You check your test results and notice something like this:
It looks uneven. Is Mida broken? Is the data wrong?
No — this is normal. Here's why.
Mida splits traffic randomly
When a visitor lands on your test page, Mida randomly assigns them to either Control or a Variant. For a standard A/B test, each group has an equal 50/50 chance.
But "50/50" is a long-run average — not a guarantee for every batch of visitors. Like flipping a coin, you can flip heads 7 times in a row and still have a fair coin. The more flips you make, the closer you get to 50/50 — but it takes time.
Low-traffic sites take longer to balance out
If your site gets a few hundred visitors per day (or fewer), an uneven split can persist for several days before it visibly corrects. That's not a data quality issue — it's just how small numbers work.
Here's a rough guide:
Total visitors so far | A 40/60 split... |
Under 300 | Very common. Almost certainly random noise. |
300–700 | Somewhat expected. Keep running the test. |
700–1,500 | Less likely by chance. Worth a closer look. |
1,500+ | Unusual for a standard 50/50 test. Contact support. |
The key thing to watch is not any single snapshot — it's whether the gap keeps widening or slowly closing over time. A closing gap is normal. A widening one is worth investigating.
When the uneven split is intentional
In some cases, an uneven split is by design, not a bug:
Multi-Armed Bandit (MAB) tests — Mida automatically shifts more traffic toward the better-performing variant. A 40/60 or even 30/70 split is expected and correct.
Custom traffic weights — If you manually configured a custom split (e.g., 30% Control / 70% Variant), the counts will reflect that.
You can verify which mode your test is running under in the test settings.
What you should NOT do
Don't pause or restart the test because the split looks uneven — resetting a running test invalidates your data and makes results unreliable.
Don't draw conclusions too early — on a low-traffic site, you may need to wait weeks to collect enough visitors for a statistically meaningful result. Mida will show a confidence indicator when you're getting close.
The short version
An uneven visitor split is almost always random noise, especially when total visitor counts are still low. Keep the test running. As more visitors accumulate, the split will trend toward equal. If after 1,500+ total visitors your split is still heavily skewed on a standard 50/50 test, reach out to our support team and we'll investigate.