Introduction
Tracking and attributing sales to campaigns are vital to understanding the efficacy of a test on user behaviour. Here, we'll explain how we do this.
Sales Event Tracking
In order to quantify a sale and attach it to a campaign, we must first capture the event. We do this by mapping your site confirmation page - in many cases this may be something that contains /checkout/confirmation or /thankyou but we can create regular expressions to account for any variations.
When a user becomes a customer and makes a sale, we track this as an event. Alongside this event, we scrape the order ID (for deduplication purposes) as well as any other relevant information like order value (which forms the basis for revenue attribution).
Event Attribution
To understand how we attribute sales, it first helps to cover how our attribution model works. We attribute based on session - meaning that any sales outside this window (in a previous or later session) will not count.
Within this attribution model, we use equal weighting (sometimes known as linear) in relation to campaigns. In comparison to banner advertising and PPC (which use last interaction), our model counts a sale equally amongst all campaigns that were viewed in the session where the sale was made.
We opted for this attribution model as it creates the fairest and most accurate representation of the impact our campaigns have on user behaviour. Using a longer window (like user) would measure over unnatural and arbitrary lengths of time, or create cannibalisation amongst tests.
Equally weighting all the campaigns seen during that session allows for independent analysis to be conducted into the efficacy of each test. Using a singularly weighted model (like last interaction) would give campaigns run at the bottom of the funnel an unfair advantage, as they're more likely to be seen just before the user makes a sale.