By William Storey

Ever find yourself wondering why the users on your site do the things they do? What types of people will buy your most valuable product? Has your new website design convinced users to convert?

Luckily for you, we have compiled a handy guide to customer modelling – a neat technique for finding out exactly what makes your users tick!

Back to blog home

Customer modelling – Why should I care?

Customer modelling may be renowned as a complex topic but it is actually a very efficient and reliable way of finding out exactly why users do the things they do on your site.

Have you ever wondered why a user would choose to purchase a particular product over another? How about which factors would make a user more likely to sign up to your newsletter?

Enter customer modelling!

This is essentially a mathematical way of saying “predicting an outcome using a variety of factors which we believe to be relevant”. So, for example, we can predict how many conversions we will get today given that it is a Saturday in December, 30% of users have visited the site on their mobile and it is a rainy day.

Not only that, we will be able to tell how many more conversions we are likely to get on a Friday as opposed to a Saturday and whether increasing the reach of mobile users by 10% will have a significant impact on this or not.

Pretty handy, right?

Sounds cool but how do we do that?

I’m glad you asked! For this type of analysis, we use a technique called regression modelling. This is an advanced statistical technique used to find out to what extent certain variables correlate and affect each other. It can be carried out by most statistical packages (R, Python, etc.) and is a technique used very heavily in the world of big data analysis. For more details on the calculations behind this, click here. If maths isn’t your strong point, a great visual explanation can be found here.

The first step in any customer modelling analysis is to decide what we wish to predict. For this example, I’m going to assume that we want to find out what factors make a user more likely to convert.

Now that we know what we’re interested in, we need to have a good hard think about what could possibly be relevant to a user converting or not. Some examples include:

  • The time of year/ day of the week
  • The channel/device through which the user arrived at the site
  • The demographics of the user (e.g. age, gender)
  • What region the user is from
  • Whether the user has signed up to your newsletter or not
  • And many more….

Top tip: It is always better to think of too many inputs than too little. If it turns out the factors aren’t actually very insightful then we can remove them from the model later.

We then input all this data into our model and run the regression optimisation to produce a list of the significance of all our factors for predicting the output. From this data, we combine our statistical expertise with our mastery of web-based marketing to derive a tonne of highly useful insights.

Visual Interpretation of Regression

Insights, you say?

Yes, indeed!

This type of analysis can provide a tonne of very useful and actionable information about how your campaigns are performing. From our example analysis piece, we have found 5 insights:

  • From Monday to Thursday, the majority of users are accessing our site via desktop. This tells us that we should be upping our bids on desktop for these days to ensure that we’re reaching as many users as possible.
  • We can see that the number of users clicking on our ads each day is not significantly affecting how many conversions we are getting. This is highly suggestive that our ad does not have a strong enough call to action (or perhaps we need to direct the user to a more appropriate landing page?).
  • There is a significant correlation between users who use the site’s search bar and their likelihood of conversion. Does this search bar lead users to what they are looking for and help them convert more easily? We could test this theory out by carrying out an A/B test with the search bar placed in a more obvious position on the homepage.
  • We get significantly fewer conversions in December than any other time of year. This means that we need to down-weight our advertising spend at this time of the year to ensure that we are not wasting money reaching users at a time where they are not likely to convert.
  • Users are more likely to convert on desktop than any other device. This provides further evidence that desktop users are very valuable. It may be worth compiling a list of users who have visited the site on a Monday to Thursday via desktop and targeting these users for remarketing.

This is just an example of some of the cool stuff we can find out with customer modelling (in a single piece of analysis!). Depending on how much data we have and how deep we wish to analyse, there are many more opportunities for useful insights to be acquired.

Wow, can’t wait to get started! What do I need?

This really does depend on what you want to find out more about.

Ideally, we’d need around 3 months of data related to what it is you are trying to find out about and also any factors that you think may have an effect on this metric (remember our big brainstorm we had before?).

The more data you can get your hands on, the more insights you are likely to gain, so make sure you’re utilising all your web analytics and CRM data.

In summary…

Customer modelling is a brilliant technique for finding really useful insights within a lot of messy data. If you aren’t utilising these techniques, you should be!

Get in touch to find out how we can get customer modelling working for you.

Share this article