Filtering by Segment
The segment filter on the Results page lets you view your experiment results through a narrower lens. Instead of looking at all visitors together, you can filter down to one group: a device type, a country, a browser, or a custom segment. Then you see how the experiment performed for just that group. This is one of the most useful tools for understanding why your results look the way they do.
Where to find the segment filter
The segment filter appears as a dropdown near the top of the Results page. It sits above the table that compares each variation (each version you're testing, including the control). Click the dropdown to see all available segment dimensions.
Tick one or more segment values, then click Apply, and the whole Results page updates. Summary cards, the variation table, metric sections, and the time series chart all show data for just that segment. Apply stays in view at the bottom of the screen however long the list is. If you close the dropdown without applying, your ticks are kept for next time and the dropdown says Not applied until you apply or reset them.
To return to the full results, click Reset in the dropdown.
Available segments
The following segments are automatically available for all experiments:
- Device type: Desktop, Mobile, Tablet. Filter to see how results differ between screen sizes and interaction modes.
- Country: any country that had visitors during the experiment. Filter to see results for a specific market.
- Browser: Chrome, Firefox, Safari, Edge, Opera, Samsung Internet. Filter to see if results vary by browser.
- Platform: the visitor's operating system (macOS, Windows, iOS, Android, Linux). Filter to see OS-level differences.
- Language: the visitor's browser language. Filter to see results for English speakers vs French speakers, etc.
- User type: New or Returning. Filter to see whether the experiment worked differently for first-time vs repeat visitors.
When an experiment uses commerce conditions (cart value, products viewed, or purchase history), the snippet records two extra values on its own. Each visitor gets a cart band and a true/false flag for whether they're a known purchaser. Both are built in, so they appear in the dropdown as soon as the experiment has recorded them. There is nothing to register.
- Cart value (
cart_band): the visitor's cart total the moment they entered the experiment, bucketed in the project currency:none(no cart known),0(empty cart),lt50,50to100,100to250,gte250. The dropdown lists them cheapest first. Filter to see whether a change worked differently for high-value carts. - Purchaser:
trueorfalse, from the visitor's purchase history. Recorded only when the experiment uses a purchase-history condition.
The snippet records these two only for experiments that use a commerce condition, so they stay out of the dropdown everywhere else. An experiment that has never recorded a cart band shows no Cart value filter at all.
A custom segment you send with avsb.track.segment() works differently. Register its key under Metrics → Segments, and it then appears in the filter dropdown under the name you gave it.
How filtering works
When you select a segment filter, A vs B reworks every number on the page. Visitor counts, conversions, conversion rates, and the per-metric statistics are recalculated from only the visitors who belong to that segment. The variation assignments themselves don't change; you're simply looking at a subset of the enrolled visitors.
Because you are now looking at fewer visitors, the credible intervals will widen and the winning probability may drop. This is expected: the smaller the sample, the less certainty. For segment analysis to be meaningful, you typically need a few hundred visitors within the segment per variation.
What the page withholds under a filter
A result found by slicing data after the fact is exploratory, not confirmatory, so while any segment filter is active the Results page withholds every decision-grade claim:
- The verdict chip and its one-line verdict are replaced by a notice explaining that the verdict is only shown on the full audience.
- The summary card that normally shows the winning probability (or statistical significance) shows an "Exploratory slice, no verdict" tag instead, and the Significant badge does not appear.
- The full decision story, including its ship recommendation, and the decision risk projection are hidden.
The descriptive numbers stay: visitor counts, conversion rates, lifts and intervals for the slice are all real and correctly computed. What the page will not do is recommend a decision from them. Clear the segment filter to see the verdict again. Changing only the date range does not trigger this: trimming the window is legitimate; picking a subgroup after seeing the data is not a confirmatory result.
The verdict, sample progress, peek protection, and the Stop action always judge your experiment's whole visitor count, never the filtered one. Date and segment filters change the numbers you see on the page; they never change what A vs B bases a decision on.
How far back you can look
The date range picker reaches back as far as your reporting window allows: 90 days without billing details on file, 365 days with them. Ask for a start date older than that, and the Results page shows the window you have instead. See Billing for the full table.
Finished experiment summaries are not affected. The lift and verdict for every experiment you have run stay on your experiment list permanently, whatever your plan.
Common use cases
Mobile vs desktop comparison
Your overall results show a modest lift, but you suspect the experiment worked better on mobile. Filter to Mobile and then to Desktop to compare. If mobile shows a strong win and desktop shows no improvement, you have a mobile-specific result. You might decide to roll out the change on mobile first.
Differences by country
You are running a global experiment but your largest market is the US. Filter by Country = United States to see how the experiment is performing for your biggest segment. If the US result looks good but results elsewhere are neutral, you can move ahead with US traffic. Give the experiment more time outside the US.
New vs returning visitors
Filter by User Type = New to see how first-time visitors responded to your experiment. Then filter by Returning. If the experiment produced different results for each group, use that to guide how you target each one. Maybe the winning variation is great for new visitors, but the control works better for returning ones.
Custom segment analysis
If you sent custom segment data like subscription plan or user tier, filter by those to understand whether premium users responded differently from free users. This kind of segment analysis often shows things the overall number hides.
Treat segment filtering as a first look, not final proof. When you filter to a segment, you're running many hidden comparisons, one per segment. This raises the odds of finding a "significant" result by chance. Use segment filters to form an idea, then test that idea with its own experiment aimed at that segment.
Bookmarking and sharing
Active filters appear in the URL. The date range writes itself as ?from=YYYY-MM-DD&to=YYYY-MM-DD. Each segment selection writes a repeated ?segment=key:value param. Values under the same segment combine with OR, and different segments combine with AND. Adding device:mobile on top of country:US narrows to mobile visitors from the US, which is what you'd expect.
Select both US and UK under Country and you get visitors from either country, added together. Values under one segment combine with OR, so the group grows as you tick more values. Values under different segments still narrow: Country = US plus Device = Mobile leaves only mobile visitors in the US. To compare two values against each other rather than pool them, filter to one value at a time and note the numbers for each.
You can bookmark a sliced view, share it with a teammate, or paste the link into Slack. Opening the link restores the same filters. Date presets like "Last 7 days" write the absolute dates into the link, not the preset name. So a link shared on a Tuesday for "Last 7 days" still shows that same seven-day window when opened on Friday. The link reproduces the exact view you saw, not a rolling window that shifts with each viewing.
The baseline variation and the statistical engine are not part of the shared link: they're per-viewer rendering preferences. The recipient sees their own baseline / engine choice for the same sliced data.