Control Group

What Is a Control Group?

The control group is the unchanged baseline experience in an experiment — the current page, the current feature, the current flow, left untouched so it can serve as the reference point. Every measured lift is a lift relative to the control. Without a well-preserved control, you don't have an experiment; you have a before-and-after anecdote contaminated by everything else that changed in the meantime.

Also Known As

• Marketing teams call it the control, the baseline, or variant A.
• Growth teams say control or baseline.
• Product teams use control, current experience, or default.
• Engineering teams refer to it as control, default branch, or the "off" state of the flag.
• Statisticians call it the control condition or baseline group.

How It Works

You launch a 50/50 test on your checkout page. 50,000 users hit the control, 50,000 hit the variant. The control experience is exactly what users saw before the test started — same copy, same layout, same code path. During the test, your team pushes a bug fix that changes the control checkout's button color. Now the control has changed mid-test. The baseline is no longer stable, and the measured "lift" of the variant is partially just the bug fix. The experiment is compromised. You restart or discard.

Best Practices

• Freeze the control experience for the entire test duration — no bug fixes, no copy tweaks.
• Run periodic A/A tests (control vs. identical control) to verify your platform produces no spurious lift.
• Document exactly what "control" is so future analysts can reproduce the test conditions.
• Keep a snapshot of the control code and content in case you need to audit later.
• If the control must change mid-test, restart the test rather than letting it continue.

Common Mistakes

• Letting a marketing team update control copy mid-test because "it was a small change."
• Forgetting that a platform-wide deploy affected the control's loading performance.
• Skipping A/A validation and trusting the platform blindly.

Industry Context

• SaaS/B2B: Risk of control drift is high because feature teams constantly iterate; freeze protocols matter.
• Ecommerce/DTC: Seasonal banners and promotions can accidentally modify the control; watch merchandising calendars.
• Lead gen: Relatively easy to freeze — landing pages are usually stable for test duration.

The Behavioral Science Connection

The control group is the placebo effect's cousin: it absorbs the Hawthorne effect, novelty effects, and attention biases that would otherwise inflate variant results. Without a control, you can't separate "people respond to something new" from "people respond to this specific change." The control is how you measure signal above the noise of observation itself.

Key Takeaway

Freeze the control, validate your platform with A/A tests, and treat any mid-test change to the control as an experiment-ending event.