Measuring GenAI Adoption: Telemetry, Surveys, and ROI Strategies

Rolling out Generative AI is a technology that uses large language models to create text, code, and images based on user prompts tools across your company is only half the battle. The real challenge comes when leadership asks, "Is it actually working?" You can’t manage what you don’t measure, but measuring AI adoption is tricky. Traditional metrics like login counts tell you who accessed the tool, not whether they used it effectively or if it saved them time.

To get a clear picture of your organization’s AI maturity, you need a multi-layered approach. Relying on just one method-like a quarterly survey-leaves blind spots. The most effective strategy combines three distinct data sources: telemetry (automatic usage data), experience sampling (specific task-time tracking), and surveys (self-reported sentiment). Together, these methods provide a complete view of adoption breadth, depth, and business impact.

Telemetry Metrics: The Objective Truth

Telemetry provides the hardest data available. It captures direct usage signals from integrated AI platforms without requiring employees to remember or report their actions. This is crucial because human memory is unreliable, especially regarding daily repetitive tasks. When you integrate AI into workflows, such as GitHub Copilot for developers or Microsoft Copilot for office workers, the platform logs every interaction.

For software engineering teams, key telemetry metrics include pull requests per developer, code review time, and cycle time. Platforms like LinearB track quantitative signals such as daily active users by team and the acceptance rate of AI-suggested code. If developers are accepting AI suggestions 80% of the time, that’s a strong signal of utility. If they’re ignoring them, the tool isn’t fitting their workflow.

In broader corporate environments, workplace intelligence platforms like Worklytics leverage Microsoft 365 integration to capture prompt frequency across Word, Excel, PowerPoint, and Outlook. They automatically measure how often employees use ChatGPT Enterprise, Google Gemini, or other integrated assistants. Benchmark data suggests beginner teams generate 15-30 prompts per employee per month. If your numbers are significantly lower, adoption may be stalled due to lack of training or awareness.

However, telemetry has limits. It tells you that something happened, not why. A high number of prompts might indicate heavy usage, or it could mean users are struggling to get good results and prompting repeatedly. Data cleaning and normalization are also required to ensure these signals are reliable across different departments and roles.

Experience Sampling: Quantifying Time Savings

While telemetry shows activity, experience sampling reveals value. This specialized methodology captures qualitative insights that raw data misses: specifically, how much time was saved on specific tasks. This is the bridge between aggregate usage metrics and concrete return on investment (ROI).

Experience sampling involves targeted inquiries where employees report the exact minutes or hours saved on particular development or administrative tasks thanks to GenAI. For example, a marketing manager might report saving two hours per week drafting social media posts using AI. By extrapolating this across the department, organizations can calculate total estimated ROI in terms of time and dollars.

This approach is particularly valuable for understanding adoption barriers. LinearB’s framework identifies qualitative signals such as developer confidence levels in AI suggestions and whether developers are integrating AI into daily tasks or ignoring it. Common barriers identified through this method include hallucinations (incorrect information generated by AI), trust concerns, and noise in suggestions. Knowing that 40% of your team avoids AI because they distrust its accuracy is actionable insight that telemetry alone cannot provide.

Surveys: Measuring Sentiment and Breadth

Surveys excel at capturing adoption breadth, satisfaction depth, and self-reported productivity improvements. They are the best tool for understanding how employees feel about the technology and whether they perceive it as helpful. However, survey design matters immensely.

The Harvard Project on Workforce launched the Generative AI Adoption Tracker, conducting quarterly surveys via the Real-Time Population Survey (RPS). In August 2024, their initial result showed genAI usage at 39.4% of the population. But by revising the question sequencing related to generative AI awareness and use, their updated result for the same period increased to 44.6%. This demonstrates how subtle changes in survey design significantly impact reported adoption rates.

As of November 2025, the Federal Reserve’s monitoring report citing the RPS indicates work-related generative AI adoption stands at approximately 41 percent among U.S. adults. Meanwhile, Pew Research Center data shows that only 23 percent of U.S. adults have used ChatGPT specifically. This discrepancy highlights the difference between "any-GenAI-use" (which includes workplace tools) and specific-platform-use among the general public. Your internal surveys should clarify whether you are measuring overall AI literacy or specific tool proficiency.

Combining Methods for a Holistic View

No single metric tells the whole story. Organizations implementing comprehensive measurement programs benefit from combining all three approaches. According to GetDX’s synthesis, telemetry is primarily useful for quantifying the impact of GenAI on developer output; experience sampling is most useful for quantifying the return on investment; and surveys are best for measuring adoption and satisfaction.

The Halston Media framework identifies four key metrics for tracking AI adoption:

• Percentage of active employee usage (capturing breadth)
• Number of AI workflows deployed (measuring enterprise integration depth)
• Number of AI experiments launched (indicating organizational learning)
• Rates of completion for AI projects (assessing operational execution)

Challenges in attribution remain significant. Code review time reductions could result from GenAI assistance, improved developer experience, or team process changes. Multivariate analysis is required to isolate GenAI’s contribution. Similarly, survey-based measurement suffers from self-reporting bias, where respondents may overstate productivity gains due to social desirability or understate them due to job security concerns.

Implementing Your Measurement Strategy

To build an effective measurement program, start by establishing baseline metrics before widespread deployment. Create automated telemetry collection pipelines to minimize manual effort. Conduct periodic surveys using consistent methodology to enable longitudinal tracking. Implement experience sampling for high-value use cases to quantify ROI accurately.

Integrate findings from all three methods into comprehensive adoption dashboards accessible to leadership. Recent developments show increasing organizational recognition that adoption metrics require as much attention as implementation planning and training programs. The future of adoption measurement involves increased automation, real-time dashboards, and integration with business outcome data to demonstrate concrete ROI connections.

By leveraging telemetry for objective usage data, experience sampling for precise ROI calculation, and surveys for sentiment analysis, you can move beyond guesswork. You’ll gain the clarity needed to justify continued investment, identify training gaps, and optimize your generative AI strategy for maximum business impact.