Most AI impact assessments start counting from the day the tool was turned on. That is the wrong starting point.
If your team started using Copilot in Q3 2024, your analytics tool almost certainly started recording from Q3 2024. Your current velocity gets compared to last month, or last quarter. The trend line starts at activation. But that starting point is arbitrary. You cannot know whether your AI tools are improving anything without knowing what the trajectory looked like before they arrived.
That is what a baseline is for.
What a baseline actually is
A baseline is a stable picture of your team’s output from before AI adoption. Velocity, churn ratio, average commit size, lines per active day, weekend commit percentage — all of it, from a representative period before the tools entered the workflow.
Most teams do not have this number in a usable form. But the data is not gone. Your git history predates AI adoption. Every commit has a timestamp, an author, and a set of file changes. If your team has been committing code for more than eighteen months, you have at least a year of pre-AI output in the repository. You just have not read it as a baseline.
Why it changes what the current numbers mean
Without a baseline, a 22% velocity increase after AI adoption looks unambiguously positive.
With a baseline, it becomes a more specific question. If your pre-AI velocity was flat, a 22% increase is significant. If you were already on a 15–18% growth trajectory, the AI contribution is closer to 4–7%. If the acceleration started before AI adoption and continued through it, the tools may not have caused it at all.
The same applies to churn. If your churn ratio was already elevated before AI tools arrived, the tools did not create that problem. If it increased sharply and the inflection point coincides with adoption, that is a different finding.
Neither of those interpretations is available without a baseline to compare against.
The specific things a baseline makes answerable
Was velocity already trending upward? A team that hired two senior engineers in Q2 and then adopted AI tools in Q3 will show a velocity increase in Q3. The baseline tells you what the Q2 trend looked like.
Was churn already a concern? Some codebases are structurally higher-churn than others — active rewrites, aggressive refactoring, early-stage architecture. A baseline contextualises what the current churn ratio means.
What is a normal commit pattern for your team? Burst patterns, weekend work, average commit sizes — these vary significantly between teams. Without a baseline, you cannot tell whether a change in pattern is meaningful.
The baseline is already recoverable
To be direct about this: you do not need to run an experiment or set up new tracking. You do not need to instrument anything prospectively. The pre-AI baseline is already in your git history, and the git history goes back as far as your repository does.
What you need is a tool that reads it with the right question in mind — drawing a line at your AI adoption date and comparing both sides with a consistent methodology. The analysis is not complicated. It is a reading problem, not a data problem.
Once you have it, every current metric has a reference point. Velocity is up 22% — up from what trend? Churn is 2.1 — was it 1.4 before? Those are answerable questions. Without the baseline, they are not.