pagent is an AI agent that automates A/B testing and personalization for conversion rate optimization teams. It generates hypotheses, builds on-brand variations in a company's design system, and runs statistically rigorous tests using Bayesian analysis, with human review and approval via email. The platform integrates with existing testing stacks through a single script tag and includes experimentation, personalization, and analytics features in one license.
Funding
Funding not disclosed
Founders
Product
Problem
Conversion rate optimization (CRO) teams face significant bottlenecks in running A/B tests: generating hypotheses, building variations, and managing the full experimentation cycle require substantial manual effort and time. This slows down the pace of testing, limits the number of experiments that can be run, and delays the discovery of improvements that could boost conversion rates.
Solution
pagent provides an AI agent that automates the entire A/B testing and personalization workflow, from hypothesis generation to test execution and analysis. The platform analyzes a website and business context to propose test ideas, then builds real-code variations that match the brand's design system. Teams review and approve changes via email without needing a login, or can enable autopilot for a fully automated cycle. pagent runs tests on live traffic, measures results with Bayesian statistics, and uses each finding to inform the next hypothesis, creating a continuous learning loop that accelerates positive outcomes.
Target Audience
Primary customers are conversion rate optimization (CRO) and experimentation teams at e-commerce companies, hospitality businesses, and other digital organizations that need to run more A/B tests efficiently and improve conversion rates.
Features
- AI-driven hypothesis generation grounded in customer behavior, business context, and user interviews
- Automatic creation of on-brand, real-code variations in the company's design system, checked on desktop and mobile
- Human-in-the-loop approval workflow via email, with options for preview, revision requests, or veto, or full autopilot mode
- Bayesian statistical analysis reporting lift and uncertainty for each variant against a control
- Single script tag installation that works alongside existing A/B testing tools without separate deploys per test
- Event-based analytics tracking every view, click, and purchase, with breakdowns by device, referrer, and visitor
- Personalization capabilities with behavioral and contextual targeting for audience-specific experiences
- Learning system that carries forward results to shape and improve subsequent test hypotheses