Cowork Super User (Preview)
Explore weekly Cowork activity, credit concentration, and directional value scenarios.
Open enablement guideWhich report should I use?
Open-source reports and tools, each built alongside the customer who needed it first.
Explore weekly Cowork activity, credit concentration, and directional value scenarios.
Open enablement guideChoose a goal
What are you trying to do — prove ROI, control cost, drive adoption? One goal at a time.
Add a product, or do not
Stack as many Copilot products as you like. Goal and product work together, and either one alone is fine.
Open the card
Inside every report: a setup guide, a walkthrough video where one exists, and the template itself to download from GitHub.
The reports most teams open first.
Lower usage, or covered by a stronger report above. Still fully supported.
Answer step 1, or pick a surface in step 2, or both. We rank the reports that answer it and put the ones built for the job first. Every recommendation opens the same detail view as the cards.
No report covers . and try a wider one.

Pick how your data gets in. Bring whichever products you have; missing ones just leave their pages empty. Each path ends the same way: open the .pbit, fill in one or two parameters, click Load.
Viva Insights ships Copilot data de-identified, so it can't be joined to your other sources until this is on. Cowork totals are correct either way.
A Global Administrator or AI Administrator, about two minutes:
Full steps and the two query names →
This processes personal data. Check whether per-person reporting needs works-council consent or employee notification where you operate — your organisation is the data controller, not Microsoft. The connector does not enforce Viva's minimum group size, so apply any privacy threshold in the report yourself.
Download your exports, drop them in one folder, point the template at it.
Anywhere on disk — C:\Consumption Central\Data is fine.
File names don't have to match exactly — the template recognises the usual variations. Grab only the products you have:
| Product | Where to get it (sign-in required) |
|---|---|
| Cowork / Work IQ | Viva Insights → Analysis → build a query → download CSV |
| Copilot Studio | Power Platform admin center → Licensing → Copilot Studio |
| GitHub Copilot | GitHub → Billing & Licensing → Usage → AI usage report |
| Azure AI Foundry | Azure Cost Analysis → export, or run pull_azure_ai.py |
Full click-paths and permissions →
Get the current Local CSV template through the Consumption Central README and setup guidance. Open it, paste your folder path into DataFolder, click Load. That's it.
Try it with sample data first: point DataFolder at 5. Local CSV/sample-data and the whole report fills in.
Only want department views? Expand Turn on Viva identification above. On this path a de-identified export's PersonPolicyMap.csv also bridges the hash if you include it.
Set it up once, then it refreshes itself. Data lands in a Lakehouse; the report reads it on a schedule and accumulates history beyond Viva's 6-month export window.
Workspace or Lakehouse setting. They write with saveAsTable(...) and read from Files/landing/…, both of which target the notebook's attached default Lakehouse. So for each notebook you use: Explorer pane → Add your Lakehouse → set it as the default (the star). That's what points the notebook at your data. The only things you edit in the top cell are source-specific (an enterprise slug, a Key Vault URL, a date window), and several notebooks need nothing beyond dropping a file in the landing folder.Fabric portal → your workspace → New → Lakehouse. Everything below writes into this one.
Viva Insights ships a certified Dataflow Gen2 connector that writes query results straight into your Lakehouse on a schedule.
viva_credits_weekly, keep the person identifiers + metric + attribute columns. Schedule Tue ~8am PST, after Viva's weekend refresh.Microsoft's Dataflow Gen2 guide →
From 2. Fabric/notebooks/ — import only the ones you need, set the imported Lakehouse as the notebook's default (see the note above), fill in whatever the top cell lists, run. What each one needs and writes:
| Notebook | Reads | What you set at the top | Writes |
|---|---|---|---|
Ingest_GitHub_API preferred | GitHub REST AI-credit endpoints | ENTERPRISE slug; TOKEN (classic PAT via Key Vault getSecret); BACKFILL_MONTHS | github_ai_usage, github_user_map |
Ingest_GitHub CSV fallback | Emailed AI usage CSV in Files/landing/github | Nothing — just drop the file | github_ai_usage, github_user_map |
Ingest_Azure_AI | Azure Cost Management + Azure Monitor | SUBSCRIPTION_ID, TENANT_ID, CLIENT_ID, KEY_VAULT_URL, CLIENT_SECRET_NAME, DAYS — needs an Entra app + Key Vault secret + Azure RBAC first | azure_ai_spend, azure_ai_tokens |
Ingest_Studio supported | 3 PPAC CSVs in Files/landing/studio | Nothing required (SNAPSHOT_MONTH only when backfilling an old export) | studio_tenant_daily, studio_agent, studio_user |
Ingest_Viva_Consumption fallback only | Viva CSV in Files/landing/viva — use Step 2 Dataflow instead where you can | Nothing required | viva_credits_weekly, viva_spending_policy |
Ingest_Org optional | Derives from viva_credits_weekly; optional CSVs in Files/landing/org | Nothing required | org_attributes |
Ingest_CommercialTerms optional | Azure Cost Management (rate = cost ÷ quantity) | SCOPE (ARM billing path); needs Cost Management Reader | commercial_terms |
Ingest_Studio_Consumption in testing | Power Platform licensing API | TENANT_ID, CLIENT_ID, KEY_VAULT_URL, CLIENT_SECRET_NAME, DAYS, RESTATE_DAYS | studio_consumption_daily, studio_capacity, studio_environment |
Ingest_Azure_AI and Ingest_Studio_Consumption read a secret from Key Vault and exchange it for an ARM / Power Platform token. Fabric's getToken audiences do not include ARM or the Power Platform API, so set up the Entra app + Key Vault secret first — Azure ingestion setup →. Leave Azure out and the Foundry page is simply empty (supported).Ingest_Studio_Consumption is not a substitute yet
It is in testing, and a scheduled Fabric refresh cannot authenticate to that API at all — the permission it needs exists only as a delegated one. Use Ingest_Studio (the CSV route) for Studio. Details →Lakehouse → Settings → SQL analytics endpoint → copy the connection string. Get the current Fabric template through the Fabric setup guide, open it and paste:
FabricSQLEndpoint | The SQL endpoint string you just copied |
|---|---|
LakehouseName | Your Lakehouse name |
Everything else has a default.
Publish to your workspace, then schedule ingestion before the semantic-model refresh — refresh only on ingestion success, never on failure or while a writer is running (a Power BI refresh alone does not call Azure or run notebooks). Tuesday morning suits Viva; Azure can run daily.
Try it with sample data first: seed_sample_data.py loads the synthetic dataset straight into your Lakehouse.
Connect Power BI straight to Viva Insights and let it refresh itself. Other products (optional) still come from files.
Get the current Viva Direct template through the Consumption Central README and setup guidance, open it, paste the identifiers, click Load.
VivaPartitionId | Partition identifier |
|---|---|
VivaQueryId | Query identifier |
(500) Internal Server Error.Adding the other products (optional): set DataFolder to a folder of whatever exports you have — found by name, nothing to rename. Leave it blank for consumption-only. Policy names need M365SpendingPolicyMetaData.csv from the Viva query download.
Heads up on department views: this live-connection path has no PersonPolicyMap.csv bridge, so if Viva identities stay hashed, department breakdowns can't be built at all. Expand Turn on Viva identification above first.
Power Automate pulls consumption on a schedule, writes it to Dataverse tables, Power BI reads those tables. Have Fabric? Use tab 2 instead.
/licensing/entitlements/MCSMessages/resources, which currently returns 403 even for a Global Administrator with every relevant scope consented. The Azure and GitHub flows do not depend on it. Studio flows must also be owned by an administrator — the Power Platform API has no application role for licensing, so a service principal can't read it. What was tried →From 4. Power Automate + Dataverse/scripts, dry-run first (prints the 11 tables / 123 columns, changes nothing):
python Deploy-DataverseSchema.py --environment https://your-org.crm.dynamics.com # then, with a token in DATAVERSE_TOKEN: python Deploy-DataverseSchema.py --environment https://your-org.crm.dynamics.com --execute
make.powerautomate.com → My flows → Import → Import Package (Legacy) → upload ConsumptionCentral-Dataverse.zip → set the Dataverse + Key Vault connections. Eight flows (a daily + a backfill per feed) import switched off.
Open each daily flow and set TenantId, ClientId (app reg from PERMISSIONS.md) and DataverseUrl in its first action. The secret is read from Key Vault under consumption-central-client-secret.
Run each Backfill flow once by hand (last 180 days, ~10–30 min each; let each finish first). Then switch on the four daily flows.
Get the current Power Automate + Dataverse template through the Consumption Central README and setup guidance, open it and set:
DataverseServer | your-org.crm.dynamics.com,5558 |
|---|---|
DataverseDatabase | your-org (first part of the env URL) |
DataversePrefix | cc_ (leave unless you changed the publisher prefix) |
Click Load, then Publish and set a scheduled refresh.
studio_tenant_daily, studio_agent, azure_ai_spend, github_ai_usage. The other seven tables have no API — load them by hand via Data → Import in make.powerapps.com, or use tab 1 / tab 2 to populate every page.PAX is the tool that produces the data, so connecting your data here means running PAX. If you do not live in the command line, start with Mini-Kitchen (tab 1): it builds the exact PAX command for you in the browser, and you run it. Prefer to script it yourself, or send output to SharePoint or Fabric? Those are the next two tabs. Every path pulls the same records from Microsoft Purview, Microsoft Entra, and the Microsoft 365 admin center, and writes analysis-ready output you point Power BI at.
PAX reads the Microsoft 365 unified audit log. With logging off, every query returns zero rows and nothing tells you why. It does not backfill, so a tenant that just switched it on starts thin.
A Global Administrator or a role with audit access, about two minutes:
Microsoft began enforcing a new AuditLogsQuery.Read.All permission in April 2026. With only the legacy permission, audit calls appear to succeed and silently return zero records, so make sure admin consent is granted. PAX v1.10.9 and later request the correct scopes.
Point and click a PAX command in the browser, copy it, then run it yourself. Mini-Kitchen never connects to your tenant and stores no credentials. It only writes the command.
Open Mini-Kitchen and start from a preset. Nothing you type there leaves the browser.
| Preset | What it collects |
|---|---|
| AI-in-One Dashboard | Copilot interactions plus Entra user info, rolled up. Does not pull the broader M365 bundle. |
| AI Business Value Dashboard | The same Copilot rollup, shaped for the AI Business Value dashboard. |
| M365 Usage Dashboard | Adds the Exchange, OneDrive, SharePoint, and Teams audit bundle alongside Copilot. |
| Entra user info only | Skips the audit query and pulls Entra user details only. |
| Microsoft Agent 365 catalog | Exports the Microsoft Agent 365 catalog only. |
None of these fit? Choose Custom audit export and set the scope yourself.
Work down the builder: scope, date range, rollup, filters, output target, and sign-in. To keep a recipe, click Export for a .paxlite file. Saved recipes live in browser storage only, so clearing site data deletes them.
Copy the command Mini-Kitchen built. Download the PAX script from the PAX releases page, open PowerShell 7, and paste it in. This is the step that reaches your tenant. Mini-Kitchen never did.
First time? Pick AI-in-One Dashboard, a short date range, and a local output folder, then run it once to see the shape of the data.
Download the script and drive it with switches. Full control, fully scriptable, no browser step.
Most people want the flagship Purview Audit Log Processor. There are three scripts, for different needs:
| Script | Use it when |
|---|---|
| Purview Audit Log Processor start here | You want Copilot, agent, and broader M365 usage from the unified audit log. |
| Graph Audit Log Processor | You want a lightweight Copilot usage plus Entra and licensing export with no Purview dependency. |
| Copilot Interactions Content | You need the actual prompt and response content, not just usage telemetry. |
Use the PAX README and setup guidance to select and obtain the current processor, then copy its current command for your reporting window into PowerShell 7. -Rollup produces the dashboard-ready summary files, -IncludeUserInfo adds Entra names and departments, and -IncludeM365Usage adds the broader workload bundle. Confirm the switches against that guidance before running.
Point your Power BI template at the rollup CSVs the run produced. The PAX repository documentation lists the current switches and output files.
-Rollup. Add -RollupPlusRaw if you also want the raw per-record CSV.PAX writes wherever -OutputPath points. Give it a local path, a SharePoint library URL, or a Microsoft Fabric OneLake URL, and it detects the destination from the URL shape. There are no destination-specific switches.
| Local folder | -OutputPath C:\PAX\ |
|---|---|
| SharePoint library | -OutputPath "https://contoso.sharepoint.com/sites/Analytics/Shared Documents/PAX" |
| Fabric OneLake (Delta tables) | -OutputPath "https://onelake.dfs.fabric.microsoft.com/Analytics/PAX.Lakehouse" |
For a scheduled Azure run with no interactive sign-in, use a managed identity or an app registration. A managed identity suits Azure Container Apps Jobs, VMs, and Functions:
Follow the current PAX unattended-run instructions for the processor filename, managed-identity command, and required permissions. Choose your date range, rollup output, tenant and client identifiers, and OneLake destination there.
App registration with a certificate works the same way, for example writing to SharePoint:
... -Auth AppRegistration -TenantId "<tenant-guid>" -ClientId "<app-id>" ` -ClientCertificateThumbprint "<thumbprint>" ` -OutputPath "https://contoso.sharepoint.com/sites/Analytics/Shared Documents/PAX"
Sites.ReadWrite.All plus Member access on the target site. Fabric output needs Contributor on the workspace, Storage Blob Data Contributor, and the Service principals can use Fabric APIs tenant setting turned on for app-registration or managed-identity runs.
Every path builds the same dashboard from the same core sources: Copilot interactions from Microsoft Purview audit logs, your licensed users from the Microsoft 365 admin center, and department data from Microsoft Entra. The template reads two pre-processed rollup files, not a raw Purview or Entra export. New here, start with Local CSV and its bundled sample data.
The template needs two pre-processed rollup files and cannot read a raw Purview CSV or a hand-exported Entra list. Mini-Kitchen is the guided, no-install way to produce them: it runs entirely in your browser, drives the open-source PAX engine, and ships an AI Business Value Dashboard preset that fills in exactly the right collection settings for this dashboard, so you do not have to understand the audit internals.
.paxlite recipe if you want to reuse it.You get an *_Interactions_*.csv and a *_Users_*.csv. To keep the dashboard fresh, re-run the command on a cadence (for example a Windows Task Scheduler job).
Already export raw data another way? The report's standalone processor turns a raw Purview export plus Entra users and licensing CSVs into the same two rollup files, no Mini-Kitchen. Get the current processor and command from the ValueLens README and Local CSV setup guidance.
Data-export roles and field list →
Producing this data reads tenant audit and directory information. Typical roles: Audit Reader (Purview), Global Reader or User Administrator (Entra and licensing). If exported names come back as 32-character hex, clear Display concealed names in Microsoft 365 admin center → Settings → Org settings → Reports, then re-run.
The packaged SharePoint pipeline and the Fabric and Dataverse paths pull from Microsoft Graph unattended, so they need one Entra app registration. The Mini-Kitchen route above does not, it signs in interactively when you run the command.
A Global Administrator, once:
AuditLogsQuery.Read.All (audit interactions), Reports.Read.All (licensed users), User.Read.All (org data). The packaged SharePoint scripts also use Sites.Selected to write to your library, and Organization.Read.All for the licence lookup fallback.Optional Agents 365 needs CopilotPackages.Read.All + Application.Read.All + User.Read.All and an Agent 365 licence. Full least-privilege breakdown: docs/PERMISSIONS.md.
See the whole dashboard now on a bundled sample dataset, or point it at your own rollup files. Manual refresh by design.
Get the current Local CSV template and sample files through the ValueLens README and Local CSV setup guidance. Open the template in Power BI Desktop (its parameters take local file paths, do not open the SharePoint template here). Point each parameter at the matching sample file, then Load. Every page fills in, no tenant needed.
| Copilot Interactions File | copilot_interactions_sample.csv |
|---|---|
| Org Data File | copilot_users_sample.csv |
| Agent 365 (optional) | agents_365_sample.csv |
Use Mini-Kitchen with the AI Business Value Dashboard preset, see Start here above. Pick a local output folder, then run the command it builds in PowerShell 7. You get an *_Interactions_*.csv and a *_Users_*.csv.
Open the same ValueLens - Local CSV.pbit, set Copilot Interactions File to your *_Interactions_*.csv and Org Data File to your *_Users_*.csv, then Load.
Re-run the Mini-Kitchen command to regenerate the files, then Refresh in Power BI Desktop.
Blank visuals almost always mean the file is a raw export, not a rollup. The template reads the pre-processed *_Interactions_*.csv / *_Users_*.csv pair only.
Large tenant? The Purview UI export caps out early, but the PAX engine partitions the query and runs unattended, so the Mini-Kitchen route above scales where a hand export does not.
Land the two rollup files in a SharePoint library and let Power BI refresh from there on a timer. Two ways to produce and land them: Mini-Kitchen with the AI Business Value Dashboard preset set to a SharePoint output folder and re-run on a Task Scheduler cadence (see Start here above, no install), or the report's own packaged scripts below, which wrap the PAX engine with app-only auth, seed-then-append de-duplication, upload and scheduling in one. The packaged route needs the app registration above, a SharePoint library, and a host with PowerShell 7+ and Python 3.10+.
Run ProvisionSiteAccess-SP-AppReg.ps1 to grant the app per-library Sites.Selected write. Save the SiteId and DriveId it prints, the upload step needs both.
The interactions data is a growing time series, so Run-PAX-AIBV.ps1 seeds a back-fill once, then appends short windows. It de-duplicates on each message identity, so overlapping days reconcile.
# First run, seed the file with a back-fill
.\Run-PAX-AIBV.ps1 -TenantId <id> -ClientId <id> -Days 30
# Scheduled runs, append only the latest window
.\Run-PAX-AIBV.ps1 -TenantId <id> -ClientId <id> -Days 2 `
-AppendFile Purview_CopilotInteraction_Rollup.csv
Add -IncludeAgent365Info for the optional Agents 365 output, or -UserInfoFile to supply your own org directory instead of pulling Entra live.
Upload-Rollups-SharePoint.ps1 lands them as fixed names copilot_interactions_rollup.csv and copilot_users_rollup.csv, overwriting the previous run.
Register-TaskScheduler.ps1 registers a daily task with -AppendFile. The secret is not stored in the task.
Get the current SharePoint template through the ValueLens README and SharePoint setup guidance. Open it, and under Transform data → Edit parameters point Copilot Interactions File and Org Data File at the two SharePoint URLs. Load → Publish, then in the Power BI service set the SharePoint data-source credentials and a scheduled refresh timed after your extract.
Tops out at Power BI Pro's 1 GB model and 2-hour refresh window. Beyond that, move to Fabric.
PySpark notebooks pull straight from Graph into a Lakehouse, a processor builds the curated fact table, and an Import-mode template reads it. No file caps, and the optional feedback and Agent 365 sources plug into the same model. Needs the app registration above, with its secret in Key Vault.
Manual setup (no installer), the same path the walkthrough video follows:
Fabric portal → your workspace (on capacity) → New → Lakehouse. Note its SQL analytics endpoint, or the workspace and Lakehouse GUIDs if you use the OneLake template.
Import from 1. Fabric/Manual setup/notebooks/, attach your Lakehouse as each notebook's default, and run:
| Notebook | Writes |
|---|---|
Copilot_Audit_Log_Direct_Ingester | copilot_interactions_parsed |
Copilot_Licensed_Users_Direct_Ingester | copilot_licensed_users |
Copilot_Org_Data_Direct_Ingester | copilot_org_data |
Set the audit ingester's MODE to backfill for the first load, then incremental for scheduled runs. See notebooks/README.md.
Run Copilot_Audit_Log_Processor after the three ingesters, immediately before the model refresh. It builds copilot_interactions_curated.
Pick one, both are Import mode over the same tables:
| Template | Set these parameters |
|---|---|
| Fabric template setup SQL endpoint | Fabric SQL Endpoint + Lakehouse Name |
| Fabric OneLake template setup OneLake | Fabric Workspace ID + Lakehouse ID |
Set RangeStart and RangeEnd to cover the history you want, keep Enable_ProductFeedback and Enable_Agent365 on Exclude until those tables exist, then Load.
Publish, set the data-source credentials in the service, then add a success-gated model refresh that runs after the pipeline succeeds. See pipelines/README.md. The shipped pipeline JSON has no refresh activity, you add it.
Copilot_Agent365_Registry_Ingester (preferred, unattended) or Copilot_Agent365_Lander (CSV fallback), never both. The shipped pipeline wires the CSV lander. Product feedback lands from Files/product_feedback/ behind Enable_ProductFeedback.A Power Automate collector writes full audit records to Dataverse, a scheduled runner uses the same ValueLens processor to prepare them, and the template reads the result. Use this only if you are already collecting Copilot interactions into Dataverse. For the settled routes, use SharePoint or Fabric.
Run Test-PowerAutomateDataverse-Preflight.ps1 against your Dataverse environment URL.
Deploy-DataverseCoreSchema.py dry-runs with no credentials, then creates the tables with --execute and a DATAVERSE_TOKEN: raw audits, curated interactions, curated users, and run manifests.
Supply your authorized unmanaged CopilotInteractionLogging.zip, adapt it for full-record retention with Prepare-CollectorRawCapture.py, then import it into an isolated Power Platform environment. Run the manual back-fill before enabling its daily schedule.
Invoke-DataverseCoreRefresh.ps1 fetches Graph users and licences, reads the retained audits, runs the processor, and emits a Core Snapshot ID on success. Pin it to a completed collector run and its exact UTC window with -SourceRunId, -RawStartUtc and -RawEndUtc.
Get the current Power Automate + Dataverse template through the ValueLens README and setup guidance, open it and set:
Dataverse URL | Your environment origin, e.g. https://contoso.crm.dynamics.com |
|---|---|
Core Snapshot ID | The successful run ID from step 4 |
Use SharePoint CSV fallback | Leave false for this path |
Load, then Publish and set organizational Dataverse credentials. To refresh, update Core Snapshot ID to a newer completed run.
Needs both an Entra app with Graph audit and directory permissions and a Dataverse application user, consent to Graph does not grant Dataverse access. See the path README.

Pick how your data gets in. All three paths read the same Copilot Studio conversation transcripts and drive the same nine pages; they differ only in how the data arrives and how it refreshes. Want to see it first with zero setup? Follow the ESS README sample-data guidance to explore all nine pages before you arrange any access.
The dashboard reads the transcripts Copilot Studio writes for each conversation. Those exist only if the agent is set to save them, and Dataverse keeps them for 30 days by default. A Copilot Studio maker or admin, a few minutes:
Export the transcript table once, point the template at the file. Best for a snapshot, a demo, or sharing outside your tenant.
Open the Power Apps maker portal, switch to the agent’s environment, go to Tables then All, search conversation, open ConversationTranscript, then Export then Export data. Download the result and unzip the CSV. The default window is the last 30 days.
Content column and the template fails to load with a Token Identifier expected error. If you opened it, re-export from Dataverse rather than trying to repair it.A roster CSV that maps UserPrincipalName to Department and Country unlocks the Organization and Country breakdowns. From the Microsoft 365 admin center: Users → Active users → Export users → Confirm (the template normalizes the column names). For the credit-consumption leaderboard, add an export from Copilot Studio → Analytics → Message Consumption. Both are optional; the report loads without them.
Get the current CSV template through the ESS CSV setup guide. Open it, paste the full path to each CSV into Transcript File (and the optional Org Data File / Agent Credits File), click Load.
Want scheduled refresh without a gateway? Host the transcript CSV on SharePoint or OneDrive and point the parameter at its direct file URL, then bind credentials in the service. See AUTO-REFRESH.md.
Connect the template straight to the Dataverse environment that hosts your agent and let it refresh itself. No file exports, no gateway.
Power Apps → switch to the agent’s environment → ⚙ gear icon → Session details → copy the Instance url (like https://orgXXXXXXXX.crm.dynamics.com). Drop the trailing slash; the connector is picky about it.
Same optional roster and Message Consumption files as the CSV path (see tab 1). Paste their paths into the matching parameters.
Get the current Dataverse template through the ESS Dataverse setup guide. Open it and fill in:
Dataverse Environment URL | The instance URL you just copied |
|---|---|
Transcript Lookback Days | Leave blank for 90. The filter runs server-side, so a smaller window refreshes faster. |
Click Load.
When prompted, choose Organizational account, sign in with an account that holds the Bot Transcript Viewer role on that environment, then Connect. If it returns zero conversations, the account is missing that role, tenant admin is not the same thing.
OAuth2, Organizational) once, then set a schedule. Steps in AUTO-REFRESH.md.One environment per file. For agents spread across several environments, either build one copy of the template per environment or use the Fabric path.
Notebooks land your transcripts and credit data in a Lakehouse, and the template reads from there. Use this only if you are consolidating multiple environments, hitting refresh limits at very large scale, or want built-in credit-consumption analytics and automatic offline topic classification. Otherwise stay on CSV Upload or Dataverse Direct.
Fabric portal → your workspace → New → Lakehouse. Everything below writes into this one.
Import Copilot_Agent_Transcript_Parser.ipynb (New → Import notebook), then in the Explorer pane add your Lakehouse and set it as the default. In the CONFIG cell keep SOURCE_MODE = 'dataverse' and set TENANT_ID, CLIENT_ID, CLIENT_SECRET for an Entra app registration (read it from Key Vault with notebookutils.credentials.getSecret in production), plus DATAVERSE_URL, or a list in DATAVERSE_URLS to consolidate several environments. Run the preflight cell to confirm auth, then Run all. It writes agent_sessions and its companion tables and classifies every conversation by topic offline, with no data leaving your workspace.
SOURCE_MODE = 'files' and drop an exported transcript CSV into the Lakehouse Files/ folder instead.The parser is scoped to ESS agents by default (a filter keys on the copilotforemployeeselfservice schema). Comment out that one filter cell to ingest any Copilot Studio agent. Stage 2 of the offline classifier is toggleable with ENABLE_SEMANTIC_FALLBACK.
Import Copilot_Credit_Consumption_Ingester.ipynb the same way. It reads the Message Consumption exports (the EntitlementConsumption... CSVs from Copilot Studio) from Files/credit_consumption in the Lakehouse, dropped there by hand or by a Power Automate flow, and writes the credit_consumption tables.
Get the current Fabric template through the ESS Fabric setup guide. Open it, point it at your Lakehouse SQL analytics endpoint (Import or DirectQuery) or Direct Lake, then Refresh. Org data is still a separate roster file, the same one used on the other paths.
Schedule both notebooks, or wrap them in a Fabric pipeline, so the Lakehouse stays current without manual runs.

Every number comes from the Microsoft Purview audit log plus Entra user and licensing data. You produce two pre-processed “rollup” files (the template can’t read raw exports), then open the edition that matches where your files live and how you want it to refresh. Pick your path below.
AuditingDisabledTenant. It’s on by default for most tenants but often off in demo / dev / new tenants, so check before you start. In the Purview portal, if you see Start recording user and admin activity, click it. Enabling it does not backfill — collection starts from that moment and can take up to 24h to appear, so a brand-new tenant’s first export looks sparse. That’s expected.The template needs two pre-processed rollup files and can’t read a raw Purview CSV or a hand-exported Entra list. Mini-Kitchen is the guided, no-install way to produce them: it runs entirely in your browser, drives the open-source PAX engine, and ships an AI-in-One preset that fills in exactly the right collection settings for this dashboard, so you don’t have to understand the audit internals.
.paxlite recipe if you want to reuse it.You’ll get Purview_Audit_..._Interactions.csv, EntraUsers_MAClicensing_..._Users.csv, and (optionally) Agent365_....csv. To keep the dashboard fresh, re-run the command on a cadence (for example a Windows Task Scheduler job).
Already export raw data another way? The standalone processor turns raw Purview + Entra exports into the same two rollup files, no Mini-Kitchen. Get the current processor and command from the AI-in-One README and setup guidance. Agent 365 can also be exported by hand from admin.microsoft.com → Agents → All Agents → Export (it’s a passthrough, no processing needed).
Data-export roles & field list →
Producing this data reads tenant audit and directory information. Typical roles: Audit Reader (Purview), User Administrator or Global Reader (Entra + licensing), and AI Admin or Global Reader for Agent 365. Your organisation is the data controller for anything you export.
Produce the rollup files, drop them on your machine, open the auto-detect edition, point three parameters at them.
No tenant access yet? The repo ships fictional demo files in the exact rollup format. Open the template, point it at sample-data/, and explore every page with no Purview, PAX, or tenant.
Use Mini-Kitchen with the AI-in-One preset — see Start here above. Pick a local output folder, then run the command it builds in PowerShell 7.
You’ll get Purview_Audit_..._Interactions.csv, EntraUsers_MAClicensing_..._Users.csv, and (optionally) Agent365_....csv.
Get the current Rollup Edition template through the AI-in-One README and setup guidance (the 3-in-1 edition accepts local paths, SharePoint, or OneLake). Open it, fill three parameters, click Load:
| Copilot Interactions File | Path to your ..._Interactions.csv |
|---|---|
| Org Data File | Path to your ..._Users.csv |
| Agent 365 (optional) | Path to your Agent365_....csv, or leave blank |
Re-run the Mini-Kitchen command to regenerate the files, then Refresh in Power BI Desktop.
Blank visuals almost always mean the file isn’t a rollup file — confirm it matches the ..._Interactions.csv / ..._Users.csv pattern from a -Rollup run.
Write the rollup files to a SharePoint library, open the SharePoint edition, publish, and let the Service refresh it on a schedule.
Use Mini-Kitchen with the AI-in-One preset (see Start here above) and set its output location to a SharePoint document library instead of a local folder. Run the command it builds on a cadence (for example a Windows Task Scheduler job) so the files stay current.
.../file.csv), with no ?....Get the current PBI-SharePoint Rollup Edition template through the AI-in-One README and setup guidance, then open it. Every parameter must be a SharePoint URL (it validates this). Put all three files in the same folder, including a manually-exported Agent 365 CSV, so you don’t mix source types.
Publish to a workspace, then in the Service: dataset → Settings → Data source credentials → sign in to SharePoint with OAuth2, set Privacy = Organizational. Then enable Scheduled refresh. No Gateway needed.
Tip: have the Mini-Kitchen command write the same filename each run so the dataset refreshes against a stable URL with no template edits.
The fastest, most reliable path at real audit-log volume. Three notebooks pull from Microsoft Graph and write Delta tables straight into a Lakehouse — no PowerShell, no CSV landing step. The report is a thin pass-through.
saveAsTable('dbo.<table>') to the notebook’s attached default Lakehouse (there is no workspace / Lakehouse variable to edit — you pin the Lakehouse as default). A Lakehouse created without schema support has no dbo schema, and every notebook then fails with a generic Spark error that names neither the schema nor the table. Tick Lakehouse schemas in the creation dialog.Create an app registration with these Microsoft Graph application permissions (admin consent required), then grab the Tenant ID, Client ID, and client secret value:
| Permission | Used by |
|---|---|
AuditLogsQuery.Read.All | Audit notebook (Purview audit query API — not the Entra AuditLog.Read.All scope) |
Reports.Read.All | Licensed-users notebook |
User.Read.All | Org-data notebook |
Fabric workspace on a capacity → New → Lakehouse, enable schemas (see the note above). Note its SQL endpoint from settings.
From Classic Editions/3. Fabric/notebooks/, import each, pin your Lakehouse as default, and paste your three app-reg values into the # === CONFIG === cell (use Key Vault via notebookutils.credentials.getSecret for production):
| Notebook | Reads | Writes |
|---|---|---|
Copilot_Audit_Log_Direct_Ingester | Graph audit-log query API (7-day windows; backfill or incremental) | dbo.copilot_interactions_parsed |
Copilot_Licensed_Users_Direct_Ingester | Graph M365 active-user report | dbo.copilot_licensed_users |
Copilot_Org_Data_Direct_Ingester | Graph /users (+ manager expand) | dbo.copilot_org_data |
Run them ad-hoc, schedule each, or import the included CopilotAdoptionPipeline to orchestrate all three (audit first, then licensed-users + org-data in parallel).
Get the current Fabric template through the AI-in-One README and Fabric setup guidance, open it and supply:
| Fabric SQL Endpoint | <workspace-guid>.datawarehouse.fabric.microsoft.com |
|---|---|
| Lakehouse Database | Your Lakehouse name (e.g. CopilotAnalytics) |
| Agent 365 (optional) | Local path to your Agent 365 CSV (no Graph loader yet) |
Publish to a workspace on the same Fabric capacity so Direct Lake gives sub-second refresh. Set scheduled refresh to match your notebook / pipeline cadence.
Sql.Database-based template also run on Azure Databricks, Synapse Spark, or any SQL endpoint — adjust one OUTPUT_TABLE line per notebook and the two template parameters. See the alternative platforms section.
Illustrative report pages, not customer results.
Three steps from export to report. Use your organization's approved data.
You need at least 50 Copilot or Viva Insights licenses and global access to the Copilot Dashboard. If Export data is missing, a Global Administrator goes to Microsoft 365 admin center > Settings > Microsoft Viva > Settings > Manage feature access > Viva Insights and turns on Copilot Metrics Export. Changes can take up to 24 hours. See export details.
Go to aka.ms/copilotdashboard > Consumption Dashboard > Explore Cowork > Export data and choose Export by week. Extract the downloaded zip file to a folder.
Open CoworkSuperUser.pbit from the Cowork Super User repository in Power BI Desktop. When prompted for DataFolder, paste the path of the extracted folder without quotes, then select Load. Start on Overview.
Optional: to show individual users instead of anonymized IDs, an administrator can turn on Identifiable Output for authorized analysts (details). Identified data is sensitive: restrict the report and its source files accordingly. The report has no preconfigured row-level security role.
Need column details or help? See the setup guide and troubleshooting.

Pick how your data gets in. Both paths use one small Python preprocessor that reads your exports once and writes the deterministic files the template loads. The one required source is a Microsoft Purview audit export; a Cowork usage export and optional consumption, organization, or identity files just fill in more pages. New to it, start with the fabricated sample: it needs no tenant role and runs in about ten minutes.
Microsoft 365 usage reports conceal user, group, and site names by default, so the Cowork usage export and any per-person or department views come back de-identified. Adoption totals are correct either way, and Reports Reader does not override this.
A Global Administrator, about two minutes:
Cowork usage report reference →
This is a tenant-wide privacy decision under your own policy, and it affects every admin-center report. Do not grant Global Administrator to the report operator.
See the whole report on fabricated data before you collect anything. Every identity is fictional and ends in @example.com.
Follow Path A in the setup guide for the current preprocessor bundle, data contract, template, and fabricated sample package. The repository is authoritative for filenames and commands.
Unzip the sample into C:\CoworkAdoptionSample and the bundle into C:\CoworkAdoptionRelease. The sample folder holds four CSVs plus a purview_audit\ subfolder of synthetic audit files.
Use the current preprocessor command from Path A, with C:\CoworkAdoptionSample as input and C:\CoworkAdoptionPreprocessed as output. Input and output must be different folders.
Open the Cowork Adoption Intelligence template obtained through the setup guide, set PreprocessedOutputPath to C:\CoworkAdoptionPreprocessed, and click Load.
Script and instructions: scripts/. Full walkthrough: Path A in the setup guide.
PreprocessedOutputPath · scheduled refresh needs an on-premises gateway
Collect your exports into one protected input folder, run the preprocessor once to a separate output folder, then point the template at that output. Bring only the sources you have; missing optional ones just leave their pages empty.
In Microsoft Purview → Audit, search your reporting window in UTC with Activities - operation names set to CopilotInteraction, then Export and keep the raw CSV unchanged.
RecordId, CreationDate, Operation, UserId, AuditData, and at least one row whose AuditData has CopilotEventData.AppHost containing cowork.AuditData.RecordId values.Each one fills in more of the report. Match is by filename or by exact headers, so keep one current file per source.
| Source | Where to get it (sign-in required) | Save as |
|---|---|---|
| Cowork usage recommended | Microsoft 365 admin center → Copilot → Cowork → Usage → Export | CoworkUserDetails.csv |
| Consumption optional | Microsoft 365 admin center Copilot Consumption - Users export | CoworkConsumptionDetails.csv |
| Organization optional | Approved HR or Entra export (department, job title, manager, geography) | CoworkUserOrgDetails.csv |
| Identity optional | Entra bulk user download (friendly display names) | cowork_users.csv |
Use a folder outside the cloned repo, for example C:\CoworkAdoptionData. Purview CSVs can sit in a purview_audit\ subfolder; discovery is recursive.
Use the current preprocessing and output-validation commands in Path B of the setup guide. Set the input to your protected raw-data folder and write the entity files to a separate output folder, such as C:\CoworkAdoptionPreprocessed, then validate that output.
A good run reports "status": "valid" and leaves manifest.json plus 13 entity-*.csv files. It never writes to the input folder, and needs no admin role, app registration, or extra Python packages.
Get the current Cowork Adoption Intelligence template through Path B in the setup guide, set PreprocessedOutputPath to the output folder, click Load, and validate against that guide.
Prefer the current template. The earlier raw-Purview and SharePoint editions are kept in release/archive/ for rollback only and are not paired with the preprocessor.

One template, three ways to feed it. The first prompt, Source Mode, is the only answer that matters: it decides which of the other parameters apply. Every page works on every path. Start on Local CSV with the bundled sample data to see the whole report in about two minutes.
The Credit Consumption pages read the Power Platform Admin Center per-message credit reports. Every other page works without them, so skip this unless you want the credit view.
A Global Administrator or Billing Administrator, once a month: open the Power Platform Admin Center → Billing → Licensing (Copilot Studio messages) and download the three reports.
| File name starts with | What it contains |
|---|---|
EntitlementConsumptionTenantDetailsReport_MCSMessages | credits per environment, per day |
EntitlementConsumptionTenantPerAgentDetailsReport_MCSMessages | credits per agent |
EntitlementConsumptionTenantPerUserDetailsReport_MCSMessages | credits per user |
There is no API for these reports, so it is a manual monthly download. On the Fabric path you can automate the drop with the ready-made email or SharePoint landing flows.
The template reads Copilot Studio transcripts from CSV files in one folder. Bundled sample data means it runs out of the box.
Get the current template through the Agent Evaluator Local CSV setup guide, then open it in Power BI Desktop. Leave Source Mode on TranscriptCSV.
Set CSV Folder Path to a folder holding the files below. The template finds them by name, so keep the names as shipped.
| File | What it feeds |
|---|---|
conversationtranscripts.csv | every transcript page |
copilot_org_data.csv | the org filter on every page |
Point it at the bundled data/ folder to fill the whole report with synthetic sample data first.
Click Load. Every page renders.
Credit Consumption pages (optional): add credit_consumption_tenant.csv, credit_consumption_agent.csv and credit_consumption_user.csv to the same folder, using the column names the ingester writes. Leave them out and those pages just stay empty.
The template reads Copilot Studio transcripts straight from Dataverse and parses them in the model. No Fabric, no notebooks, no app registration.
Get the current template through the Agent Evaluator Dataverse setup guide, open it and set Source Mode to Dataverse.
Set Dataverse Url to your environment URL, e.g. https://yourorg.crm.dynamics.com. Find it in the Power Platform Admin Center → Environments → your env → Environment URL. One environment per report.
Set CSV Folder Path to a folder holding copilot_org_data.csv, which drives the org filter. A SharePoint library URL or a local folder both work.
Click Load. On first refresh choose Organizational account and sign in with a login that can read the Conversation Transcript table. The native connector uses that login, so there is no app registration or client secret.
Credit Consumption pages (optional): as on Local CSV, add the three credit_consumption_*.csv files to the CSV folder.
Notebooks land your data as Delta tables in a Lakehouse and the template reads them over the SQL endpoint. This is the path for large volumes, history beyond 30 days, several environments at once, and the turnkey Credit Consumption pages.
Fabric portal → your workspace → New → Lakehouse. Note its SQL analytics endpoint (Lakehouse → Settings).
Import Copilot_Agent_Transcript_Parser.ipynb, fill the # === CONFIG === cell, run. Two ways to source transcripts:
| In the CONFIG cell | What it does |
|---|---|
SOURCE_MODE = 'dataverse' | Pulls live from one or many environments (DATAVERSE_URL or DATAVERSE_URLS). Needs an Entra app added as an Application User in each environment, with TENANT_ID, CLIENT_ID and a CLIENT_SECRET read from Key Vault. Each run accumulates history past the 30-day window. |
SOURCE_MODE = 'files' | Reads a conversationtranscripts.csv you drop in Files/copilot_transcripts. Nothing else to set. |
Writes agent_sessions, agent_turns, agent_errors, agent_subagents, agent_catalogue, agent_performance.
Turn on the credit exports (see the callout above the tabs), drop the three files in Files/credit_consumption, then run Copilot_Credit_Consumption_Ingester.ipynb. It writes credit_consumption_tenant, _agent and _user. To automate the monthly drop, import an email or SharePoint landing flow.
Land copilot_org_data for department breakdowns. The join uses the AAD object ID, so the export must include the id column, which a Graph /users call returns. Without it, org breakdowns come back blank.
Get the current template through the Agent Evaluator Fabric setup guide. Open it, set Source Mode to Fabric, then Fabric SQL Endpoint and Lakehouse Name. Load, Publish, then schedule the notebooks to run before the dataset refresh.

Nothing to export, and no tenant data leaves anyone's machine. You install two Copilot Cowork skills wired to one shared Teams channel: every teammate runs the member skill, which emails a de-identified summary of their Cowork activity to the channel, and you run the manager skill, which reads the channel and emails the team one anonymized dashboard. You get both skills from the Installer Studio, which personalizes each download for your channel.
Both skills point at one dedicated, email-enabled Teams channel. Create it before you build or install anything, and keep it just for these reports so the totals stay clean.
Cowork report - your team, and add only the people whose work should count.The Installer Studio page walks through this with screenshots →
You get both skills from the Installer Studio. It bakes your channel into both downloads, so no one is ever asked to paste a link. It runs entirely in your browser, and your channel link and email are never uploaded.
Go to the Cowork Team Report Installer Studio and paste your channel link and email address from the step above.
Click Build my install links. The page shows the channel name it found. Check it is the right one, then download both zips: the manager skill and the member skill.
Open Copilot Cowork, upload the manager .zip as-is (do not unzip it), and say "Install this manager skill and walk me through setup." Those last words matter. Installing on its own runs nothing, and setup is what sends the member skill to your team.
Send everyone the member .zip. Each person uploads it in Cowork, installs it, and runs it once. After a required privacy review, their de-identified summary is emailed to the channel, and from then on it posts on its own 15-day cadence.
Ask the manager skill to build the team report, then let it run on a schedule, a day or two after the member cadence so every post is in before it aggregates and emails the newsletter.
See it first: open the live sample dashboard (invented data) to see the finished report before you roll it out.

Both templates read the same thing: a Viva Insights person query, one row per person per week. Set the query up once below, then pick how the report reads it. Every path ends the same way: open the .pbit, fill one or two parameters, click Load.
An Insights Analyst (the role is assigned in Viva Insights admin) builds this in the Analyst Workbench. The first run can take a few hours, so start it early.
Is Active = True, and under employee attributes add Organization, FunctionType, TimeZone and SupervisorIndicator.| Metric group | Select |
|---|---|
| Microsoft 365 Copilot | All metrics |
| Collaboration activity | All metrics |
| Collaboration network | Diverse ties, External network size, Internal network size, Strong ties |
| Collaboration by day of the week | All metrics |
| Working hours collaboration | All metrics |
| After hours collaboration | All metrics |
| Meeting types | All metrics |
| External collaboration | External collaboration hours |
| Focus metrics | All metrics |
Reusing an older query? Re-open it, open the Microsoft 365 Copilot metric selection and re-check Select all, so the consolidated Copilot Chat metrics are included. Check the repository for the current metric requirements.
Download the query result once and point the template at the file. Simplest path; the report holds a snapshot until you re-download.
In Analysis results, find your completed query and click the page icon in the Actions column to download the CSV.
Get the current Super User Adoption CSV template through the Decoding Super Usage README and setup guidance. Open it, set SourceType to CSV, paste the full path of the downloaded file into the CSV box, leave the Direct Query boxes empty, click Load.
To refresh: re-run the query, export a fresh CSV, and re-point the template at the new file.
Connect the template straight to the live query. No download, and the report follows each weekly Viva Insights refresh.
In Analysis results, click the link icon in the Actions column for your completed query. It gives you a Partition ID and a Query ID.
Get the current Super User Adoption DirectQuery template through the Decoding Super Usage README and setup guidance. Open it, set SourceType to DirectQuery, paste the two values into PartitionID and QueryID, leave the CSV box empty, click Load. Sign in with your Organizational account when prompted.
Publish to a Power BI workspace, then set a scheduled refresh on the semantic model so it keeps pace with Viva's weekly update.

Both templates read a Viva Insights person query, one row per person per week, the same query the Super User Adoption report uses. If you already ran it there, reuse the completed query and skip to the tabs. Otherwise set it up once below, then pick how the report reads it. Every path ends the same way: open the .pbit, fill one or two parameters, click Load.
An Insights Analyst (the role is assigned in Viva Insights admin) builds this in the Analyst Workbench. The first run can take a few hours, so start it early. One export feeds both this report and Super User Adoption.
Is Active = True, and under employee attributes add Organization, FunctionType, TimeZone and SupervisorIndicator.| Metric group | Select |
|---|---|
| Microsoft 365 Copilot | All metrics |
| Collaboration activity | All metrics |
| Collaboration network | Diverse ties, External network size, Internal network size, Strong ties |
| Collaboration by day of the week | All metrics |
| Working hours collaboration | All metrics |
| After hours collaboration | All metrics |
| Meeting types | All metrics |
| External collaboration | External collaboration hours |
| Focus metrics | All metrics |
Reusing an older query? Re-open it, open the Microsoft 365 Copilot metric selection and re-check Select all, so the consolidated Copilot Chat metrics are included. Check the repository for the current metric requirements.
Download the query result once and point the template at the file. Simplest path; the report holds a snapshot until you re-download.
In Analysis results, find your completed query and click the page icon in the Actions column to download the CSV.
Get the current Super User Impact CSV template through the Super User Impact README and setup guidance. Open it, set SourceType to CSV, paste the full path of the downloaded file into the CSV box, leave the Direct Query boxes empty, click Load.
To refresh: re-run the query, export a fresh CSV, and re-point the template at the new file.
Connect the template straight to the live query. No download, and the report follows each weekly Viva Insights refresh.
In Analysis results, click the link icon in the Actions column for your completed query. It gives you a Partition ID and a Query ID.
Get the current Super User Impact DirectQuery template through the Super User Impact README and setup guidance. Open it, set SourceType to DirectQuery, paste the two values into PartitionID and QueryID, leave the CSV box empty, click Load. Sign in with your Organizational account when prompted.
Publish to a Power BI workspace, then set a scheduled refresh on the semantic model so it keeps pace with Viva's weekly update.

Pick how your data gets in. Every path ends the same way: open the .pbit, point its five parameters at the CSVs you produced, click Apply, then Refresh. Start with a one-month window on your first run.
The whole report is built from the Microsoft 365 unified audit log. With logging off, every Purview search and every PAX run returns zero rows and nothing tells you why. It does not backfill, so a tenant that just switched it on starts thin.
Audit permission change (April 2026): Microsoft now enforcesAuditLogsQuery.Read.Allfor the Graph audit query API. With only the legacyAuditLog.Read.All, Copilot audit calls appear to succeed and silently return zero records. PAX v1.10.9 and later request the correct scopes, so grant admin consent for the new permission.
One command pulls your audit data, rolls it up into the shape Power BI needs, and exports your Entra users at the same time. This is the recommended path, and it skips the processing step entirely.
Easiest: open Mini-Kitchen, choose the M365 Usage Dashboard preset, work down the builder, and copy the command it writes. Mini-Kitchen runs entirely in that browser tab, so it pulls no tenant data and stores no secrets.
Or write it yourself:
.\PAX_Purview_Audit_Log_Processor.ps1 -IncludeM365Usage -Rollup -IncludeUserInfo -StartDate "2026-04-21" -EndDate "2026-05-21"
Download the PAX script from the PAX repository (Mini-Kitchen's own Download PAX script button takes you to the same place), open PowerShell 7, paste the command, and sign in. This is the step that reaches your tenant. It produces five import-ready CSVs:
| File | Contents |
|---|---|
Rollup | Per-user, per-app, per-day event counts |
UserStats | One row per user with tiers and engagement segments |
SessionCohort | One row per (user, app) with a session-count bucket |
SessionStats | Copilot prompt / session / response counts |
EntraUsers | Department, job title, and Copilot license status |
Get the current M365 Usage Dashboard template through the README and setup guidance. Open it, go to Transform data → Edit parameters, and paste the five file paths:
M365 Rollup Data | Rollup CSV |
|---|---|
M365 User Stats Data | UserStats CSV |
M365 Session Cohort Data | SessionCohort CSV |
M365 Session Stats Data | SessionStats CSV |
Entra Users Data | EntraUsers CSV |
Apply, Refresh, then Save as .pbix so you do not re-enter the paths next time.
No PAX? Export from Purview by hand, then flatten the exports with the Python processor that ships in the repo.
In Purview audit search, run four searches over the same date range, leaving Users blank. Export each with Download all results and save under the filename shown. Keep each pull under the 50K / 100K row cap; split a pull into weekly chunks if it hits the cap.
| Pull | Paste into the filter | Save as |
|---|---|---|
| Files (SharePoint / OneDrive) | FileAccessed, FileViewed, FilePreviewed, FileModified, FileDownloaded, FileUploaded | pull1_files.csv |
| Outlook / Mail | MailItemsAccessed, MailboxLogin, Send | pull2_outlook.csv |
| Teams | MessageSent, MessageRead, MessagesListed, ChatRetrieved, ChatCreated, MeetingParticipantJoined, MeetingStarted, MeetingEnded, MeetingParticipantDetail, MeetingDetail, TeamsSessionStarted | pull3_teams.csv |
| Copilot + Connected AI | CopilotInteraction, ConnectedAIAppInteraction (paste into Record types, not Activities) | pull4_copilot.csv |
Extend the Teams pull by +1 day: Purview batches meeting events up to 24 hours after a meeting ends.
In the Microsoft Entra admin center go to Identity → Users → All users → Download users, including userPrincipalName, displayName, department, jobTitle, and assignedLicenses.
hasLicense column
The plain Entra download has no hasLicense flag, so the Enablement Strategy quadrant and license ranking cannot tell licensed users apart. The repo's PowerShell export snippet auto-detects every Copilot SKU in your tenant and adds the column. PAX (tab 1) adds it for you.The raw Purview CSVs carry a nested AuditData JSON column Power BI cannot read. Use the current usage-bundle processor and command from the M365 Usage Analytics README and setup guidance to turn the files, Outlook, Teams, and Copilot pulls into the four import-ready CSVs.
Have a single raw PAX CSV instead of four pulls? Use --pax "Purview_Export.csv".
--skip-precompute or --no-session-stats: Department Readiness needs UserStats and SessionStats, and Adoption Momentum needs SessionStats plus at least 60 days of history for its current-versus-previous 30-day comparison.Get the current template through the M365 Usage Analytics README and setup guidance. Open it and set the same five parameters as tab 1: the four processor outputs (Rollup, UserStats, SessionCohort, SessionStats) plus your Entra CSV. Apply, Refresh, Save as .pbix.
Want to see the finished report before you touch your tenant? Open the synthetic demo.
Follow the M365 Usage Analytics README sample-data guidance for the current demo report, then open it in Power BI Desktop. It ships with cached synthetic @example.com users, so every one of the ten pages fills in with no parameters and no exports.
This file is for preview and testing only. When you are ready for your own tenant, use tab 1 or tab 2, both of which drive the .pbit template.

This report runs two ways: use the self-refreshing Direct Query variant for live Viva Insights data, or use the CSV variant for Microsoft 365 admin center (MAC) exports, with two CSV exports joined on user principal name (Copilot credit consumption and a Microsoft Entra org directory). Wire up your own data on either path, or explore the shipped demo dataset first.
Produce the two CSV exports, point the template at them, refresh.
Microsoft 365 admin center → Copilot → Cost management → Consumption tab → Export CSV. One row per user with the monthly credit limit, credits used, session count, % used and last activity date.
Microsoft Entra admin center → Identity → Users → All users → Download users (or Microsoft Graph GET /users). Supplies department, jobTitle, city, country and manager — the org context the credit export doesn’t carry.
Export 1 (credit consumption): Display Name, User Principal Name (join key), Monthly credit limit, Monthly credits used, User ID, Microsoft 365 Copilot license, Last activity date, Session Count, % Used.
Export 2 (Entra directory): userPrincipalName (join key), displayName, department, jobTitle, jobFamily, city, country, costCenter, manager, businessUnit.
jobFamily,costCenterandbusinessUnitare not standard Entra fields — an admin maps them in (for example from an HR feed), or those slicers stay blank. Every user in the credit export needs a matching directory row, or their credits aren’t attributed to a department.
Get the current Cowork Chargeback CSV template through the CreditUsage README and setup guidance. Open it, set CreditCsvPath to the credit CSV and EntraCsvPath to the directory CSV, click Load.
Click Refresh and confirm the Consumption visuals populate (total credits used, allowance, utilization %, top user) and all six pages render.
Need to group by something the standard fields don’t cover (say CustomDivision)? Add it as an extra column in your Entra directory CSV (Export 2), keyed by userPrincipalName — the template passes any extra directory column straight through. Click Refresh and the column shows up on the Org table in the Data pane. Then select the Slice By table and, in the formula bar, add one line before the closing }:
("Custom Division", NAMEOF('Org'[CustomDivision]), 11)
Label shown in the dropdown · the real column · the next sort number. It appears instantly as a new checkbox in the Slice By slicer on every page.
Connect Power BI straight to Viva Insights. No CSV exports, and once published it refreshes on a schedule, so monitoring is hands-off.
In the Viva Insights Advanced Insights portal, an analyst builds and saves a custom Consumption query, ticking Select spending policy and employee attributes so policy, limits and org context travel on the rows. Note the query’s Partition and Query identifiers.
Get the current Cowork Chargeback DirectQuery template through the CreditUsage README and setup guidance, then open it in Power BI Desktop.
VivaPartitionId | The Partition: GUID shown in the Advanced Insights portal |
|---|---|
VivaQueryId | The identifier of your saved Consumption query |
Optional: BillingPeriodWeeks (default 4) sets how many trailing weeks roll into the current billing window.
Sign in to the Viva Insights connector with your organizational account, click Load, then Refresh. Every page populates from live tenant data.
Publish to the Power BI Service and set a scheduled refresh on the dataset. The report then keeps itself current with no manual step.
To group by a custom employee attribute (say CustomDivision), include it in your Viva Consumption query’s employee attributes so it rides along on the people rows. Click Refresh and the template carries custom people columns through to the Org table, so your attribute appears in the Data pane. Then select the Slice By table and, in the formula bar, add one line before the closing }:
("Custom Division", NAMEOF('Org'[CustomDivision]), 11)
Label shown in the dropdown · the real column · the next sort number. It appears instantly as a new checkbox in the Slice By slicer on every page.
VivaQueryId parameter so refresh keeps resolving.Want to see all six pages before you collect anything real? The repo ships a synthetic dataset in the exact format the template expects.
Get the current Cowork Chargeback CSV template and sample data through the CreditUsage README and setup guidance, then open the template in Power BI Desktop.
When prompted, set the parameters to the two files in the demo data folder, then click Load:
CreditCsvPath | Path to cowork_billing_synthetic.csv |
|---|---|
EntraCsvPath | Path to entra_org.csv |
All six pages render on synthetic data: Consumption, Chargeback (PayGo), Prepaid Allocation, Optimization, Forecast, and Glossary.
The report ships two roles: All Org (Admin) (unrestricted, for finance and program leads) and Department Admin (only the viewer’s own department, via a USERPRINCIPALNAME() filter). Set it up in two places:

This web app reads two CSV exports in your browser: a Microsoft Entra user export and a Copilot credit consumption export. Nothing is uploaded. Drop the files in, reconcile to your invoice total, then export the journal or line items.
Export the two CSVs, open the app, then drop both files into the upload area.
Microsoft Entra admin center (entra.microsoft.com) → Identity → Users → All users → Download users (CSV; supplies department, job title, cost centre, business unit; you can load more than one).
Microsoft 365 admin center (admin.microsoft.com) → Copilot → Cost management → Consumption tab → Export CSV (one row per user with credits consumed).
The join key is user principal name. The Entra file supplies department, job title, cost centre and business unit. The credit export supplies one row per user with credits consumed.
You can load more than one Entra CSV if your tenant exports users in batches. Files are parsed locally in the page.
Open Cowork Chargeback, then drop the Entra export and the Copilot credit export into the app. It joins the files on user principal name and produces the chargeback.
Enter your Microsoft invoice total so the app can reconcile billed cost against the file-level chargeback calculation.
Export the journal or the line items as CSV for finance review or downstream allocation.
Run the entire chargeback on synthetic data before you use tenant exports.
Open the app and click View Demo.
The demo uses synthetic users and credits. Nothing is uploaded.

This web app reads two CSV exports in your browser: a Microsoft Entra user export and a Copilot credit consumption export. Nothing is uploaded. It joins the files on user principal name, assigns usage cohorts and lets you tune spend-tier policies before exporting assignments.
Export the two CSVs, open the app, then drop both files into the upload area.
Microsoft Entra admin center (entra.microsoft.com) → Identity → Users → All users → Download users (CSV; supplies department, job title, cost centre, business unit; you can load more than one).
Microsoft 365 admin center (admin.microsoft.com) → Copilot → Cost management → Consumption tab → Export CSV (one row per user with credits consumed).
The two files are joined on user principal name. Each user is placed in a usage cohort by percentile of credits used, then given a recommended spend-tier policy.
The Entra file supplies department, job title, cost centre and business unit so you can review policies by organization context.
Open Cowork Policy Helper, then drop in the Entra export and the Copilot credit export.
Adjust each policy's monthly credit allowance live and review how the recommendations change.
Override individual users where needed, then export the assignments as CSV.
Explore the policy workflow on synthetic data before you use tenant exports.
Open the app and click View Demo.
Files are parsed locally and demo assignments are held in memory only.

This report reads three admin exports: one GitHub Copilot Usage Insights JSON, one GitHub Enterprise Members CSV, and one Entra users CSV. Point the Power BI template at local files, or use the sample files first.
Produce the three exports, open the template, and paste each full local file path when prompted.
Sign in as a GitHub Enterprise Owner, go to github.com/settings/enterprises, pick your enterprise, then use the top-right (...) menu → Insights → Copilot Usage → select Last 28 days → Download. Save the .json file locally.
In the same enterprise, go to People → Members → right side CSV Report. Fewer than 1000 members downloads immediately. For 1000 or more members, GitHub emails a download link.
In Microsoft Entra admin center, sign in as User Administrator or Global Reader, then Identity → Users → All users → Download users. Include at least UserPrincipalName and Department.
Get the current GitHub Copilot Usage Analytics template through the GitHub Copilot Impact README and setup guidance, then open it in Power BI Desktop. When prompted, paste the full local path to the GitHub Usage JSON, the GitHub Members CSV, and the Entra users CSV, then click Load.
To refresh later, re-export the files, keep the same paths or update the parameters, and reload the template.
The inputs are 2 CSVs + 1 JSON, not 3 CSVs. There is no API token or PAT anywhere in this setup, access is through interactive admin sign-in.
The GitHub export is empty unless the enterprise Copilot usage metrics policy is opted in under AI Controls. Department segmentation is blank unless SAML SSO is enforced between GitHub and Entra.
The repo ships sample files that let every page render before you collect tenant data.
From GHCP Sample Files, download these three files:
membership_export - Copy.csv | Sample GitHub Enterprise Members CSV |
|---|---|
OrgData file.csv | Sample Entra org CSV |
test output.json | Sample GitHub Copilot Usage Insights JSON |
Get the current template through the GitHub Copilot Impact README and setup guidance. Open it, point the three prompts at those files, and click Load.
All pages render on synthetic data for the enterprise nicknamed Byte-Sized-Donuts.

This Power BI template reads the Viva Insights GitHub Copilot Export query output. Use the reliable CSV route first, or switch the single DataSource parameter to Viva when using the connector route.
Run the Viva Insights GitHub Copilot Export, then choose CSV files or the Viva connector by setting DataSource.
In Viva Insights, go to Analysis → Analyst Workbench, run a GitHub Copilot Export query, then open Analysis results. This requires the Insights Analyst role. GitHub Copilot data in Viva Insights is in preview.
From Analysis results, download the CSVs and put them flat in one folder. Get the current template through the GitHub Copilot Panel README and setup guidance. Open it, leave DataSource=CSV, set DataFolder to that folder, and click Load.
PersonGitHubActivityMetrics.csv, GitHubActivityBreakdownByFeatureMetrics.csv, GitHubActivityBreakdownByModelFeatureMetrics.csv, GitHubActivityBreakdownByLanguageFeatureMetrics.csv, GitHubActivityBreakdownByLanguageModelMetrics.csv, and PeopleMetaData.csv.
In Analysis results, click the query link icon, then Copy identifiers and connect to Power BI to get the Partition ID and Query ID. Set DataSource=Viva, then set PartitionId and QueryId. In connector settings use Schema=Pivoted, Granularity=Row-level, and Connectivity=Import.
DataSource | CSV or Viva |
|---|---|
DataFolder | Folder that contains the six CSV files for the CSV route |
PartitionId | Viva connector route only |
QueryId | Viva connector route only |
The shipped template connector TableName options currently carry the CSV names, so the CSV route is the reliable one today.
The sample data route is the preloaded Route 1 path, no tenant connection required.
Download the CSVs from the sample-data folder. The sample contains 200 people, 16 orgs, and 26 weeks. Copy the files flat into the default folder C:\GitHub Copilot Panel\Data.
Get the current template and sample files through the GitHub Copilot Panel README and sample-data guidance. Open the template, leave DataSource=CSV, and click Load. No parameter change is needed when the files are in the default folder.
When you later point at real export data, set config is_synthetic=0.

This tool reads your existing local GitHub Copilot session logs on your own machine, including GitHub Copilot CLI session-state events and VS Code Copilot Chat sessions. Nothing is uploaded. It requires Python 3.10+, the GitHub CLI (gh) signed in for AI analysis, and real Copilot usage history. A machine with no Copilot history produces an empty report.
Install the plugin inside Copilot CLI, then run the report against your own recent sessions.
In a Copilot CLI session run /plugin install whatidid@awesome-copilot.
Run whatidid. The default lookback is 7 days. You can also pick it via /agent.
The tool writes a self-contained HTML report for the selected period. If there is no local Copilot history, the report is empty rather than populated with sample data.
It reads GitHub Copilot CLI logs under ~/.copilot/session-state/<uuid>/events.jsonl and VS Code Copilot Chat session history on the same machine. It is for GitHub Copilot, not Microsoft 365 Copilot.
Install the same agent plugin from VS Code, then ask Copilot Chat for a report.
Open the Command Palette, choose Chat: Plugins (@agentPlugins), search whatidid, then click Install.
In Copilot Chat ask, for example, give me a 7-day Copilot report.
The plugin uses the Copilot history on your machine. There is also a separate first-party VS Code extension named What I Did with a Report Builder panel.
You need real GitHub Copilot activity on the machine and gh signed in when AI analysis is requested. The tool does not ship demo or sample data.
Clone the repo and run the Python CLI directly when you want to inspect or modify the tool.
Run gh auth login one time.
git clone https://github.com/microsoft/What-I-Did-Copilot.git
cd What-I-Did-Copilotpython whatidid.py
Common flags: --7D, --14D, --30D, --60D, --90D, --date 2026-03-19, --from ... --to ..., --refresh, --email [addr], --out-dir <dir>.
The output is report_<label>.html, saved to the script folder and auto-opened.

This skill reads your own Copilot Cowork session history under OneDrive Documents/Cowork/ through Microsoft Graph, using the Cowork app allow-list. It is not reading local disk files. You need Copilot Cowork access and existing Cowork session history; synthetic demo artifacts are available only for preview.
Install the skill zip in Cowork and let the agent read your allowed OneDrive Cowork session folders.
Go to the Cowork ROI report repository on GitHub and follow the README to get the latest skill zip. Do not unzip it.
Open a Cowork session at aka.ms/cowork, click + to attach the zip, then send Add this skill.
Ask Generate my impact summary report. The skill asks two questions: the period, 7, 15 or 30 days, and whether to run once or automate plus email.
The data source is your OneDrive Documents/Cowork/ session history. Access is through Microsoft Graph and the Cowork app allow-list, not local files.
Run the bundled scripts directly when you are developing the skill or testing report generation.
You still need a working working/cowork_raw.json, normally produced by the Cowork agent.
Use standard-library Python 3. The scripts reconcile taxonomy, classify sessions, compute metrics, then build the report.
python reconcile_taxonomy.pypython classify.pypython compute.pypython build_report.pypython to_csv.py optional
This path is for development. It does not replace Cowork access or the agent-produced raw session harvest.
Use the prebuilt demo to see the output shape without Cowork access, OneDrive data or a license.
Open cowork-roi-report-demo.pdf.
The repo ships synthetic inputs at skill/examples/demo_sessions.json behind the prebuilt demo report.
The demo does not connect to OneDrive or read Cowork session folders. For your own impact report, use the Cowork skill path.

This template has one real connection mechanism: three Power Query parameters that point at local CSV files. The M365 Admin Copilot usage export is required. Survey responses and Entra users are optional. No demo data ships with the repo.
Export the files, open the .pbit, set the parameters to local paths, and load.
Go to admin.microsoft.com → Reports → Usage → Microsoft 365 Copilot → select the 180-day period → Export in the top right. The file name looks like CopilotActivityUserDetail_YYYY-MM-DD.csv.
In the M365 admin center, go to Settings → Org settings → Reports, uncheck Display concealed user, group, and site names in all reports, wait 48 hours, then re-export.
The repo does not ship a downloadable survey file, and its survey link is broken. Recreate the 12 recommended questions in Microsoft Forms or another survey tool from the README table. Capture an Email or UPN field by turning on Forms Record name or by adding a required email question, then export responses as CSV.
Overall satisfaction, Helps me work faster, Improves work quality, Improves productivity, Benefits my role, Copilot readiness, Agents usage (all 1-5); Time saved using Copilot (buckets); Time repurposed; Use cases; Barriers to adoption; What else? (free text).
Go to entra.microsoft.com → Identity → Users → All users → Download users. This supplies organization, department, and display names.
Get the current Adoption & Sentiment template through the README and setup guidance, then open it in Power BI Desktop, Feb 2025 or later. Set Copilot Activity (MAC) to the required usage CSV full path with no quotes, Raw Survey Data to the survey CSV path or / to skip, and User Details (Entra) to the Entra CSV path or / to skip. Click Load.
Re-export monthly and append to build month-over-month trends.
The Comments page uses a Smart Narrative visual. It needs Power BI Pro or PPU with Copilot in Power BI enabled, otherwise that one visual is blank.

This dashboard has three genuine setup paths. Start with the Copilot Dashboard export if you only need your own view, or use Viva Insights person-query paths when you need analyst-controlled CSV or Direct Query inputs. No demo data ships with the repo, and report data refreshes weekly.
You need an M365 Copilot license and Power BI Desktop. The Viva person-query paths also require Viva Insights Analyst Workbench access.
Use exports from the Copilot Dashboard in Viva Insights. Load any combination of Copilot, Agent, and Cowork data.
Open the Copilot Dashboard via Viva Insights, then click Export data. This downloads your Copilot usage CSV.
For Agent data, export the folder containing AgentMetadata.csv, PersonAgentResponsesMetrics.csv, PersonAgentCreditsRetentionMetrics.csv, and PeopleMetadata.csv. For Cowork data, export the folder containing PersonServiceCreditsMetrics.csv, PeopleMetaData.csv, and SpendingPolicyMetadata.csv.
Get the current Personal Copilot, Agent, & Cowork template through the README and setup guidance. Open it, set CopilotCsvFilePath, AgentDataFolderPath, and CoworkDataFolderPath. Leave any path you do not have blank, then click Load.
Use the PersonId slicer to filter the report to yourself.
Create a role named ViewOwnData with [PersonId] = USERPRINCIPALNAME(), then assign it in the Power BI Service.
Build one Viva Insights person query and export it as CSV.
Go to analysis.insights.cloud.microsoft → Analysis results → Create analysis → Person query → Set up analysis.
Set Time period to Last 6 months (rolling), Group by to Week, filter Is Active = True, add attributes Organization, FunctionType, and TimeZone, and under Microsoft 365 Copilot select All metrics. Missing even one metric can leave visuals blank.
Save and run the query, wait for Completed, then export the CSV.
Open Viva Insights Personal Dashboard V9 (CSV Import pbit).pbit from the repository and enter the CSV path.
Use the same person-query build, then connect the template to the completed query IDs.
Use the same setup as the CSV path: Last 6 months rolling, group by Week, Is Active = True, attributes Organization, FunctionType, TimeZone, and all Microsoft 365 Copilot metrics.
From the completed query, copy the Partition ID and Query ID.
Open Viva Insights Personal Dashboard V9 (Direct Import pbit).pbit and enter the Partition ID and Query ID.

CustomizeCopilot is a library of ready-made add-on pages that graft onto an existing Super User Adoption or Super User Impact report, version 4 or later. The data lives in that host report, fed by Viva Insights, and the add-on inherits it. Today the library has one live add-on: Champion-ID.
Copy the add-on pages into a host report that is already connected to your Viva Insights data.
Open your Super User Adoption or Super User Impact report, version 4 or later, in Power BI Desktop. It should already be connected to your data.
Follow the CustomizeCopilot README and Champion-ID setup guidance for the current report, then open it in a second Power BI Desktop window. It ships with sample data and opens instantly.
In the host report, add a new blank page.
In Champion ID.pbix, click the empty canvas, press Ctrl+A, then Ctrl+C. In the host report blank page, press Ctrl+V. The visuals land with formatting.
Repeat the copy and paste flow for page 2, Extended Analytics.
The 25 required DAX measures auto-bind on host v4+. Verify the visuals resolve, then save the host report.
Open Champion ID.pbix on its own. It ships with sample data and renders both pages populated, so you can inspect the add-on before grafting it into your host report.
For source-controlled .pbip reports, copy the page GUID folder under definition/pages/ and register the GUID in pages.json pageOrder, adding any missing measures to the model.

This web app reads two CSV exports in your browser: a Microsoft Entra user export and a Copilot credit consumption export. Nothing is uploaded. It joins the files on user principal name and prices usage into FOCUS-aligned cost columns.
Export the two CSVs, open the app, then drop both files into the upload area.
Microsoft Entra admin center (entra.microsoft.com) → Identity → Users → All users → Download users (CSV; supplies department, job title, cost centre, business unit; you can load more than one).
Microsoft 365 admin center (admin.microsoft.com) → Copilot → Cost management → Consumption tab → Export CSV (one row per user with credits consumed).
The app joins the two files on user principal name and prices usage into the FOCUS cost columns: List, Contracted, Effective and Billed.
Use department, cost centre or business unit from the Entra export as the allocation lens.
Open the FinOps & FOCUS Cost Report, then drop in the Entra export and the Copilot credit export.
Tune the list and contracted rates live, then allocate by department, cost centre or business unit.
Use the List, Contracted, Effective and Billed cost columns to compare finance views and export the result for follow-up.
Preview the full FOCUS-aligned cost view before you use tenant exports.
Open the app and click View Demo Report.
The demo uses synthetic data and runs in the browser.
Start with manual assumptions and no files, or use supported exports to inform a scenario. Selected files are processed locally in your browser.
Open Cowork Credit Forecasting. Choose Modeling to start without a file; users, usage, growth, credits, prices and budget start at zero, so enter your own values.
In Your Data, try the fictional demo or import a Viva Insights Consumption query ZIP or a joined CSV with usage and organization details.
To run a Viva Insights Consumption query: you need the Insights Analyst role and access to the global partition. Manual modeling and the demo do not require these permissions.
ServiceName condition for one Microsoft 365 Copilot service, such as Cowork. Select the organizational attributes you want to explore.See Microsoft's query instructions. A service-level ZIP must contain a single service; GitHub credits are separate. Check the source yourself: opaque service names or IDs and joined CSVs cannot be authoritatively verified by the app. Person-level exports require appropriate data handling.
Review the data and assumptions, select a group and starting month if using imported data, then adjust the inputs in Modeling. For a first imported-data forecast without an edited plan, prepaid credits, prices, budget and growth start at zero unless you explicitly select historical growth. Enter and review these values; the fictional demo uses its own defaults. Applying real data to an edited demo or unknown-origin plan requires confirmation and leaves a warning to review any retained assumptions. Use day or calendar-month data for calendar-month forecasts; check historical months for partial periods or incomplete data. A single month's change can exaggerate compounded growth, so review the rates before relying on them.
The consumption value is a scenario based on your assumptions, not an invoice or a billing reconciliation. A modeled shortfall does not guarantee that pay-as-you-go is available or enabled; confirm your actual commercial terms and settings. Review exported forecasts and presentations for sensitive inputs before sharing.

There is no file to upload, no export to request and no admin to chase. The model runs entirely in your browser: type task volumes and assumptions, then read the live assisted-hours, value and ROI outputs.
Open the model and enter the scenario assumptions you want to test.
Open the Cowork ROI Model. It runs locally in the browser.
Give the scenario a clear name so exported or shared outputs are easy to identify later.
Enter your task volumes per category. Adjust minutes per task if you want to depart from the research band; use the reset control to return minutes to the band.
Minutes-per-task bands are pre-filled from published research. The model keeps the assumptions visible so finance and business owners can challenge them before relying on the ROI result.
Set annual cost, blended hourly rate, monthly active users and unlicensed seats.
Review assisted hours, value and ROI as they recalculate live, then export or share the scenario.

The calculator reads one CSV: a Viva Insights person-query export, one row per person per week. Everything is parsed in your browser — the file is never uploaded anywhere. Bring your own export, or open the built-in demo first.
Run one Viva person query, open the calculator, drop the CSV in.
Open the Viva Insights Analyst Workbench (sign-in required), then Analysis results → Create analysis → the Person query card → Set up analysis. Set Time period to Last 6 months (rolling), Group by to Week, and the filter to Is Active = True. Under metrics choose Microsoft 365 Copilot — all metrics, add the attributes Organization, FunctionType and TimeZone, then run it and download the result as CSV from Analysis results.
At minimum the export needs PersonId, MetricDate and Total Copilot actions taken. Grouping by Week is required — the usage-tier cohorts need at least 12 consecutive weekly rows per person (the 9-of-12-weeks habit rule). British-English and Spanish column names are auto-translated, so a localized export works as-is.
The retired Super Usage heatmap export is rejected on upload — its wide YYYY-MM-DD Metric columns are detected and the app tells you to re-export as a Person query. The person query is the only accepted format.
Want to see the exact shape first? sample-data.csv is a real person-query example.
Open the ROI Calculator — it runs entirely in your browser, nothing to install. Prefer an air-gapped copy? Download & run locally (a ~2 MB ZIP with a one-click launcher, zero external calls).
Pick an industry (or a custom blended hourly rate), the licence cost per user per month, the minutes saved per Copilot action, and the analysis period. Every figure recalculates live as you sweep these.
Drag the export onto the upload area. Parsing happens locally in the page — the file never leaves your machine — and the Executive Summary, per-org value, usage-tier ROI, break-even, expansion and unlicensed-opportunity views populate immediately.
See the whole value story on a synthetic dataset before you touch tenant data.
Open the demo report (or click View Demo inside the calculator). It loads a built-in Viva person-query dataset — 300 people, 7 organizations, 14 weeks — with sensible defaults (350 licences, $30/user/mo, 6 min/action, $78/hr). No file to pick, no sign-in.
Executive Summary, Organizations, Usage Tier Value Distribution, ROI & Forecast, Expansion Projections, Unlicensed User Opportunity Cost and the Adjusted Copilot-Assisted-Hours model all render on the sample data, exactly as they will on yours.
Every export from demo mode is watermarked DEMO DATA, so swap in your own person-query CSV (tab 1) before sharing anything with a sponsor.