A/B test
Help Feedback
GK
Demo · Try StatEVN on its sample files: no login, nothing to upload. Ask for access to use your own data
StatEVN's server is busy or can't be reached. Trying again…
+ New analysis
Recent
Models
OLS Regression
GLM Families
Logistic Regression
Gradient Boosting
Random Forest
Decision Tree
XGBoost
K-Means
Time Series
μ Statistical modelling for everyone
Get a real answer from your numbers.
Analyse a full dataset, or check whether an A/B test result is real. Every result is explained in plain language, with no code.
01Dataset analysis

Analyse a dataset

Find what drives a number, predict an outcome, group similar items or forecast a series.

Drop your file here, or click to browse
Processed on our server to produce your results, and not kept
.csv .xlsx .xls .json
or
No data at hand? Try a sample:
02A/B test

Compare two versions

Did version B really win, or is it chance? No file needed: enter your visitors and conversions.

Significance
Is the difference real?
Lift
How much better, in percent
Sample size
How many more visitors you need
Segments
Up to 10 versions, or many segments
You can also upload a file of results.
Archive
Your three most recent analyses are in the sidebar. Older ones are kept here — click one to open its summary.
What do you want to understand?
Describe your goal in plain language. StatEVN will detect the right model and adapt the next step to match.
How do you want to work?
📊
OLS Linear Regression
Get data from a database
Choose what to fetch from lists, no SQL needed. Table rules you set once (filters, measures, dates) are applied every time. The result is analysed like an uploaded file.
1 · Connect
Use a login that can only read. The password is used for this session only and is never saved.
What kind of prediction do you need?
"Predict" can mean two different things. Choose the one that fits, and StatEVN will open the matching variable selection.
📊
Understand the effect
Prediction models
See which variables affect your outcome and by how much, for example what drives revenue. You choose an outcome and its predictors.
📈
Forecast variable(s)
Forecasting models
Project future values of a series over time. You choose the column to forecast and, optionally, external variables. Needs a date column.
A/B test
Find out whether the difference between two versions is real or just luck. Enter your totals, upload a file, or plan how many visitors a test needs.
I have totals
I have a file
Segments table
Plan the sample size

What did you measure?

Conversions (yes / no) Averages (revenue, time, …)
For each version enter the number of visitors and how many of them converted (bought, signed up, clicked). Choose the control, the version you compare against.
Compare models
Fit several kinds of model on the same data and the same test rows, and see which does best.

Which models?

Every model uses the same training rows and is scored on the same test rows it never saw. Pick at least two. For a yes/no outcome StatEVN uses the classification version of each. When you are grouping items or forecasting over time, you choose methods for that instead.
Select your variables
Click columns on the left to assign them.
Available columns
Outcome variable (drag a column here)
Drag a column here
Predictors (drag columns here)
Drag columns here
ARIMA
Forecast how many periods ahead? leave empty = automatic (one season, or about 12)
StatEVN finds your date column automatically, and asks you when it isn't sure which column holds the dates. Your data should have one row per day, week, month, quarter or year.
Help us understand your column names
Some of your selected columns appear to contain abbreviations. Knowing what they mean — and whether higher values are good or bad — helps StatEVN interpret results correctly. This step is optional.
Other abbreviations in your column names
These appear at the start of column names (e.g. CC_BR). Tick the ones you want to define — optional.
⚠ Results interpretation may be reversed for columns where the direction (higher = better or worse) is unknown. For example, a positive effect from a "Bounce Rate" variable is actually negative if you haven't told StatEVN that higher Bounce Rate is bad. You can always re-run with direction specified.
Some columns have too many different values
A column with thousands of different names cannot be used as categories in a model. StatEVN can sort the values into a few groups first. Pick the grouping that makes sense for your data — the preview shows exactly what each option does.
Analysing your data…
Analysis failed
OLS Linear Regression
Results
Why this model was chosen
How was this analysis? Two minutes of feedback helps shape StatEVN during the beta.Give feedback