Personalise, predict, recommend… What’s the difference?

In my last blog, I offered a technical perspective on personalisation – what it is and how it works. In this blog I am going to elaborate a little on some common confusion that arises when we talk about customisation through personalisation, predictive analytics and recommendation systems.

 

In the context of Internet based companies, personalisation implicitly refers to a system that automatically delivers dynamic content such as textual elements, links, adverts, product recommendations, etc. that are tailored to needs or interests of a particular user or a segment of users [1]. The key ingredient of any personalisation system is user profiles accommodating attributes (e.g. gender, age) and behavioural (e.g. clicks on items, website navigation) characteristics of the system users. Automatic personalisation is also substantially different from customisation and the difference lies with where the user profile is created. In customisation, it is the user herself who controls the creation of the profile and, indirectly, its content, configuration and presentation. In personalisation systems, however, it is the system that creates and updates the profile, with minimum user intervention [1].

 

Predictive analytics, on the other hand, refers to technology and methods (such as statistical methods) that learns from user’s experience (i.e. data) to make prediction the future behaviour of individuals and the objective is to drive better decisions [2]. The size of the dataset being examined is not as important as the quality of data, while the most important factor is the predictive power of the data. In other words, the value of data is determined by how much you can infer from it, which in turn is unleashed using machine learning or statistical methods. Examples of predictive analytics are fraud detection (identifying which transactions are fraudulent) and targeted marketing (finding potential customers for the given product or service).

 

It should be noted that predictive analytics has its own limitations as well as potentials. While, in most cases, accurate predictions are quite hard to achieve (if not impossible), with some certain error thresholds valuable predictions can be made. Predictive analytics is not the same as forecasting: While the latter deals with aggregated predictions (e.g. total sale of a specific shoe brand for next month), predictive analytics aims to predict, for instance, which individuals are likely to purchase a pair of shoes of that brand [2].

 

Finally, recommendation systems can be seen as data intensive applications of predictive analytics [3]. The input to a recommendation system is a past history of user interactions with products or services (accommodated in user profiles) while the output is a set of (ranked) services or items matching the active user personal preferences and satisfying her current needs. A well-known example of a recommendation system is NetFlix, which recommends new movies to see based on the ratings you gave to movies you have watched before. This system is also an example of an automatic personalisation system, where suggested movies match users’ tastes of genre, director, actor, actress, etc. and the recommendation logic is not hardcoded, but decided by collective intelligence unleashed by machine learning techniques.

 

In spite of their differences, all three variations of customisation can be instrumental in delivering an unrivalled customer experience.