Loyalty Marketing & Rewards Programs
A commissioned study conducted by Forrester Consulting on behalf of ComarchDownload Your Free Copy
Machine learning techniques enable marketers to go beyond traditional personalization methods by processing huge datasets and detecting patterns within customers’ behavior. This may be used as an input for the next generation of recommender systems, enabling organizations to reach the level of hyper-personalization of their marketing activities and sales offers.
Before discussing the actual machine learning tools and techniques that can be used for achieving this, let’s consider three general types of data that organizations are usually able to collect, and which can later be utilized by the ML systems for the discussed use case:
Customer interaction histories provide a very broad scope of information about clients and a gold mine for ML systems. Whether such data refer to purchases, queries, reservations, requests, meetings or other interactions, they are a valuable payload of information about the customer. The CRM system is an obvious source of such data, but it is important that it captures as many interactions as possible, both online and offline if feasible.
Customer ratings are the core of recommender and personalization services in organizations such as Amazon, Netflix and Spotify. What is important is that ratings do not necessarily need to be explicit, for example, in the form of a numerical score or a free-form descriptive comment. Ratings may also be implicit, reflected for example in a customer’s choices and behavior. For instance, browsing through products from the same category on an e-commerce website and then picking one can be considered as an implicit positive rating of a chosen offer. Similarly, if a regular customer suddenly becomes inactive after receiving a particular product or service, it may mean that the quality did not meet their expectations and an action is required in order to retain them. These are very subtle decisions that become extremely complex on a large scale, which is exactly the kind of scenario where ML/AI systems may provide assistance.
Although the history of customer interactions is generally considered as a superior source of data for predicting a given customer’s behavior, demographic information about the customer remains an important input for marketing personalization systems. Traditionally, this information has been used for collaborative filtering recommender systems. These were designed to suggest products chosen by customers similar to the potential buyers. Similarity in this context usually meant age group, location, preferred sales channel, purchase history, etc.
Now that we have established the main types of data that can be fed into a machine learning system, let us turn our attention to some of the available ML/AI methods that can be used for marketing personalization purposes, and which we are currently exploring as part of the Research and Development arm of the Comarch Loyalty Management platform.
An artificial neural network [ANN] is a category of software modeled on the way the human brain and neural system operates. From the marketing perspective, ANN-based systems can be utilized to assist or fully automate some of the decision-making processes. In order for ANNs to calculate predictions, they need some reference data for a process called “training”. In the discussed use case, the entirety of historical data collected about an individual customer could be used, including their demographic data, ratings and details of all interactions with the originating organization. ANNs are capable of processing enormous volumes of data, and adapt their calculations depending on the training datasets they receive. This means that the same model can be re-used for every customer, as it can adapt to the training data without any additional manual input or corrections. A trained ANN can then be utilized in a number of ways, for instance:
ANN is a very powerful and versatile technology with numerous sub-types and use cases. A vast majority of machine learning solutions are either fully or partially based on some type of ANN system.
Another interesting machine learning technique is a method originally developed for the purpose of natural language processing [NLP]. Its goal is to detect the latent meaning of a word by analyzing other words that are frequently used in close association or in a similar context. From the marketing perspective, this approach allows similar products to be recognized by analyzing items that are frequently purchased together. This relies on the assumption that, if products A and B are frequently purchased with products C, D and E, it is probable that there is a relation between A and B ; the products belong to the same category, are purchased by similar customers, are complementary, etc.
Detecting such relationships between products may enable an organization to run personalized up-selling or cross-selling campaigns. Using the previous example, if product B was a more expensive version of product A, a marketing activity promoting product B to customers that have purchased A in the past could be launched, with the goal of up-selling a premium brand to a likely target group. Similarly, if B was out of stock for some time, customers could be offered A as a potential replacement.
Collaborative filtering is one of the methods frequently used by the conventional recommender systems. It is based on an assumption that customers with a similar purchase history are likely to purchase similar products in the future. Therefore, if one out of two similar customers purchased some item, recommending it to the other one can potentially be very effective. This straightforward assumption powers a vast majority of commercial recommender systems.
As machine learning gained momentum, there have been various attempts to use it to achieve a similar or better result. The most popular solution is to use a deep (multi-layered) artificial neural network, most frequently in the form of an autoencoder.
To sum up, the biggest advantage of machine learning systems is that they are designed to process very large datasets, and to automatically detect patterns that would otherwise be very difficult to discover. The actual results and added value of using ML for offer personalization depend on a variety of factors, most importantly on the volume and quality of data used for the process of algorithm training. Although, they may not be the ultimate solution for certain use cases, investing in ML-driven systems definitely gives potential for optimizing marketing processes, reducing overall campaign costs and improving conversion rates.
Transforming the way the world works