Statsig Release Notes

Today, we’re launching agent-skills, our new public repository for reusable Statsig skills. It’s designed to help teams run common Statsig workflows faster and more consistently from AI agents.

What you can do now

• Create Dashboard: Generate Statsig dashboards with a repeatable, structured workflow instead of manual one-off setup.
• Create Cloud Metric: Define cloud metrics through a guided skill flow, including key configuration steps that are easy to miss in ad hoc API calls.

Why this matters

Skills turn complex Statsig workflows into repeatable, shareable agent instructions you can personalize or share and reuse across your team's projects. With Skills, you can direct your agents to execute multi-step logic, stitching together Console API calls, MCP tool calls, and prompt instructions.

Getting started

1. Ensure you have a Console API Key -- this is required for the skill to carry out Statsig Console API actions.
2. Install the Statsig agent-skills repo with the Vercel skills CLI:

npx skills add statsig-io/agent-skills

1. Instruct your agent to use the skill (e.g., "Codex, help me create a cloud ratio metric for checkout rate).
2. Watch your agent follow your direction and the skill instructions to work with Statsig!

Explore the repo and start building repeatable Statsig workflows: statsig-io/agent-skills.

Statsig MCP now supports for both Segments and Layers, so you can more seamlessly manage user targeting and experiment configuration using your AI workflows.

What you can do now

• View full segment definitions and create new segments (rule-based or ID-based)
• Update existing segments, including rule-based segments and ID-based segment membership
• View all layers and their parameter details
• Create layers and create experiments with assignment to a layer

Why this matters

Segments and Layers are core building blocks for safe, precise experimentation. Segments unlocked faster targeting definition based on a set of users or rules. Layers unlocked cleaner parameter management under high experiment volume. Now, empowering your agents with these tools will help accelerate iteration velocity and improved engineering efficiency, all while maintaining safe and consistent experiment configurations.

Try it out

If you have the Statsig MCP set up, try the below example prompts and workflows to explore the new segment and layers functionality:

• "List all segments, then show details for the segment [segment_name].”
• “Create a layer for shared signup experiment parameters.”
• "Create an experiment testing new signup flow UI and add it to the signup_tests layer."

Learn more in the docs

Statsig MCP Overview.

Abort long-running queries from Metrics Explorer to reduce warehouse load and avoid unnecessary compute usage.

What You Can Do Now

• Prevent long-running queries from tying up warehouse resources
• Avoid accidental full-dataset scans
• Limit the cost impact of exploratory queries

How It Works

Metrics Explorer queries can be canceled after 5 seconds when running on supported warehouse integrations (BigQuery, Databricks, Snowflake, and Athena). Query cancellation currently applies to individual charts and does not yet extend to dashboards.

Impact on Your Analysis

Cancel Queries let you interrupt a run, refine the query, and try again immediately. This reduces unnecessary warehouse usage while keeping exploratory workflows fast and responsive.

What You Can Do Now

• Understand product stickiness out of the box
• Separate growth driven by new vs sustained engagement
• Spot churn and reactivation patterns at a glance

How It Works

Select an event, define a unique unit, and choose a time interval. Lifecycle Charts automatically classify activity by one of four lifecycle states:

• New:
  Active in the current interval with no prior activity within the lookback window (up to one year)
• Resurrected:
  Active in the current interval, not active in the previous interval, but had activity earlier in the lookback window
• Recurring:
  Active in both the current and immediately previous interval, indicating continued engagement.
• Dormant:
  Active in the previous interval but inactive in the current one, highlighting potential churn

Impact on Your Analysis

Lifecycle Charts reveal why usage changes by showing shifts in engagement composition over time. Teams can distinguish growth from retention changes, identify drop-off earlier, and understand product stickiness without building custom retention analyses.

Check out our docs for more information.

What You Can Do Now

• Generate dashboards from an API request
• Add time series, rich text, and categorical widgets
• Integrate dashboard creation into workflows powered by tools like Codex Skills

Impact on Your Analysis

Dashboards can now be managed at scale through code. Teams can automate setup to save time and integrate it with the tools they already use. The console remains available for exploration and refinement.

Private Beta

This feature is currently in private beta for Pro and Enterprise customers.
If you'd like access, reach out over Slack.

Add structure to dashboards by organizing widgets into focused sections. Dashboard Pages help teams separate workflows and context so related signals live together.

What You Can Do Now

• Navigate dashboards with clearer context
• Group related widgets into dedicated views
• Keep dashboards performant as they scale

How It Works

Add pages inside a dashboard to organize widgets into distinct sections while keeping everything in one place.

Impact on Your Analysis

Loading fewer widgets at once improves dashboard performance and responsiveness. Teams can move between workflows faster while working with large or complex dashboards

Stay informed on key metrics through scheduled dashboard reports. Dashboard Subscriptions deliver a PDF snapshot of your dashboard directly to Slack or email on a cadence you choose.

What You Can Do Now

• Receive automated dashboard snapshots in Slack or email
• Schedule recurring updates for teams or stakeholders
• Keep visibility on important metrics without manually checking dashboards

How It Works

From any dashboard, open the “…” menu and select Add Dashboard Subscription. Configure the delivery schedule and subscribed audience. Statsig generates a PDF snapshot at the scheduled time and delivers a read-only version of the dashboard via Slack or email.

Impact on Your Analysis

Dashboard Subscriptions makes it easier for teams to monitor ongoing trends asynchronously. Stakeholders receive recurring updates as dashboards update.

🎚️ Switchback Enhancements for WHN

We rolling out an improved Switchback Experimentation model to WHN customers. The new Switchback experiment utilizes a regression-based analysis method that replaces our previous bootstrapping approach. This update brings greater flexibility and analytical power, including the ability to break down results by pre-computed dimensions, more configurable burn-in/out periods, and improved scheduling and clustering.

What is a Switchback experiment?

By alternating treatments over time for the same units, switchbacks help control for interference and capture more realistic system-level effects. Use a switchback experiment when you can’t reliably randomize at the user level—typically because treatments affect shared systems or environments (e.g., marketplaces, pricing, routing, or infrastructure).

How do I enabled the enhanced features?

Cutting over to the new Switchback model is a breaking change, and we’ll work closely with customers running legacy switchback experiments to plan a smooth migration. For customers who haven’t previously used switchback experiments in Statsig, the feature will be rolled out in the coming days. If you’re interested in learning more or getting started, feel free to reach out via Slack or your account manager.

Inline Power Analysis

• Run and view power analysis results without switching to another page
• Set and iterate on target MDEs and experiment duration
• Instantly see recommended experiment duration based on your inputs
• Quickly access results anytime—results are saved and visible on the setup page

This makes it easier to align on realistic expectations before you launch, ensuring your experiments are both efficient and statistically sound.

Inline Power Analysis is rolling out in the coming days to all of our Cloud customers.

We’ve added metrics and metric source tools to the Statsig MCP, so your agents can now easily read and analyze Statsig metrics metadata from within their workflows.

What You Can Do Now

• List metrics and metric sources in your Statsig project
• Retrieve metric definitions

This makes it much easier for MCP-powered workflows to find the right metric, inspect how it’s defined, and understand which metric sources are available before creating, updating, and analyzing experiments and gates.

Why This Matters

Before this, the Statsig MCP supported adding existing metrics to gates and experiments, but had less visibility into the metrics layer itself. Now with these tools, agents can reason about Statsig metric definitions and sources directly, making it easier to discover the right metrics, understand their definitions, and set up experiments with confidence.

Try It Out

If you have the Statsig MCP set up, try the below example prompts and workflows to explore the new metrics functionality:

• "What metrics do we have related to user retention? Pull their definitions and suggest which would work best for a 7-day activation experiment."
• "Pull up the definition for the metric [metric name]"
• (For Warehouse Native projects) "What metric sources do we have configured?"

To set up the Statsig MCP server and explore all the capabilities it supports today, see our docs page.

We’ve enhanced our variance reduction methodology in Cloud by automatically applying Stratification to better account for users with missing pre-experiment data.

How it Works

Users are grouped into strata based on available pre-experiment information. Treatment effects are then calculated within each group before being aggregated into a single estimate and applying standard difference-in-means and variance calculations.

Impact on your Analysis

This improvement retains more users in analysis while still applying variance reduction wherever pre-experiment data exists. The result is tighter confidence intervals for more reliable decision-making.

What it can do

With the Statsig ChatGPT App, you can:

• Bring Statsig into your analysis by pulling experiment context and results, and asking questions like “what moved?” or “what should we do next?”
• Compound knowledge across your connected tools by tying Statsig data to the docs, notes, and discussions you already have connected in ChatGPT
• Build a flywheel of rapid iteration by generating insights and instantly creating or updating Statsig gates, experiments, and dynamic configs, all within ChatGPT.

Why this matters

ChatGPT is increasingly embedded in the day-to-day workflows of product builders. The Statsig ChatGPT App enhances that workflow, letting you plan with real results and ship the next step immediately —all in the same thread.

Getting started

• Install the Statsig ChatGPT App.
• Connect your Statsig workspace (you’ll be prompted to authenticate).
• Start exploring — and shipping.

For more setup instructions, see our docs page.

Try these prompts

• “I’m writing an experiment review — pull the latest results and context for [experiment name] and draft a summary.”
• “Based on this PRD, create an experiment with two variants and the key metrics we mention.”
• “Create a feature gate for [feature] and roll it out to 5% of users in [segment].”
• “Create a dynamic config called [name] with these keys, and suggest a safe rollout plan.”

🎚️ Analyze Only Switchback in Beta

WHN customers can now provide their own Switchback assignment data for a regression based analysis in Statsig. This allows users to have maximum flexibility by controlling their own assignment method while leveraging already existing metric data in Statsig.

To get started you will need to first create a dedicated assignment source for Switchback experiments.

And select Switchback as an Experiment type in the Analyze Only experiment creation modal.

The feature is available today in beta. If you’re interested in learning more or need help getting started please reach out to us via Slack or through your account manager!