Control group

B2B growth teams constantly test changes: new email subject lines, website copy variations, pricing structures. Without control groups, you can't tell which changes actually improved your metrics and which were coincidental.

Control groups are especially important for longer-duration tests and tests with high implementation costs.

A control group is the slice of users or customers you deliberately leave alone during an experiment. They carry on with the existing experience while a test group gets the change you're trying. Compare the two and you can tell whether your change actually moved the numbers, or whether something else, the season, a big launch, a slow news week, would have moved them anyway. No control group means you're guessing.

The whole point is causation, not correlation. If you change your email copy and opens go up, you only know the copy did it if a comparable group who got the old copy did worse over the same window.

Three flavours of control

• No intervention: the control keeps the current experience, untouched.
• Placebo: they get a change that shouldn't do anything, so you can separate the effect of the change from the effect of any change.
• Alternative intervention: they get a different version, so you're comparing two real options rather than something-vs-nothing.

Why it matters

B2B growth teams test constantly: subject lines, landing pages, pricing, onboarding flows. Without a control you'll happily credit a redesign for a lift that was really just a strong quarter. Controls matter most on long, expensive tests, where guessing wrong costs you weeks.

How to apply it

Make the control genuinely comparable to the test group, same source, same segment, same time window. If the groups differ to begin with, you're measuring the difference between the groups, not the effect of your change. And run it long enough to ride out natural week-to-week noise; two to four weeks is a sensible floor for most B2B tests.

Examples

Subject-line test. Say you're running a re-engagement campaign in Brevo. Send the personalised subject line to half your list and hold the other half on your usual line as the control. Four weeks later you compare open rates, and because the control ran in the same window, a lift isn't just "people open more in March".

Landing-page test. Say you're testing a new pricing page in Unbounce. Split traffic 50/50 between the new design and the existing page (your control), and let the existing page keep running so seasonality hits both equally. The gap in conversion is the real effect of the redesign.

Measuring it honestly. Whatever tool runs the test, wire both cohorts into Amplitude so test and control are the same chart, same metric, same dates. That's what stops you from comparing this month's test group against last month's vague memory of "normal".

A consulting firm split traffic 50/50 between a new page design and the existing one. The new design converted 8% of visitors compared to 6% for the control group.

A B2B software company tested personalised subject lines with half their list. After four weeks, the personalised group had 18% open rates compared to 15% for the control.

Ensure your control group is truly comparable to your test group. If groups are mismatched, differences in outcomes might reflect group differences rather than the effect of your test.

Run your test for long enough to account for natural variation. Aim for at least 2-4 weeks for most B2B tests.

A sales organisation assigned 20 reps to a new discovery framework and 20 to the existing process. After three months, the new process group had 68-day average deal length versus 71 days for control.