Phase 3 — Map columns and compute

Inspect headers of each CSV (read the first two lines) and match each file to a report flag using the signatures in the reference: per-app Loop/OneNote columns → --usage; agents-used column → --agents-user; an agent-name row grain → --agents-agent; a web-chat surface / unlicensed users → --chat.
Run the build (from the skill directory) — write to working/ first:
```
python scripts/build_dashboard.py \
  --company "<Company>" [--logo input/<logo>] \
  --usage input/<usage>.csv --agents-user input/<agents_user>.csv \
  --agents-agent input/<agents_agent>.csv --chat input/<chat>.csv \
  --out working/<Company>_Copilot_Adoption_Dashboard.html
```
Omit --period — the window comes from the export's Report Period column (MAC default 28 days). Pass --period only when the user explicitly asks for a different window, and --report-date only to override the export's Report Refresh Date.
Read the JSON summary the script prints. Check periodDays and periodSource — confirm the window matches the export the user supplied (e.g. 28 from the Report Period column) and mention it when you report back. If warnings names an unmapped column, write a small --col-map JSON (see the reference) and re-run — do not hand-edit numbers. Confirm template_reference_check is clean.

Phase 4 — Review and deliver

Show the user a short table of the computed KPIs (activation, habitual, agent adoption, conversion), the number of recommended next steps the dashboard raised, plus the report date, the reporting window and where it came from (e.g. "28-day window, from the export's Report Period column"), and which sources were used, for a quick sanity check before delivery.
Publish the finished file to output/ with CopyArtifact(surface="output", source="working/<file>.html", destination="<file>.html"), then confirm it exists with Glob output/**/*.
Tell the user the dashboard is ready and note any pending panels (missing sources).

Output

Primary deliverable: one self-contained .html dashboard in output/, named <Company>_Copilot_Adoption_Dashboard.html, opening directly in any browser.
Chat reply: 3–5 lines — the KPI summary table, the report date/window, and a note of any missing source. Keep internal paths and tool names out of it.

Guardrails

Never fabricate a metric: every KPI, segment, and share is computed by the script from the exports. If a column can't be mapped, provide a --col-map and re-run; never type a number in by hand.
Always inspect headers before computing — tenants name columns differently; confirm each file maps to the right report flag.
Never hard-code the reporting window. It is read from the export's Report Period column (MAC default 28 days) and shown on the dashboard badge and footer; only pass --period when the user explicitly asks for a different window. If the script warns that reports disagree or that no Report Period column was found, tell the user which window was used and why.
If a report is missing, leave its sources flag false so the panel renders as pending — do not guess or copy sample values as a fallback.
Thresholds (segments, lead tiers, KPI targets) come from the framework and are tunable; if the user asks, adjust them in the reference, not by editing outputs.
Review before delivery: show the KPI summary table and confirm the company name spelling before publishing the final file; keep the four exports in working/ and deliver only the dashboard.
Recommendations are threshold-driven, never invented. KPI-triggered cards appear only when a measured value trips its framework threshold, and each cites that number; the ongoing training/awareness cards are labelled as such because they are not threshold-driven. Do not add, reorder or soften cards by hand — if a threshold needs to change, change the template's RECO_T and say so.
Dormancy is not a recommendation trigger. The framework treats it as a trend needing two reporting periods; this is a single-period view, so the Dormant segment is charted but never used to raise a play.
This skill reports aggregate adoption only — never rank or evaluate individual users. Redirect performance questions to the user's manager or HR.

create-copilot-adoption-dashboard

Overview