Business Context

Splice is considering an experiment on the Sounds landing page signup flow to increase the rate of visitors who create an account and start a free trial. Before launching, the Growth team wants to choose a minimum detectable effect (MDE) that is realistic, statistically defensible, and worth the engineering cost.

Problem Statement

You are planning a two-arm A/B test with equal traffic split between the current signup flow (control) and a redesigned flow (treatment). Use the baseline conversion rate and traffic estimates below to determine the smallest lift the team can reliably detect in a 21-day experiment, and whether a proposed target lift is feasible.

Given Data

Assume the primary metric is binary conversion at the user level.

Requirements

1. Compute the per-variant sample size available in 21 days.
2. Estimate the MDE in absolute percentage points for a two-sided test at the stated  and power.
3. Convert that MDE to a relative lift over baseline.
4. Determine whether the proposed 0.8 percentage point lift is detectable with the planned runtime.
5. If the team wants to detect a 0.6 percentage point lift instead, estimate the required sample size per arm and approximate test duration.
6. Briefly explain how you would think about choosing MDE in practice at Splice beyond just plugging numbers into a formula.

Assumptions

• Random assignment is valid and independent across users.
• Use the normal approximation for power/sample-size calculations for two proportions.
• Ignore variance reduction methods such as CUPED.
• No peeking or early stopping adjustments.