Documentation
What is ExperimentEngine?
ExperimentEngine is an AI-powered CRO insight engine built by the Adasight team. It connects to your Amplitude instance, automatically analyzes funnels and user behavior, and generates ranked experiment hypotheses every week.
The goal: continuous optimization cycles that don't depend on manual data pulls or ad-hoc analysis.
Architecture
The system runs on a weekly loop:
- Data Pull — Mac Mini cron job fetches Amplitude data via API (funnels, events, segments).
- Analysis — Claude API processes the data, identifies anomalies, trends, and opportunities.
- Report Generation — Weekly insight report with visualizations and actionable takeaways.
- Hypothesis Ranking — AI generates and ranks experiment ideas based on expected impact and effort.
- Knowledge Base — Every report and experiment result is stored and searchable.
- Feedback Loop — Your approvals, rejections, and notes train the next cycle.
Getting Started
- Navigate to Clients and select your brand.
- Review the weekly dashboard for the latest insights and metrics.
- Browse the Hypotheses tab to see ranked experiment ideas.
- Approve or reject hypotheses — your feedback improves future suggestions.
- Use the Chat tab to ask questions about your data or brainstorm ideas with the AI partner.
FAQ
How often does the data refresh?
Weekly. The Mac Mini runs an automated Amplitude data pull every Monday morning.
Can I add my own experiment ideas?
Yes. The Hypotheses tab will include a way to submit your own ideas alongside AI-generated ones.
Who has access?
Currently limited to Adasight team members with @adasight.com Google accounts. Client-specific access (e.g., Zain for OnTopUp) can be configured via Cloudflare Access.
What data does ExperimentEngine see?
Only Amplitude analytics data for the configured brand/project. No PII, no payment data. The system reads events, funnels, segments, and experiment results.
How do I report issues?
Reach out to Gregor or Elvi on Slack, or open a ClickUp ticket in the Adasight Hub project.