Your testing tool shows a green arrow and a percentage. Galley tells you whether that number is a real result or just noise, in one plain sentence, with the statistics done properly underneath.
Free to use. No account. Your CSV is read in the browser, never uploaded.
You run a test for a week. The dashboard says the new version is up 12 percent. You ship it, and revenue does not move.
That happens because a percentage on a dashboard is not a result. It is a snapshot of two numbers that would differ from each other even if the two versions were identical. The question that matters is whether the gap is bigger than chance can explain, and almost no ecommerce dashboard answers it clearly.
Galley answers it. It takes the same counts your platform already gives you and runs the tests a statistician would run, then says what they mean in a sentence you can forward to your client or your boss.
Everything below runs automatically. You do not choose a test or set a parameter.
Ship it, keep running, or call it flat. One sentence at the top of the report, written in words rather than notation, with the confidence level and the corrected p-value stated plainly beside it.
Conversion rate can rise while revenue per visitor falls, usually when the winning version sells more of something cheaper. Galley reports revenue per visitor and average order value alongside the rate, so a hollow win shows itself.
If your traffic split came out uneven, the whole experiment is suspect no matter how good the numbers look. Galley runs a sample ratio mismatch check and warns you before you read anything else.
Export the results from your testing platform, drop the file in, and the form fills itself. Galley recognises the common export shapes and lets you map the columns by hand for anything it does not know.
Galley is not a spreadsheet formula with a nice skin on it. It is 906 lines of Python built on scipy, statsmodels and numpy, the same libraries that sit behind published research and professional data science work. All of it executes the moment you press Analyse, and none of it is your problem.
def perform_srm_test(observed, expected_split=None):
n = len(observed)
if expected_split is None:
expected_split = [1 / n] * n
total = sum(observed)
expected = [total * p for p in expected_split]
stat, p = chisquare(observed, f_exp=expected)
return stat, pThe sample ratio mismatch check, exactly as it runs. It is what tells you the traffic split broke before you trust a single number above it.
Drop in a CSV from your testing platform, paste the results table straight out of the dashboard, or type four numbers per variation: visitors, conversions, revenue and units.
Duration and start date, so the report can tell you whether you have collected enough data and how much longer you would need to reach a call.
Significance, confidence intervals, the revenue check and the traffic split, all in one page you can export to PDF and hand to whoever signs off the change.
Galley reads the export formats below without any setup. Anything else falls through to manual column mapping, which takes about fifteen seconds and works with any spreadsheet.
Plus any spreadsheet at all, through manual column mapping.
Every number on the report, the verdict, the significance tests and the charts, is produced without any AI at all. It is arithmetic, and it costs nothing. The AI layer is optional and sits on top of it.
Connect an API key and the same figures get read back to you: a written report structured for whoever has to sign off the change, and a short explanation under each chart saying what it shows and why it matters.
An executive summary, a trade-off analysis across conversion rate, average order value and revenue per visitor, a risk assessment weighing the Bayesian probabilities against your test duration, and a recommendation. Written for a stakeholder, not for an analyst.
Bootstrap distributions and Beta posteriors are not self-explanatory. Ask for it and each chart gets a couple of sentences telling you what you are looking at, grounded in your actual numbers rather than a generic description of the chart type.
Prefer not to use AI at all? Galley also writes a rule-based report with no key and no external call, covering the same result in plainer terms.
The CSV is parsed in your browser. It is never uploaded, and nothing about the file leaves your computer. Only the summary counts you see in the form, visitors and conversions and totals, are sent to run the analysis. There is no account, so there is nothing stored against your name.
Bring the export from your last experiment and find out what it actually said.
Analyse a test