Project description
TranslatedWe have a growing web product and want to stop guessing which changes actually impact metrics. The goal is a full-featured experimentation platform that lets product and marketing teams run controlled tests without backend rewrites each time. An experiment is defined by choosing a target audience, describing two or more variants, assigning traffic distribution (e.g., 50/50 or weighted), and selecting one or more conversion goals: signup, click, completed purchase. After launch, the system must reliably keep each visitor in the same variant across sessions so results stay clean.
The heart is the statistics engine and results dashboard. For each active experiment, you need to see visitors, conversions, and conversion rate by variant, plus correct significance calculation to understand when results are trustworthy and when it's just noise. We ask for confidence intervals, lift versus control, and a clear indicator of whether the test reached significance and gathered enough sample size. We also want protection from premature conclusions: the dashboard should tell us when it's too early to decide. This is an experimentation tool, not a feature flag switch and not standard web analytics — the focus is on controlled tests and honest significance math.
For site tagging, we need a lightweight SDK, preferably JavaScript for the browser plus a simple server version that receives the assigned variant and sends conversion events back. It should be fast, cache assignments, and degrade gracefully so a broken experiment never breaks the page. The result must be production-ready and documented so our team can run experiments independently after handoff.
— Experiment builder: variants, weighted traffic split, audience targeting, multiple goals
— Stable per-visitor assignment, consistent across sessions and ideally devices
— Statistics engine: conversion rates, confidence intervals, lift, significance check and sample size
— Results dashboard with variant metrics and clear recommendation to continue or stop
— Client and server SDK for variant assignment and event tracking with graceful fallback
— Documentation and short handoff so the team runs experiments independently
The heart is the statistics engine and results dashboard. For each active experiment, you need to see visitors, conversions, and conversion rate by variant, plus correct significance calculation to understand when results are trustworthy and when it's just noise. We ask for confidence intervals, lift versus control, and a clear indicator of whether the test reached significance and gathered enough sample size. We also want protection from premature conclusions: the dashboard should tell us when it's too early to decide. This is an experimentation tool, not a feature flag switch and not standard web analytics — the focus is on controlled tests and honest significance math.
For site tagging, we need a lightweight SDK, preferably JavaScript for the browser plus a simple server version that receives the assigned variant and sends conversion events back. It should be fast, cache assignments, and degrade gracefully so a broken experiment never breaks the page. The result must be production-ready and documented so our team can run experiments independently after handoff.
— Experiment builder: variants, weighted traffic split, audience targeting, multiple goals
— Stable per-visitor assignment, consistent across sessions and ideally devices
— Statistics engine: conversion rates, confidence intervals, lift, significance check and sample size
— Results dashboard with variant metrics and clear recommendation to continue or stop
— Client and server SDK for variant assignment and event tracking with graceful fallback
— Documentation and short handoff so the team runs experiments independently
Software development Cash, Bank transfer
