Freelance › Projects › Web development › Web programming › Build an A/B testing and experimentation platform
Build an A/B testing and experimentation platform
Employer
Content
Project parameters
Type of cooperationOne-time project
SectionWeb development, Web programming
Prepaymentwithout prepayment
Payment methodsCash, Bank transfer
Acceptance of requestsfrom today, 10:50 until Sep 8, 2026
Project description
We run a growing web product and want to stop guessing which changes actually move the numbers. The goal of this project is a full experimentation platform that lets our product and marketing teams launch controlled experiments without touching backend code every time. An experiment should be defined by picking a target audience, describing two or more variants, assigning a traffic split (for example 50/50 or a weighted rollout), and choosing one or more conversion goals such as a signup, a click, or a completed purchase. Once an experiment is live, the system must consistently keep each visitor in the same variant across sessions so results stay clean.
The heart of the work is the statistics engine and the results dashboard. For every running experiment we need to see visitors, conversions, and conversion rate per variant, along with a proper significance calculation so we know when a result is trustworthy rather than noise. Please include confidence intervals, uplift versus the control, and a clear indicator of whether the test has reached significance and has enough sample size to call. We would also like guardrails against peeking, so the dashboard should communicate when it is still too early to decide. This is explicitly an experimentation tool, not a feature-flag toggle and not a generic web-analytics dashboard: the focus is controlled experiments and honest significance math.
To instrument our sites we need a lightweight SDK, ideally JavaScript for the browser plus a simple server-side option, that fetches the active variant assignment and reports conversion events back. It should be fast, cache assignments, and fail gracefully so a broken experiment never breaks the page. Deliverables should be production-ready and documented so our team can create and manage experiments on our own after handover.
— Experiment builder: variants, weighted traffic split, audience targeting, multiple goals
— Sticky per-visitor assignment consistent across sessions and devices where possible
— Statistics engine: conversion rates, confidence intervals, uplift, significance and sample-size checks
— Results dashboard with per-variant metrics and a clear stop/continue recommendation
— Client and server SDK for assignment and event tracking, with graceful fallback
— Documentation and a short handover so our team can run experiments independently
The heart of the work is the statistics engine and the results dashboard. For every running experiment we need to see visitors, conversions, and conversion rate per variant, along with a proper significance calculation so we know when a result is trustworthy rather than noise. Please include confidence intervals, uplift versus the control, and a clear indicator of whether the test has reached significance and has enough sample size to call. We would also like guardrails against peeking, so the dashboard should communicate when it is still too early to decide. This is explicitly an experimentation tool, not a feature-flag toggle and not a generic web-analytics dashboard: the focus is controlled experiments and honest significance math.
To instrument our sites we need a lightweight SDK, ideally JavaScript for the browser plus a simple server-side option, that fetches the active variant assignment and reports conversion events back. It should be fast, cache assignments, and fail gracefully so a broken experiment never breaks the page. Deliverables should be production-ready and documented so our team can create and manage experiments on our own after handover.
— Experiment builder: variants, weighted traffic split, audience targeting, multiple goals
— Sticky per-visitor assignment consistent across sessions and devices where possible
— Statistics engine: conversion rates, confidence intervals, uplift, significance and sample-size checks
— Results dashboard with per-variant metrics and a clear stop/continue recommendation
— Client and server SDK for assignment and event tracking, with graceful fallback
— Documentation and a short handover so our team can run experiments independently
