Generally, experimentalists make decisions by comparing means and using standard deviation to assess spread. There's exceptions, like using percentile metrics, but the vast majority of comparisons are done in this way.
It's effective, but it's also well known that means mask a lot of information. To help experimentalists on Statsig understand what's going on behind the scenes, we're adding an easy interface to dig into the distributions behind results.
Here, we can see a pulse result showing a statistically significant lift in revenue for both of our experimental variants.
By opening the histogram view (found in the statistics details), we can easily see that this lift is mostly driven by more users moving from the lowest-spend bucket into higher buckets
This is available today on Warehouse Native - and we're scoping out Statsig Cloud.
Many customers on Statsig run hundreds of concurrent experiments. On Warehouse Native, this means that interactive queries from the Statsig console can run slowly during peak hours for daily compute
Now, users on Snowflake, Databricks, and Bigquery can specify separate compute resources for 'interactive' console queries vs. scheduled 'job' queries - meaning the interactive queries will always be snappy. This also means a large compute resource used for large-scale experiment analysis won't get spun up when running small interactive queries like loading data samples.
For those warehouses, we've also added the ability to specify different service accounts for different "Statsig Roles" within the Statsig roles system. This means that the scorecard service account has the necessary access to user data for calculating experiment results, but customers can specify privacy rules like masking for sensitive fields to prevent access to sensitive data through interactive queries in the Statsig console.
Three exciting new improvements to our recently launched Topline Alerts product-
Embed variables in your alert message- You can now insert the event name, value of the alert, warn threshold, alert threshold, and (soon) the value you've grouped your events by directly into your notification text body to provide more context when viewing alert notifications.
Test your notification manually- You can now trigger each state of your alert (Raise, Warn, Resolve, No Data) to ensure your alert is configured as desired at setup time.
View Evaluated vs. Source data- In the "Diagnostics" tab of your alert, you can now toggle between Evaluated (aggregated data for the end alert evaluation) and Source modes (underlying event data pre-aggregation, used as input to your alert calculation). While Evaluated data mode is still restricted to a 24 hr event window, you can look back further for Source data to get a sense of how the event you're alerting on has trended over a longer window.
Surrogate metrics are now available as a type of "latest value" metric in Warehouse Native.
Surrogate metrics (also called proxy or predictive metrics) enable measurement of a long-term outcome that can be impractical to measure during an experiment. However, if used incorrectly adjustment, the false-positive rate will be inflated.
While Statsig won't create surrogate metrics for you, when you've created one you can input the mean squared error (MSE) of your model, so that we can accurately calculate p-values and confidence intervals that account for the inherent error in the predictive model.
Surrogate metrics will have inflated confidence intervals and p-values compared to the same metric without any MSE specified.
Learn more here!
You can now compare up to 15 groups in funnel charts when using a group by, up from the previous limit of 5.
Select and compare up to 15 groups in a funnel analysis
Use a new selector to control exactly how many groups to display
Once you apply a group by (e.g., browser, country, experiment variant), a group count selector appears. Use it to choose how many top groups to include based on event volume.
This gives you more flexibility to analyze performance across more segments—especially helpful for large experiments, multi-region launches, or platform-specific funnels.
Let us know how this works for your use case—we’re always looking to improve.
Drilldown now includes two new visualization options to help you better understand the distribution of your metrics: Donut charts and a World Map view.
Use Donut charts to visualize proportional breakdowns of any event or metric
Use the World Map to see event counts or metric values by country
Apply these views to any Drilldown chart with a group-by
In Drilldown, after selecting a metric and grouping it by a property (like country, browser, or device type), choose either the Donut or World Map view from the chart type selector. The map overlays values by country, while the donut shows relative proportions.
These views make it easier to spot geographic trends, visualize dominant segments, or present clean summaries of how usage breaks down across key dimensions—especially useful for sharing or monitoring at a glance.
A new dedicated chart settings panel gives you more control over how charts are displayed—making it easier to fine tine your analysis data and how that data is visualized.
From the gear icon in the top-right of any chart, you can now:
Start Y-Axis at 0 for more consistent visual baselines
Filter Out Bots to clean up automated or test traffic
Include Non-Prod Data when needed for QA or staging checks
Show Table/Legend Only to highlight key values without showing the full plot
Split Charts by Metric (in Metric Drilldown only) to display each metric on its own chart—ideal for comparing metrics with different units or scales
Click the gear icon to open the chart options panel. These settings are chart-specific and persist as part of the chart configuration. When using Drilldown, splitting by metric creates a stacked view—turning one chart into a mini dashboard.
These controls help tailor each chart to its purpose—whether you’re cleaning up noisy data, presenting key takeaways, or exploring metrics with vastly different scales.
You can now filter entire dashboards using behavioral cohorts—alongside existing property-based filters.
Apply a behavioral cohort as a filter to any dashboard
Combine cohort filters with property filters for scoped, layered analysis
Easily compare how different cohorts interact with your product across multiple charts
From the dashboard filter panel, select a saved cohort to apply it globally. All charts on the dashboard will update to reflect data only for users in that cohort. You can still apply property filters in parallel.
This makes it simple to compare behaviors across user groups like “First-time Users,” “Power Users,” or “Users Who Churned Last Month.”
Cohort filters unlock more targeted analysis across dashboards, allowing you to focus on the patterns and behaviors of specific user groups without editing each chart individually.
You can now measure how frequently users (or other unit IDs) perform a specific event with the new Count per User aggregation option.
Analyze the average, median, or percentile distribution of how often an event is performed per user (or per company, account, etc.)
Select from: average (default), median, min, max, 75th, 90th, 95th, or 99th percentile
Choose the unit ID to aggregate on - user ID, company ID, or any custom unique identifier
When you select Count per User in Metric Drilldown charts, Statsig calculates how many times each unit ID performed the chosen event during the time window. You can then apply summary statistics like median or 95th percentile to understand the distribution across those users.
This aggregation only includes unit IDs that performed the event at least once in the time range—it doesn’t factor in users who did not perform the event.
This gives you a more nuanced view of engagement patterns, helping you answer questions like:
What’s the median number of times a user triggers a key action?
How often do your most active users complete a workflow?
How concentrated or spread out is usage of a particular feature?
Ideal for understanding usage depth, not just reach.
You can now use User Journeys in the Warehouse Native version of Statsig to visualize the most common paths users take through your product.
Build user journeys directly on your WHN setup
Choose a source table, specify your event name column, and select a starting event
Analyze the most frequent sequences of events after that starting point
To get started, select the table where your events are stored, specify which column contains the event names, and choose the event that marks the beginning of the journey. Statsig will generate a path view showing the most common user flows from that point forward.
At this stage, User Journeys on WHN are designed for schemas where all events live in a single source table. We’re actively working to support setups with:
One source table per event
Multiple source tables each containing many events
This feature gives you visibility into how users move through your product, where they drop off, and which paths are most common—directly within your warehouse environment.