Measuring Content Success Part 2: Normalising data

To measure success, you need two key things: something to measure, and a metric for success. 

In Part 1 (gathering data) of this series, we covered how important it is to gather enough data. Now you should have your ‘something to measure’, we need to look at how we measure and analyse the data.

Every piece of data is quite unique. The success of an individual post relies on many things: how many followers the company has, the time of day, the contents of the post - we could go on. 

How do you accurately measure success while taking these into account? Did a post get low engagement because it was genuinely bad, or because the organisation posting it had a small following?

To determine this, we need to normalise the data.

What is data normalisation?

Data in isolation can be misleading. 

For example: Johnny’s house has five bathrooms, but Areeba’s only has one. From that data, we could guess that Johnny is more successful.

Context changes that. It turns out that Johnny lives with four other people, while Areeba lives on her own. A better measurement of success would be to look at how many bathrooms there are per person.

To level the playing field, we normalise the data by the number of people. Johnny and Areeba are on equal footing: they both have one bathroom per person. 

In the same way, looking at a social post in isolation may give you unreliable metrics. We need context.

We need to compare apples with apples

What does success look like for your company? Impressions, engagements, subscribes? In this case, we’re going to look at impressions (A.K.A. the number of times your content was seen, counting multiple views from unique users).

In the previous example Johnny had five bathrooms and five people in the house. 

Imagine we’ve got a LinkedIn post that has 5 impressions and 500 followers on the company page. 

Previously, we measured the number of bathrooms per person. 

Now we’re measuring the number of impressions per follower. 

“Normalisation involves removing elements of a signal in data, which are not related to the effect we are interested in observing. Think of it like noise cancelling headphones,” our Chief Data Scientist, Oliver Paul, shares. 

Normalisation involves removing elements of a signal in data, which are not related to the effect we are interested in observing.

“For example, a post from Google getting 40,000 more impressions than a much smaller organisation’s post has nothing to do with the quality of each post respectively. It is likely that this difference is caused by Google having a much larger audience than The Content Engine. 

“If this hypothesis is true, then we want to remove the audience size from the equation. We can do this by normalising impressions by follower count or engagements.”

In the image below, Ollie compares impressions across several organisations. On the left is the original number of impressions, and on the right the data has been normalised. 

The resulting metric is the number impressions per follower - and the companies’ rankings have changed. Johnny’s five bathrooms are less impressive once the question goes from ‘how many bathrooms’ to ‘how many bathrooms per person’. We’ve remove some of the ‘noise’ with a filter. 

These types of insights help us understand what works and what doesn’t. The content posting cycle becomes data-driven, constantly adapted to help companies grow a more engaged audience.

If you’re interested in learning more about how we create data-driven content, we’d love to chat.

In Part 3 of this series we’ll look more closely at how we can use the normalised data to get an answer to the eternal question: when is the best time to post content?