Customer Segmentation With AI: Finding Your Unknown Unknowns

Customer Segmentation With AI: Finding Your Unknown Unknowns

Customer Segmentation With AI – One of the biggest insights I got when I first read the book Lean Analytics was that data analysis helps us to uncover different information from customers’ behaviour. There are so-called “known unknowns”, or those things that we now aware we don’t know. For example, how recent and frequent the customers are.

But there are also “unknown unknowns”. These are things we are not aware we don’t know, so we are missing out. In this article, I am going to show you how artificial intelligence can be used to uncover this information from customer survey data, web page visitors and CRM databases.

By doing this, you’ll not only be able to optimise communication with your existing customers, but you will also be able to identify characteristics, interests, preferences of your power users. Then, you can target potential power users with awareness campaigns at the top of your marketing funnel.

Imagine you offer some type of consulting or advice to your new customers. For example, which software subscription is the best for their business purposes or which clothes to buy based on their style and needs. Let’s take this example a bit further with a practical use case. Consider a company selling clothes for men in the form of a subscription model.

They can sign up to receive mystery boxes: packages with preselected clothing items. When they arrive to your website you can design an onboarding process, for example some type of survey to map out their needs better. You ask your new customers to respond to a 10 item questionnaire in which they provide you with yes or no answers. For instance how, when, and why they will be using your product. The business as usual situation would be to collect this data customer by customer.

Then, give these answers to a specialist to inspect them. Then the specialist – a marketer, business developer or a UX designer – would try to use their human judgement and domain expertise to infer patterns from this data. But even a small questionnaire like this, with 10 items, can generate thousands, if not millions of different combinations of answers. It is impossible for a human to memorise and identify patterns from data of this kind. It becomes a typical case for artificial narrow intelligence!

These algorithms are in the domain of unsupervised learning that mathematically measure how similar or dissimilar are the users’ responses. These are the so called clustering algorithms, in which the K-Means and hierarchical clustering are classic examples. Instead of giving the responses of new users to a human specialist to decide which cluster the new user belongs to, You can send this information first to a clustering algorithm.

The algorithm could spot that you have 3 or 4 latent segments of customers based on the type of clothes they would like to buy for work or socialising. For example, that you have two groups who buy clothes for work. One is composed of c-level corporate executives who seek for quality and convenience of delivery. Another group composed of hipsters who want to try new top brands. You just uncovered the unknown-unknowns. By uncovering these data insights from the clustering model, you’ll probably increase your uplifts and return on advertising spend.

Now, you’re going to use this information to create a lookalike audience on Facebook, Instagram or Google Ads to reach out to similar users with awareness campaigns. You should also adjust existing communication with current customers according to their cluster labels. It is up to you as a skilled professional to come up with messaging and imagery tailored that segment, instead of just relying on the generic ads you spread to everyone. Everything goes back to experimentation. Just be careful.

Providing personalised content without sounding too creepy and invading privacy can be a challenging task. For the hipsters, an ad with an avocado matcha latte could be a good start! So, we started with an example of survey data collected during the onboarding. But you can do the same process with “top of the funnel” data, for example, which urls, which sections they are visiting, how many blog articles they are reading, how frequently they are visiting and for how long they stay on the website.

You can also do the same thing with retained customers, using a CRM database. For this, try to start collecting data on what we call the first order features for clustering, which are Recency, Frequency and Monetary value and customer tenure. These come from the classic marketing framework, the RFM model. Let’s take a look at another use case. Consider a website where registered users can donate to different crowdfunding projects.

In this case you can measure recency with the last date of donation or the last time they logged in to the website. Frequency can be the number of sessions on the website, the number of different projects they donated to, or any feature that reflects the level of engagement. Last, the monetary factor can be measured with the total amount donated. Then, you should add second order features like customer preferences. This is usually based on which products and type of services they interact with. This might reveal real unknown factors, for example, if you donate to a photographer, you might also donate to a contemporary painter, but never to a classical pianist.

Finally, you can also inspect how socio-demographic characteristics, for example, age or gender, vary within and between each cluster. Typically, running a clustering algorithm is the first pilot project we advise companies to start with. They are the “low hanging fruit” of machine learning in business: mature, informative, easy to implement and interpret. They can create a positive momentum for you and for your company to acquire the taste for using more artificial intelligence to improve products and services.

After that, you’ll probably want to go deeper in the analysis and use other machine learning techniques that can, for example, uncover the personality and the mindsets of your typical customers. This would lead to an even better personalisation, but let’s leave that for another video… Hope you are now eager to uncover some of these unknown-unkowns in your business model. Please leave your comments and questions below, and see you next time!


Useful links:

reference – Customer Segmentation With AI

Share this post ...

1 thought on “Customer Segmentation With AI: Finding Your Unknown Unknowns”

  1. Pingback: What is Customer Segmentation? Customer Segmentation in Marketing | My Universal NK

Leave a Reply

Your email address will not be published. Required fields are marked *