Imagine you have just built a landing page from scratch. You have 10 sections, each presents different content to customers – your goal is to get the most engagement possible:
Now that the page has launched, you want to optimize the engagement (let’s say clicks) to this new experience. When you look at the clicks to each section, you see a fairly predicable decline in clicks as customers scroll lower and lower on the page. How do you know that Section 1 is actually performing well or if it’s only doing well because it’s at the top of the page? The answer lies in knowing your impressions. Not all customers are going to see Section 10 and if we knew exactly how many customers saw Section 10 we would know just how valuable those clicks are to that widget. We need to know the denominator:
Getting impressions can be hard for even the most sophisticated digital platforms. This is because the content in each section needs to be aware of how far down the customer has scrolled, then register an impression when the customer reaches that point on the page. If you are unable to get impressions, you can also look at the average scroll depth for the page to approximate how many customers saw each section. For example: if you know only 10% of customers scroll down to Section 10, you can reasonably assume that 10% of page visits or 100 impressions would have encountered that area of the page.
Now that we have our impressions, we normalize the data. For example: if we know that only 50% of visits scroll down to Section 7, we can reasonably assume that those clicks are worth 2X as much as our sections that are above the fold and receiving 100% of the impressions. Thus, Section 7’s normalized clicks = 400, not 200. By normalizing the data we can find opportunities in how we prioritize content on our page:
If we have decided to optimize this page based on clicks, the correct order for the content on the page would be by “Actual Rank” which uses the normalized clicks to tell us what is actually important to customers.
The ONLY thing better than manually optimizing your page layout is automatically adjusting the layout based on a multi-armed bandit algorithm or multivariate test of different layouts, but that requires further sophistication and an explanation for another day!