As digital marketing techniques have gained acceptance and matured over the last few years the momentum to collect and utilise data has rapidly gathered pace.
In the customer experience and communication space ‘data’ has become the catch-all term for anything digital including contact details, transaction history, behavioural information and even content such as images or video. This has inevitably led to speculation about the value of these data.
So here is the way we think about this issue at Working Three.
The value of data doesn’t come from collecting and storing data. Rather, it follows from the process of analysing the data, developing deep insights and taking actions based on those insights. It comes when brands provide better outcomes for the creator of the data – the customer – through improved engagement or better-tailored goods and services.
In an environment where it is becoming increasingly difficult to get people to engage with marketing messages this process can generate considerable value for brands. Customer data utilised in this way can help make marketing efforts more engaging, efficient and tailored – and, thus, valuable.
Broadly speaking, there are three types of models that marketers can employ to utilise data, better optimise the marketing spend and drive marketing-focused innovation. These are:
- Improving segmentation through the use of pattern recognition algorithms
- Making accurate predictions through propensity modelling
- Filtering information served to a customer to make recommendations
Customer segmentation, otherwise known by data professionals as clustering, becomes far more sophisticated when algorithms are used to analyse customer data sets. Humans can only process a few variables related to customer segmentation. Software is not bound by that restriction. This is particularly important when a business is trying to calculate the real value of a specific customer.
Additionally marketers can rapidly break out of traditional segmentation models, which are usually based on a small number of basic demographic data points and look for far more meaningful segmentation models. These include product-based segments (algorithms that discover the type of products, and groupings of products, which people do and do not buy from), brand-based segments (algorithms that discover the type of brands, and groupings of brands, people do and do not like) and behavioural segments (the type of behaviours which people display such as purchase frequency, time spent on site, high or low engagement with marketing content and the degree to which they are influenced by discounting).
Propensity models allow you to predict the future behaviour of an individual customer or customer segment. Assuming you capture the right data, it is possible to use algorithms that compare one customer to many others to predict how much that customer is likely to spend with you over their ‘lifetime’. For example, while one customer may make a higher initial purchase they may not be as valuable as another customer who makes more frequent but smaller purchases. In this case it would make sense to focus acquisition marketing spend on the customer with a higher overall lifetime value.
It is also possible to predict a customer’s propensity to engage. Understanding how likely it is that a certain customer will click on your content marketing efforts or email communications can result in significant efficiency gains.
Another valuable propensity model measures the propensity to buy. This tells you which customers are ready to make a purchase, enabling you to target these customers with the right kind of offer. This kind of model also highlights the customers who are not ready to purchase, so that brands can target them with a more aggressive offer.
Amazon made automated recommendations famous with their ‘people who bought this product also bought…’. Employing recommendation algorithms it is now possible to go beyond the simple up-sell and provide a digital service that really helps customers discover new products and services that they will like.
Cross-sell recommendations can become one of the most useful to consumers. Rather than trying to sell a bigger version of the same product you can suggest what type of products are bought with it, thereby bundling a set of products up. This works well for apparel but can work equally well for the entertainment industry (‘pre-purchase your refreshments with your movie ticket to get express service’) and many other markets.
‘Next sell’ recommendations take into account a broader set of data to suggest the next item a customer may wish to purchase. This works best when it is presented as a value-added service. For example, if a bike company knew a customer had just upgraded her bike they could then offer a range of tools and accessories to help that customer get more use from her purchase.
Using the types of models described above companies can, and are, starting to realise the true value of the data they are collecting. As they start to understand what insights the data can generate they will begin to uncover even more value.
Ultimately, this is how the value equation needs to be thought about. It is not storing data that is valuable but the act of doing something based on utilising the data that creates value.
Some of that value can be guessed at, but in most cases, the biggest leaps will only be discovered once you actually get started.