Split test — is a way to compare several versions of an ad, landing page, or setup to understand which one brings in more money and conversions.
Traffic is split evenly between the tested versions in real time. For example, 50% of users go to version A, while the remaining 50% — go to version B. Based on the results, the buyer compares the metrics and keeps the version that brings in more money, while the weaker one is turned off.
In traffic arbitrage, a split test is needed to find a working setup and eliminate unprofitable options. The tool shows precise metrics: conversion rate, cost per lead, and ROI for each creative or page.
What is tested with a split test
-
Creatives: different headlines, visuals, video sequences, hooks in the first 3 seconds, calls to action (CTA).
-
Pre-landing pages: different stories (storytelling versus a news format), the presence of comments, timers, or discount wheels.
-
Landing pages: order form, button color and placement, different price or discount options, reviews from real customers.
-
Offers: comparing two similar offers from different advertisers using the same traffic.
-
Audiences and traffic sources: different targeting settings, GEOs, age groups, or ad placements.
How a split test works: an example
An affiliate marketer found a nutra offer and is unsure which pre-landing page converts better: an interview with a well-known doctor or a personal weight-loss story from an ordinary girl.
-
A flow with two landing pages is created in the tracker.
-
Traffic is distributed 50/50: every even-numbered click goes to the «doctor» version, and every odd-numbered click — to the «girl» version.
-
After both versions receive 1,000 clicks each, the buyer checks the statistics:
-
Version A (doctor): 1,000 clicks → 12 leads → conversion rate 1.2%
-
Version B (girl): 1,000 clicks → 35 leads → conversion rate 3.5%
Version A is turned off, and the entire traffic volume is redirected to the profitable version B.
Main rules for a proper split test
-
Do not change several elements at once: if you simultaneously change the headline, button color, offer price, and product image on the landing page, it will be unclear what exactly caused the conversion increase or drop.
-
Do not run the versions at different times: you cannot run version A on Monday and version B on Saturday. Audience behavior differs depending on the day of the week and time of day. The test must be run strictly at the same time.
Important: Do not stop the test too early. In traffic arbitrage, it often happens that one version suddenly pulls ahead during the first couple of hours simply because of a lucky coincidence, but over a few hundred clicks it falls into the red. Let the test collect enough data before reallocating the budget.
