Exporting Results

A vs B lets you export your experiment results as a CSV for further analysis, record-keeping, or sharing with stakeholders.

CSV export

The Export CSV button sits in the action button group at the top of the Results page. It only appears once the experiment has at least one metric with results. It also needs the exportData permission on your role, so ask an org admin if you cannot see it. The file downloads as <experiment-name>-results.csv, with any non-alphanumeric characters in the name replaced by hyphens.

The export reflects the metrics and date range currently loaded on the page. The file is named after the experiment and the moment you exported it, in UTC: Homepage-hero-results-20261001-1818.csv.

The file opens with a short block of key,value lines saying what it covers: the experiment, when it was exported and computed, the date range, any segment filter, the engine of record and the engine shown, and the decision thresholds. Each line appears only when that fact is known. A blank line separates the block from the table.

Metric results

The main table has one row per variation per metric, with these sixteen columns:

Column	What it holds
Metric	The metric's name.
Direction	Which way is a win for this metric: `INCREASE` or `DECREASE`.
Primary	`Yes` for the primary metric, `No` for every other metric, guardrails included.
Variation	The variation's name.
Visitors	Unique visitors enrolled in this variation.
Conversions	Conversions on this metric.
Conv. Rate	Conversions ÷ visitors, to two decimals (e.g. `3.42%`).
Goal-Aligned Lift	Change vs control, to one decimal, signed so that a positive number is always good news: on a `DECREASE` metric, a drop shows as positive (e.g. `+12.4%`). `—` on the control row.
Probability Better (goal-aligned)	The Bayesian chance that this variation beats control in the metric's good direction (e.g. `97.3%`). `—` on the control row, and on every row of a Frequentist or Sequential experiment.
p_value	The p-value from the Frequentist or Sequential engine, as a plain number. `—` on the control row, and on every row of a Bayesian experiment.
ci_low, ci_high	The two ends of the lift interval the engine computed, goal-aligned like the lift. `—` when there is none.
Revenue	Total revenue attributed to this variation, in your project currency.
AOV	Average order value for this variation, in your project currency. `—` when unavailable.
honest_lift_pct	The winner's-curse-corrected lift (e.g. `+6.2%`; see Statistical Methodology). Filled only on the primary metric's leading non-control arm; `—` on every other row.
wrong_sign_risk	The chance that arm's true effect actually points the other way (e.g. `4.7%`). Filled on the same single row as `honest_lift_pct`; `—` elsewhere.

Revenue and AOV are stored internally in minor units (cents) and formatted into your project's currency on export, so the cells read as money rather than raw integers.

A result that only got noticed because it looked big is usually a little smaller in reality. honest_lift_pct and wrong_sign_risk correct for that. A vs B fills them on exactly one row per export: the primary metric's leading non-control variation. Every other row reads —. When a prior is active on the experiment (see Effect Priors), both columns read — everywhere instead. The prior already corrects for the same thing in the headline lift, so using both would correct it twice.

Segment Lift

When the experiment has segment-lift data, a Segment Lift section follows the metric rows with one row per variation per segment value:

Segment Key, Segment Value, Variation, Direction, Visitors, Conversions, Conv. Rate, Goal-Aligned Lift, CI Low, CI High, Adj. p-value, Probability Better (goal-aligned), Significant

A line above the header marks the section as exploratory: it points you to leads worth investigating, not decisions.

Adj. p-value is the p-value after multiple-comparison correction: segment analysis tests many subgroups at once, so the raw p-values would overstate significance. Significant is true / false.

Top Movers

Two further sections, Top Movers: Positive and Top Movers: Negative, list the segments with the largest lift in each direction:

Segment Key, Segment Value, Variation, Goal-Aligned Lift, Visitors

Segment and top-mover rows are for generating hypotheses, not for declaring winners inside a subgroup: the experiment was not powered to detect effects at segment level. See Segment Lift.

If the experiment was stopped early while running under the Frequentist engine, the CSV ends with an Early stop footer recording that the run "Stopped under Frequentist with reduced validity", plus the reason and timestamp if they were captured.

This travels with the file on purpose. A Frequentist p-value assumes one look at a planned sample size, so a result stopped early is weaker than its p-value implies, and a CSV sent to a stakeholder should not be able to hide that. See Early stopping.

Saving a PDF snapshot

There is no PDF export. To capture a visual snapshot, use your browser's print-to-PDF:

Expand every metric section you want captured. Secondary metrics are collapsible, and a collapsed section will not appear in the output.
Set your date range so the time-series chart shows the window you want.
Open the print dialog (Ctrl+P on Windows/Linux, ⌘+P on Mac) and choose Save as PDF as the destination.
Adjust layout. Landscape usually suits the wide results tables, and reducing the scale fits more per page.

Printing renders the page as it appears on screen: nothing is re-laid-out for paper, so expect to fiddle with scale and orientation. For anything you need to look deliberate, screenshot the specific panel or use the CSV and build the summary yourself.

Use the CSV export when the recipient needs to run their own analysis in a spreadsheet or BI tool.
Use a print-to-PDF snapshot when you want a visual summary in a document or deck.
For stakeholders who have A vs B access, share the Results page URL: it shows live, up-to-date results.

The CSV reflects the data at the time of download. If the experiment is still running, the file is a snapshot: re-download it if you need updated numbers later.