Predict A/B test outcomes before you launch.
Replica runs simulated user sessions across your control and treatment website variants, calibrated against your past A/B tests. Get predicted impact, confidence intervals, replays, and diagnostics in minutes.
Forecast the metric. See the behavior behind it.
Replica simulates user sessions across your control and treatment website variants, predicting lift and confidence intervals while showing the replays, reasoning, themes, and diagnostics behind the result.
Metric forecasts
See how each metric is predicted to move, with lift estimates, 95% confidence intervals, and a clear ship-or-skip recommendation. Segment results by user attributes for deeper analysis.
Session replays
Watch each simulated user session from start to finish, with every click, scroll, pause, and input paired with what the user was thinking in a searchable transcript.
Simulated user interviews
Interview individual simulated users to understand specific moments of conversion, hesitation, or drop-off in their sessions — or ask across all sessions to uncover broader patterns behind the forecast.
Auto-clustered themes
Replica analyzes session transcripts across control and treatment to identify recurring behavioral patterns, then ranks themes by frequency, relevance, and impact to show what mattered most across the simulation.
Searchable transcripts
Search across all sessions to quickly find moments of conversion, hesitation, confusion, or drop-off, then click any line to jump directly to that moment in the replay.
Calibrated against your real A/B test history
Replica backtests against past experiments where outcomes are already known, then tunes the models and simulated users until forecasted lift closely tracks actual lift. Once calibrated, Replica can forecast future website tests before launch.
Validated on real experiments and real user behavior
Replica’s case studies show forecasts matching live A/B test outcomes. Our research shows that simulations become more accurate when grounded in real user attributes, behavior data, and finetuning — the core methodology behind Replica.
“The fact that Replica can accurately predict whether A or B is better means we can test extremely rapidly.”
Use your existing data stack to simulate real users
Replica uses your existing product and session data to create simulated users, finetune their behavior, and run thousands of browser sessions across your control and treatment variants. In minutes, you get predicted lift, confidence intervals, session replays, transcripts, and behavioral themes before launching the test.
Replica connects to your analytics, experimentation, session replay, and warehouse tools to create simulated users matched to your real audience. We use user attributes and traffic patterns to define each simulated user, then finetune their behavior on session recordings and action transcripts.
Replica uses these simulated users to run thousands of web sessions across your control and treatment variants in minutes. Each simulated user views, thinks, scrolls, clicks, and types like a real user.
Predicted lift and 95% confidence intervals show what changed. Session replays, transcripts, and clustered behavioral themes show why. Ship or skip with quantitative signal and qualitative evidence.
StatsigAmplitudeOptimizely
Mixpanel
PostHog
Google Analytics
HotjarStatsigAmplitudeOptimizelyMixpanelPostHogGoogle AnalyticsHotjarStatsigAmplitudeOptimizelyMixpanelPostHogGoogle AnalyticsHotjar
Snowflake
Databricks
PostgreSQL
MongoDB
MySQL
Redis
SQLite
Google CloudSnowflakeDatabricksPostgreSQLMongoDBMySQLRedisSQLiteGoogle CloudSnowflakeDatabricksPostgreSQLMongoDBMySQLRedisSQLiteGoogle CloudPrioritize the right tests before production
Replica runs before live experimentation. Use it to forecast impact, inspect behavioral evidence, and prioritize which website changes deserve real A/B test traffic. It helps teams test more ideas, filter out weak candidates, and make every live experiment count.
Built by experimentation veterans, supported by the best
See how Replica performs on your past A/B tests
Share past website A/B tests where you already know the outcomes. Replica calibrates its simulations against your experiment history, compares predicted lift to actual lift, and gets Replica ready for production use on future tests.