Reading the Results Dashboard

The Results tab is where you spend most of your time once a test is live. This guide walks through the experiment page and every panel on its Results tab in the order you see them, with a quick read on what each number means.

The experiment page

Open any experiment from the Experiments list. The header shows the test’s name and a status badge: Draft, Scheduled, Running, Paused, Completed or Archived.

Next to the badge, a running test has a Pause button and a paused test has a Resume button. Pausing stops sessions from being recorded immediately, and you can resume from the same point later. The ⋯ menu holds the rest of the actions:

• Export CSV (see Export CSV below)
• Reschedule and Cancel schedule on a scheduled test
• Duplicate, which creates a copy you can edit and launch
• Archive (on anything except an archived test)
• End test on a running or paused test
• Delete on anything that isn’t running

Below the header are four tabs: Results, Setup, Activity and Alerts. Setup holds the General, Targeting and Variants steps; Activity is the test’s history (see Experiment Activity); Alerts lists the test’s anomaly alerts and results warnings, with a badge counting the open ones.

Once a test is Completed or Archived, a note above the results reads Final snapshot. This test has ended, so its results are frozen. The numbers no longer change and the refresh button disappears.

Overlapping experiments

When another test has been serving the same visitors, a banner above the filter bar reads N overlapping tests · X% shared visitors, with an (i) that explains which test overlapped and since when. It never appears on a draft or scheduled test, and it only counts overlaps inside the date range you have picked.

The banner carries an Overlap view switch, which is off every time you open the page. Turn it on and the whole Results body is rebuilt around experiences: control, each variant of this test on its own, and each combination with the other test’s variants. An Overlap analysis card then appears before Performance by variation. It plots each variant’s lift by what the other test showed and gives one of three verdicts: Interaction likely, Interaction inconclusive or Not enough shared data. When several tests overlap, a Compare with menu picks which pair you are looking at; it starts on the pair with the most shared visitors.

Overlap view honours every filter in the bar except Exposure: with Mixed only selected the switch reads Not available with mixed exposure. If the overlap is too large to analyse in the browser, the banner asks you to narrow the date range.

The filter bar

A sticky bar under the tabs narrows everything below it: Devices · Visitors · Exposure · Advanced · Dates, with the Updated just now button on the right. Exposure only appears when the test’s stickiness lets one visitor see more than one variant, or once such visitors exist. A control that is narrowing the page gets a highlighted edge, and on a phone the controls fold behind a Filters button that shows how many are active.

The bar’s notes appear directly under it: the audience and data window for a visitor segment, a note on which day attribute data starts for Advanced filters, a low-sample warning on a thin segment (“Low sample in this segment.”), and a reminder to refresh when the numbers are more than a day old (“Results last computed 2d ago. Refresh for current data.”). Like the anomaly alerts, each warning shows here the first time you see it and is listed in the Alerts tab after that.

Every filter is explained in Filtering Results.

Alerts

The Alerts tab lists every open anomaly alert and results warning for the test, with the dismissed ones in a collapsed Dismissed (n) group below. A new alert also appears as a banner under the filter bar the first time you see it; after a reload it is in the Alerts tab only. Each alert has a one-line message, a What to do link that opens the recommendation, and a dismiss button. See Anomaly Alerts for what each check looks for.

Headline numbers

Four stat cards open the results:

• Visitors: unique visitors who entered the test in the current view, across all variations.
• The primary goal, labelled by its name (for example Purchases): the count and the share of visitors it represents, or the mean for a value metric.
• Variant uplift: the relative change in the primary metric of the best variation versus control.
• Chance to beat control: the probability, from the Bayesian model, that the best variation truly beats control. The sub-line reads from Not enough data and Low confidence up to Very high confidence at 95% or more.

The verdict box below them summarises the test in one line. Its title is one of:

Verdict	What it means
Not enough data yet	At least one variation is under the sample floor, so there is no comparison to make.
Winner blocked by guardrail	A variation cleared 95% but breached one of your guardrail goals.
<name> is winning	That variation has a 95% or higher probability to be best.
Control is winning	Every variation has a 5% or lower probability to be best.
<name> is trending ahead	The best variation is at 80% or higher, but the result is still uncertain.
Control is trending ahead	Every variation is at 20% or lower.
No clear leader yet	Everything in between.

The sample floor behind Not enough data yet is ten conversions on the primary goal per variation, control included. Until every variation has ten, the engine withholds a winner however the probabilities look, and the footer under Performance by variation reads Estimated uplift is not yet available.

A badge in the corner reads Collecting data while the test runs, Paused when it is paused and Final results once it has ended. When the title reads <name> is winning and the test has not ended, a Declare winner button appears in the box. It opens the same End this test? confirmation as End test in the ⋯ menu; see Declaring a Winner.

Right under the verdict, a completed test with an honest winner shows the rollout card. See Winner Rollout.

Conversion rate comparison

Beside the Conversion rate over time chart sits the comparison card. A Bars | Distributions toggle switches between a bar per variation on a shared scale and the posterior curves from the Bayesian model: wider curves mean fewer visitors. Both views draw Control and every variation, even at 0%: a bar keeps a small mark at the start of its track, and a curve the model is all but sure of is drawn as a narrow block. Distributions lists Control first, its rate also marked as the dashed baseline; a variation with no visitors yet shows a flat curve across the whole axis. For a revenue or other continuous primary metric (Revenue per session, Average order value, a calculated metric) each curve is the Normal posterior on that variation’s mean, drawn from its standard error, and the axis reads in that figure. Your choice is remembered on this browser.

Under the bars, one line states the uplift in words (Variant B has a 12.4% higher conversion rate than control) and a ring shows Chance to beat control with its confidence title. When the primary goal is a value metric such as Average Order Value, the card is titled after that metric and shows means instead of rates.

Charts over time

Conversion rate over time draws each variation’s rate day by day. A Daily | Cumulative menu picks each day’s own values (the default) or running totals, and a Line | Bar toggle picks the shape. Early days swing a lot; trust the trend once the lines settle. The axes always cover the selected date range; with nothing to plot yet the chart reads No conversions yet over them. Chart unavailable with a Try again button appears only when the data could not be fetched.

On Shopify, two more charts follow the funnel cards further down: Average order value and Revenue per visitor, both per variation and both drawn from the same date range and filters as the rest of the tab.

AI analysis

Under the chart and comparison row, workspaces with AI enabled see the AI Analysis card with its Generate analysis button. It reads the unfiltered results, so it carries an All visitors tag whenever the Exposure filter is on Single-variant only. See AI Analysis.

Performance by variation

One row per variation, control first with an Original chip, with five columns:

Column	Meaning
Variation	The name, its colour dot, and a tooltip saying what it changes.
Visitors	Unique visitors who entered this variation.
Conversion rate (or the metric’s name)	The rate with a bar, and X of Y visitors underneath.
Uplift	The relative change versus control.
Chance to beat control	The probability this variation truly beats control.

The footer reads Estimated uplift: +a% to +b% · 95% credible interval once every variation clears the sample floor. A wide interval means the data has not converged yet; a narrow one shifted away from zero is a confident result.

Other goals

Every other goal the test tracks gets a row: your conversion goals, custom events that carried a value, calculated metrics, and Subscriptions when any were recorded. Each variation has a cell with the rate (or mean) and the visitor count, and each non-control variation has a Chance to beat control column. Use it to spot side effects: a variant that lifts add-to-carts but lowers purchases is not a win.

Funnels and revenue

Each funnel you have set up in Settings gets a card, Visitors first and then every step, with a column per variation and the drop-off at each step. A Left | Center toggle changes the bar alignment, and tests with many variations page through them.

Revenue breakdown (always on Shopify, or Value breakdown for an HTML goal with a value) opens with a totals strip (Total revenue, Orders, Visitors), then the rings side by side above the table: a share ring and, on Shopify, a Revenue per session ring holding a slice per variation sized by its revenue per session, with each figure named in the key beneath it (before any revenue arrives that ring is a neutral grey and every figure reads $0.00). The table under them has one row per variation: Visitors, Orders, Revenue, Avg. order value and Revenue per visitor, each with its change versus control. Revenue per visitor is the fairest comparison when visitor counts differ.

A facts row closes this part of the page: Traffic split, Started, Duration and Audience.

Shopify-only panels

Shopify workspaces get two more groups at the bottom of the tab.

Advanced DOM Insights appears once the Web Pixel has recorded element data. It has three sections: Variant Visibility (the share of sessions that scrolled to the changed element, with the average scroll depth), Click Analysis (the most-clicked elements per variation, with the difference from control) and Render Health (the share of sessions in which the variant rendered, flagged under 95%). An Exposed sessions only switch narrows the rest of the results to sessions that scrolled to the change.

Then come three cards built from ShopifyQL, each marked store-wide because they describe the whole store during the test period rather than the variations: Order Insights, Geographic Breakdown and Product Performance.

Neither group honours the Exposure filter: Advanced DOM Insights and Order Insights wear an All visitors tag while Single-variant only is selected.

Export CSV

Export CSV in the ⋯ menu downloads a per-day table for the filters currently in force. The first row is the header, Date,Variation,Visitors,…, followed by the measured columns: on Shopify Added to cart, Began checkout, Purchases and Revenue with its currency; on an HTML site the primary goal’s name. There is one row per day per variation, and the Visitors column sums to the Visitors stat card over the whole range. With Overlap view on, the second column is Experience instead.

Refreshing

Results are computed on the server and cached, and the button on the right of the filter bar says how old they are: Updated just now, Updated 12 min ago, and so on. Clicking it recomputes the view you are looking at from the latest data and, on an unfiltered view, refreshes the other device views in the background. While the request runs the icon turns and the button is disabled (with reduced motion on, it reads Refreshing… instead), the cards dim under a moving sheen with a Refreshing results… note at the top of the results, and the numbers you were looking at stay on screen; every card, chart and table re-renders when the fresh numbers land. If the revenue figures were not ready in time, the comparison card, Performance by variation and the Revenue breakdown read Revenue is still computing — try Refresh again shortly. while everything else shows the fresh data; the last saved revenue numbers are kept until a refresh brings new ones. If the recompute fails, the numbers on screen stay and a red banner at the top of the results says Couldn’t refresh results. Try again. Views with a visitor segment, Advanced filters or a custom date range are computed live every time and are never cached. If the cached numbers are more than a day old, a banner under the bar reminds you to refresh.

The button is hidden on Completed and Archived tests, whose results are frozen.

What to Look For

A quick mental checklist when you open the Results tab:

1. Chance to beat control: is the leading variation past 95%? If yes, candidate winner.
2. Sample sizes: does each variation have at least 300–500 visitors and ten conversions? Below that, even a high probability is fragile.
3. Time elapsed: has it run a full week minimum? Day-of-week effects can swing day-to-day.
4. Devices: switch the Devices filter between Desktop, Mobile and Tablet. Does the result hold, or is it a one-device effect?
5. Funnel breakdown (Shopify): is the lift coming from the right stage of the funnel, or are you just shifting the bottleneck?

If all five check out, you have a real result. See Declaring a Winner for what to do next.

Common Mistakes

• Stopping at 80% or 90% probability. This is the single most common A/B testing mistake. Wait for 95%. See Common Mistakes.
• Reading the conversion rate without looking at the credible interval. A 5.0% vs 4.5% comparison is meaningless if the interval under Performance by variation runs from −10% to +30%. The interval is what tells you whether the difference is real.
• Never switching the Devices filter. A flat aggregate result with a strong mobile win is still actionable: apply the variant to mobile only.
• Refreshing repeatedly during a low-traffic experiment. The math doesn’t move faster because you watched it. Set a reminder to check tomorrow.

Next Steps

• Get the full picture of the Bayesian model behind the numbers: Bayesian Stats Explained.
• Narrow the results by device, visitor type, exposure, attributes or dates: Filtering Results.
• Get an AI-generated read on the result: AI Analysis.