Copilot Analytics Scripts

This page contains specialized scripts for analyzing Microsoft Copilot usage data from Viva Insights.

Beyond covering key analyses around Copilot usage volume and breadth (range of actions and applications), these scripts also include a method for measuring Copilot habituality based on behavioral research. This approach determines whether a user can be considered a habitual Copilot user, enabling visualization through user segmentation that identifies Power Users and Habitual Users within an organization. This segmentation provides a framework for continuously tracking Copilot adoption success and measuring the effectiveness of your deployment strategy.

See our DAX Calculated Columns page for detailed instructions on how to identify Copilot Usage Segments using Power BI templates and pre-built DAX formulas.

For more information on the Copilot Usage Segments, see this introduction.

For more inspiration on analyzing Copilot adoption and impact, have a look at our advanced examples playbook.

Advanced Analysis Scripts

Copilot Advanced Analysis

📄 copilot-analytics-examples.R

Purpose: Comprehensive analysis of Copilot usage patterns and trends
Prerequisites: vivainsights R package, Copilot usage data
Key Analysis: Usage segmentation, trend analysis, adoption metrics
📥 Download

📄 copilot-analytics-examples.py

Purpose: Comprehensive analysis of Copilot usage patterns and trends
Prerequisites: vivainsights Python package, Copilot usage data
Key Analysis: Usage segmentation, trend analysis, adoption metrics
📥 Download

📓 copilot-analytics-examples.ipynb (Jupyter Notebook)

Purpose: Interactive analysis of Copilot usage with visualizations
Format: Jupyter Notebook
Prerequisites: vivainsights Python package, Copilot usage data
Key Features: Step-by-step analysis, interactive visualizations
📥 Download

Copilot Adoption Journey and Ways of Working

Where the scripts above give you a tour of the core Copilot metrics, this notebook is the next step: an end-to-end assessment of how far adoption has travelled and what sustained use is associated with. Reach for it once you have at least 12 weeks of Copilot data and need to answer a leadership question rather than explore the data.

Start with	When
Copilot Advanced Analysis (above)	You are getting oriented, and you want the core metrics, segmentation, and standard visuals.
Copilot Adoption Journey (below)	You have 12+ weeks of data and need cohorts, habit formation, conversion targeting, and associated ways of working.

📓 copilot-adoption-journey-analysis.ipynb (Jupyter Notebook)

Purpose: End-to-end assessment of Copilot reach, habit formation, conversion opportunity, and associated ways of working
Format: Jupyter Notebook
Prerequisites: vivainsights, pandas, NumPy, SciPy, statsmodels, matplotlib, and a Copilot-rich Person Query export with at least 12 weeks of data. Add pyarrow if you are loading a Parquet export
Key Features: 12-week Power/Habitual/Novice/Low/Non-user segmentation, entry-cohort and function analysis, native Viva Insights scans, adjusted associations, and same-person diagnostics
Interpretation: Observational associations only. Set INPUT_FILE to your own export before running; any metric or attribute your query lacks is reported as unavailable and skipped
Related: Copilot Usage Segments for the segment definitions, and Causal Inference when you need an effect rather than an association
📥 Download

This example is currently available in Python only. Switch the toggle above to Python to see the notebook details, or use the R scripts in the other sections on this page.

Adoption Tracking & Causal Analysis

The examples in this section focus on measuring Copilot adoption credibly over time. The two causal examples build small, clearly labelled seeded simulations so that the models have something to recover; swap the simulation block for your own export before drawing conclusions.

Copilot Usage Segments Over Time

Sums individual Copilot-action columns, classifies each person-week with identify_usage_segments(version = "12w"), and visualises how the mix of Power, Habitual, Novice, Low, and Non-users evolves week by week.

📄 copilot-usage-segments-trend.Rmd

Purpose: Track how the mix of Copilot usage segments evolves week by week
Format: R Markdown
Prerequisites: vivainsights R package, dplyr, tidyr, ggplot2, scales
Key Features: identify_usage_segments (12-week rolling), stacked-area segment mix, action trend
📥 Download
🌐 View HTML Output

📄 copilot-usage-segments-trend.py

Purpose: Track how the mix of Copilot usage segments evolves week by week
Prerequisites: vivainsights Python package, pandas, numpy, matplotlib
Key Features: identify_usage_segments (12-week rolling), stacked-area segment mix, action trend
📥 Download

Difference-in-Differences Metric Scan

Runs a within-person DiD per metric across two both-licensed groups (Power vs Low Copilot users) and assembles the effects, confidence intervals, and significance into one sortable table plus a forest plot, honestly surfacing the metrics that do not move.

📄 did-metric-scan.Rmd

Purpose: Run a within-person DiD per metric (Power vs Low Copilot users) into one sortable table
Format: R Markdown
Prerequisites: vivainsights R package, fixest, dplyr, tidyr, ggplot2, purrr, scales
Key Features: Per-metric TWFE DiD, significance stars, forest plot, honest reporting of null effects
📥 Download
🌐 View HTML Output

📄 did-metric-scan.py

Purpose: Run a within-person DiD per metric (Power vs Low Copilot users) into one sortable table
Prerequisites: vivainsights Python package, linearmodels, pandas, numpy, matplotlib
Key Features: Per-metric TWFE DiD, significance stars, forest plot, honest reporting of null effects
📥 Download

Event-Study & Difference-in-Differences

Aligns each adopter on their own event time, checks the parallel-trends assumption before trusting a single headline number, and reads the within-person change net of a non-adopting control.

📄 event-study-did.Rmd

Purpose: Measure within-person behaviour change around Copilot adoption with a TWFE event-study/DiD
Format: R Markdown
Prerequisites: vivainsights R package, fixest, dplyr, tidyr, ggplot2, scales
Key Features: Event-time alignment, pre-trend check, person + week fixed effects, z-scored composite index
📥 Download
🌐 View HTML Output

📄 event-study-did.py

Purpose: Measure within-person behaviour change around Copilot adoption with a TWFE event-study/DiD
Prerequisites: vivainsights Python package, linearmodels, pandas, numpy, matplotlib
Key Features: Event-time alignment, pre-trend check, person + week fixed effects, z-scored composite index
📥 Download

Interactive Report Templates

Where the scripts above answer one question at a time, these two templates are complete multi-page HTML reports you can open in a browser and hand to a stakeholder. Both are built with R Markdown and flexdashboard, and both ship with synthetic data so they render end to end before you point them at your own export.

They are built on two of the newer Viva Insights query types, both available from the Customised query tab under Create analysis in the Viva Insights analyst experience. See the AI cost query documentation for the official schema and access details:

Query	What it adds	Used by
Consumption query	Copilot credit and token consumption per person and per delegated task type	Copilot Consumption and Ways of Working
GitHub query	GitHub Copilot activity, model mix, and language mix for linked developer accounts	Developer Experience and Copilot

Treat the numbers in the published outputs as illustrative. The simulations are seeded so the charts have something to show, and each report states its assumptions in an appendix.

Copilot Consumption and Ways of Working

Reads the Consumption query alongside a Person Query to separate how much Copilot is consumed from how it is consumed. It covers the concentration of token consumption across users, credit intensity as a cost-mix measure, delegated task types, and the association between consumption and collaboration patterns.

Token consumption is typically right-skewed, so the report leads with a concentration curve and banded percentiles rather than an average that a small number of heavy users can pull upward.

📄 copilot-consumption-ways-of-working-simulation.Rmd

Purpose: Separate Copilot consumption volume from consumption mix, and relate both to ways of working
Format: R Markdown (flexdashboard), seven pages
Prerequisites: vivainsights R package, dplyr, tidyr, ggplot2, scales, flexdashboard, knitr
Key Features: Concentration curve and percentile bands, credit intensity per 1,000 tokens, delegated task-type mix, usage segments, function drill-down
Interpretation: Associations only. Credit intensity is a cost-mix measure rather than a measure of value or quality
🌐 View the live report
📥 Download

This template is currently available in R only. Switch the toggle above to R to see the details, or open the live report directly.

Developer Experience and Copilot

A manager view of developer working conditions and recorded AI use, combining the GitHub query with a Person Query so that GitHub Copilot and Microsoft 365 Copilot are read together rather than in isolation. It establishes a baseline before it looks at change, and it keeps eligibility and coverage visible rather than treating missing activity as zero.

Joint product use is reported with coverage exceptions retained, so a team that simply lacks a licence is never mistaken for a team that has chosen not to adopt.

📄 github-copilot-developer-productivity-simulation.Rmd

Purpose: Give an engineering manager a baseline of developer working conditions and joint GitHub and M365 Copilot use
Format: R Markdown (flexdashboard), seven pages
Prerequisites: vivainsights R package, dplyr, tidyr, ggplot2, scales, flexdashboard, knitr
Key Features: Team league table, ranked distributions of meeting and uninterrupted hours, joint product-use footprint with coverage exceptions, model and language mix, trends
Interpretation: Associations only. Acceptance rate is not a measure of code quality, after-hours activity is not a measure of burnout, and calendar space does not establish coding time
Also required: github-developer-experience-helpers.R and render-github-developer-experience.R
🌐 View the live report
📥 Download

This template is currently available in R only. Switch the toggle above to R to see the details, or open the live report directly.

📖 Documentation for both templates, covering the expected schema of each query, the simulated data contracts, and how to swap the simulation block for your own export.

Power BI Integration

DAX Calculated Columns

📁 DAX Calculated Columns

Purpose: Pre-built DAX formulas for Copilot usage segmentation in Power BI
Language: DAX
Format: Individual .dax files
Prerequisites: Power BI Desktop, Copilot usage data

Available Columns:

12-Week Rolling (RL12W) - Recommended for long-term analysis

📄 _Total Copilot actions_RL12W.dax: Average weekly actions over 12 weeks
📄 _IsHabit_RL12W.dax: Habit formation indicator (9+ weeks of usage)
📄 _CopilotUsageSegment_RL12W.dax: User segmentation (Power/Habitual/Novice/Low/Non-users)

4-Week Rolling (RL4W) - Recommended for short-term/pilot analysis

📄 _Total Copilot actions_RL4W.dax: Average weekly actions over 4 weeks
📄 _IsHabit_RL4W.dax: Habit formation indicator (4 weeks of usage)
📄 _CopilotUsageSegment_RL4W.dax: User segmentation (Power/Habitual/Novice/Low/Non-users)

📖 DAX Documentation

Usage Segmentation

User Segments Defined

These five segments form a single mutually-exclusive ladder, evaluated top-down so every user falls into exactly one tier (full definitions and decision tree on the Copilot Usage Segments page):

Power Users: Habitual and averaging 15+ weekly Copilot actions
Habitual Users: Habitual (9+ of 12 weeks in RL12W, all weeks in RL4W) but averaging < 15 weekly actions
Novice Users: Not habitual, averaging 1+ weekly Copilot actions
Low Users: Not habitual, some usage but averaging < 1 weekly action
Non-users: No Copilot usage in the measurement period

Sample Data

Example Datasets

📁 Example Data

📄 copilot-metrics-taxonomy.csv: Copilot metrics reference
📄 viva-insights-org-data-sample.xlsx: Sample organizational data

Analysis Workflow

Export Copilot Usage Data from Viva Insights
Choose Your Analysis Method:
- R/Python scripts for detailed analysis
- DAX columns for Power BI dashboards
Select Time Frame:
- RL12W for long-term habit analysis
- RL4W for pilot programs or short-term analysis
Run Analysis using the appropriate script

Copilot Usage Segments: how Power, Habitual, and Novice segments are defined
DAX Calculated Columns: ready-to-use Power BI formulas for segmentation
Copilot Causal Toolkit: measure the causal impact of Copilot on business outcomes
Causal Inference in Copilot Analytics: methods for isolating Copilot’s true effect
Frontier Prompt Library: generate Copilot reports and dashboards with coding agents
Advanced Analytics: machine learning, regression, and statistical testing
Network Analysis: organizational network analysis (ONA)
Essentials: utilities and visualizations to prepare your data
Getting Started: environment setup and first steps

Need Help?

Copilot Analytics Documentation: Viva Insights Copilot Guide
Power BI Integration: DAX Documentation
Sample Data: Example datasets