About CounterFact Lab
Reduce failed experiments with offline evaluation
Why We Exist
A/B testing is essential but expensive. Most ideas don't improve your product, yet they still consume traffic, time, and resources. CounterFact Lab filters candidates offline before they reach your experimentation queue, so you only test what's likely to succeed.
What We Believe
- Decisions matter more than predictions. Offline accuracy doesn't guarantee online impact. We bridge that gap with counterfactual evaluation.
- Trust the data. Every candidate must pass rigorous quality checks before we recommend shipping. No guesswork.
- Work with your stack, not against it. We reduce the number of tests you need to run. Your A/B test stays in control.
What You Get
Upload your historical logs and new policy recommendations. We will tell you whether to ship, block, or collect more data (with clear reasons why).
You get statistical confidence intervals, quality diagnostics (data overlap, sample size, weight reliability), stress tests that simulate real-world constraints, and a complete audit trail for every evaluation.
Everything works through our web interface or REST API. No migration required.
Ready to ship smarter? Start evaluating policies in the Playground.
Somayeh is the founder of CounterFact Lab, an evaluation layer that helps ranking and recommendation teams triage ideas offline before committing traffic to A/B tests.
Before CounterFact Lab, she worked on production recommender pipelines at Capital One, built an audit-automation platform at Agero, and developed predictive models and internal tooling at Fannie Mae. She holds a PhD in Physics from Duke University.
Her focus: help teams move faster with fewer failed experiments by bringing off-policy estimators, robust diagnostics, and transparent uncertainty to the earliest stages of the ML lifecycle.
Trust markers
- PhD in Physics (Duke); peer-reviewed publications
- Led production recommender and real-time ML systems in industry
- Built and shipped fraud detection, predictive modeling, and automation in regulated settings