What an Enterprise Copilot Trial Showed
GitHub’s earlier lab studies established that Copilot can speed up coding and boost developers’ confidence in their output. The open question was whether those results would survive contact with a large, real-world engineering organization. To find out, GitHub teamed up with Accenture on a randomized controlled trial plus a company-wide adoption analysis and user survey. The headline results: developers were quicker to pick up the tool than expected, pull request volume and merge rates rose, and self-reported satisfaction improved.
Study Design and Data Collection
The study had three parts. First, a randomized controlled trial split Accenture developers into two groups: one with Copilot access, one without. The participants spanned engineering, design, and testing roles, from entry-level to team management, working on a variety of software products. Telemetry from DevOps pipelines captured output metrics related to regular coding activity.
Second, a broader adoption analysis looked at installation rates, code acceptance rates, and how long it took developers to accept Copilot’s first suggestion. Success in this phase was defined as accepting at least one suggestion. Third, a survey of Copilot users gathered perceptions about how the tool affected their workflows. Combining telemetry with survey responses gave a fuller picture of the tool’s impact than either source alone.
Adoption Was Fast and Sticky
The usage data shows that developers did not need a long ramp-up. 81.4% installed the IDE extension on the same day they received a license, and 96% of those who installed it started receiving and accepting suggestions that same day. On average, it took about one minute from seeing the first suggestion to accepting one.
That early enthusiasm carried over into sustained use. 67% of respondents said they used Copilot at least five days per week, with an average of 3.4 days per week. A substantial 70% reported relying on it for tasks in a familiar programming language. In the survey, 51% rated Copilot “extremely useful” and 43% found it “extremely easy to use.”
Quality Held Up as Throughput Rose
Pull request counts are a standard proxy for delivered value, and Accenture developers saw an 8.69% increase in pull requests. But more output only matters if it clears review. The pull request merge rate rose 15%, and successful CI builds increased 84%. Both human reviewers and automated test runs were more likely to pass the code produced with Copilot assistance.
On acceptance behavior, developers accepted around 30% of Copilot’s suggestions. 90% of developers committed code that Copilot suggested, and 91% reported that their teams had merged pull requests containing such code. Retention was high: developers kept 88% of Copilot-generated characters in their editor.
Developers Reported a Better Workday
The survey portion revealed a strong jump in job satisfaction. 90% of developers felt more fulfilled with their jobs when using Copilot, and 95% said they enjoyed coding more with it. The effect scaled with usage: fulfillment increased only “a little” for those using Copilot less than two days per week, but “quite a bit” for those using it more frequently.
Beyond satisfaction, 70% of developers said they expended quite a bit less mental effort on repetitive tasks, and 54% spent less time searching for information or examples. A majority also reported being able to maintain flow state while using the tool. The reduction in cognitive load on routine work could let developers redirect attention toward design, collaboration, and other higher-value activities.
From Lab Conditions to Production Code
The Accenture results are an early sign that Copilot’s lab-measured benefits persist in enterprise settings. The combination of faster adoption, higher throughput that still passes review, and improved self-reported developer experience suggests the tool can be integrated into daily workflows without a quality trade-off. The findings also point to a secondary effect: as developers engage more with Copilot, they may build proficiency in their organization’s codebase more quickly, raising contribution levels across teams.
