Campaigns & A/B. Group links. Split traffic. Measure everything.

Two-proportion z-test at the click level. The null hypothesis is that both variants have the same click-through rate. Confidence is 1 - p-value, expressed as a percentage. We do not apply Bonferroni correction for multiple variants (running 3+ variants increases the false positive rate — for formal multi-variant experiments, export the raw click stream and apply the correction in your warehouse). The dashboard confidence number is a directional indicator for teams without a data scientist reviewing every test; for high-stakes decisions, export and apply sequential testing corrections.

Default minimum is 200 clicks per variant before the confidence indicator is shown. You can configure this threshold per link (50 minimum, 5,000 maximum). With typical 50/50 splits, 200 clicks per variant gives you roughly 80% power to detect a 10% absolute difference in click rate at 95% confidence — adequate for directional decisions. For smaller effects (1–5% differences), you need thousands of clicks per variant. The A/B system surfaces the number and doesn't hide it; if you need a larger sample, wait.

Yes. Links can be added to or removed from a campaign at any time. Adding a link doesn't backfill its historical click data into the campaign aggregates — campaign analytics show clicks that occurred while a link was assigned to the campaign. If you're using UTM templates, the template is applied at link-creation time, so links added after campaign launch need to be created inside the campaign (or manually UTM-tagged) to inherit the template values.

Round-robin distributes traffic sequentially: first click → Variant A, second → Variant B, third → Variant A, and so on. Every variant gets exactly equal traffic over any large enough window. Weighted random assigns each click to a variant probabilistically according to the configured weights (e.g., 70% Variant A, 30% Variant B) — the actual distribution converges to the weights over time but will show variance over short windows. Use round-robin when you want exact equality; use weighted when you want to favor a control variant or ramp up a new variant cautiously.

Yes — you can pause an individual variant (set its weight to 0 in weighted mode, or exclude it from round-robin rotation). Paused variants stop receiving traffic but retain their historical click data. This is useful if a variant has a bug or shows a negative effect mid-test: pause it without ending the test or losing the accumulated data. Pausing the variant is logged in the audit trail.

Changing the UTM template does not retroactively update existing links. Links created before the template change keep their existing UTM values (set at creation time). Links created after the change inherit the new template values. This is intentional: UTM values in analytics are a fact about what was tagged, not a configurable display label. If you need to re-tag existing links, export them, update the UTM fields in the CSV, and re-import — the importer updates existing links by slug match.

No limit on links per campaign. Campaigns with tens of thousands of links are supported — the campaign analytics query is backed by ClickHouse and handles arbitrary link counts. The dashboard link table paginates at 100 links per page with search and filter. For campaigns at very large scale (100,000+ links), the API is the recommended interface rather than the dashboard.

Yes. Export the full click event stream for a campaign from the analytics export API. Filter by link_id for each variant, join with conversion events, and run any statistical model you prefer. The export includes the variant_id field on each click event so you can distinguish which variant was served. For BigQuery or Snowflake users, the scheduled export includes variant_id and the raw click + conversion tables.