Why is my A/B test not splitting traffic evenly?

Uneven A/B traffic splits usually come from one of a few places:

• The randomization isn’t truly random
  If assignment is based on a weak hash, time, IP, browser fingerprint, or “first come, first served,” the split can drift.

• You’re splitting at the wrong level
  If users can see the experiment on multiple devices, sessions, or pages, and the unit of randomization is inconsistent, traffic won’t look even.

• Filtering after assignment
  If you assign 50/50 but then exclude bots, internal traffic, errors, or users who fail tracking, the observed split can become uneven.

• Sample ratio mismatch from implementation bugs
  Common causes: one variant failing to load, redirect loops, JS errors, one bucket dropping more requests, or server-side routing issues.

• Caching or load balancing effects
  A CDN, edge cache, or sticky session can send more traffic to one variant than intended.

• Consent or ad blockers
  If analytics only records users who accept consent or don’t block scripts, measurement can be biased.

• The experiment is segmented unevenly
  If you’re looking at a subset by geography, device, or campaign, the split may be even overall but not within that segment.

• Small sample size / normal variance
  Early in a test, a 50/50 split can still look off just by chance.

What to check:

1. Compare assigned traffic vs tracked traffic.
2. Verify the split at the assignment point, not just in analytics.
3. Look for variant-specific errors, missing events, or redirects.
4. Confirm the randomization unit is user/session/device as intended.
5. Check for filters, consent loss, and bot exclusion.

If you want, I can help you debug it systematically—tell me whether your test is client-side or server-side, and what platform/tool you’re using.