The Next Step in Customer Loyalty: Predictive Marketing 

Maria Marinina | Sep 4, 2017 401 views No Comments

Share on LinkedIn

A good business will give their clients exactly what they asked for by implementing a state of the art search engine with all the necessary filters. A great business will employ a natural step forward and add an AI-powered recommendation engine, since, as Steve Jobs put it, “Customers don’t know what they want until we’ve shown them.” This is because people tend to stick with the things they know and usually rely on their social circle to get new ideas about things they might like. However, a company should short-circuit this long adoption cycle and get their products and services in front of potential buyers as fast as possible, using technology.

The Role of Big Data

The current hype is all about creating experiences, customizing the relationship with the client by learning their preferences and acting in consequence, while maintaining high standards of data privacy. The customer-centric revolution started by Amazon is now unfolding in different industries, and it has big data at its heart.

The most valuable lesson from Amazon is that tailor-made recommendations based on the products selected, the profile of the customer and similarities with other clients’ shopping carts can boost sales up to 35%. Of course, not every online store has an army of developers and data science engineers like Amazon, but even simple, free recommendation plugins on a web-site can generate 15-20% more sales.

Recommendation Engines

However, exactly how do recommendation engines work? A good comparison is the one with a librarian. If you are visiting your local library where the librarian knows you personally and has a good idea of what you have liked so far, the next book will be based on your personal preferences. If you have just moved to a new town, the librarian has no idea what you like, so you will probably end up with a generic best-seller.

Content-based Filtering

If the system generates recommendations from the user’s input, it is content-based filtering. This technique is very closely related to search engines. The algorithm searches for items that have tags attached which are related to the initial search and gives a relevance score used to list the results. The great advantage here is that the system needs less time to learn about the user’s preferences, as those are given, such as a travel location or the content of a cart.

Collaborative Filtering

On the other hand, collaborative learning takes a different route and monitors items bought together or items purchased by people who share different characteristics. For the algorithms to work properly, it is necessary to have massive amounts of data to use as input to derive meaningful associations, not just pure coincidences. This is a major obstacle that most small companies could not overcome with their pool of data, but could be feasible by accessing the services of an artificial intelligence vendor. On the other hand, the advantages of collaborative filtering include being able to extend results to multiple categories.

The data fed into the algorithms is related to the items (name, price and attributes), the user (demographics and behavior) and contextual information (device, location and URL). These items are coded into large item-feature and user-feature matrixes that are used to compute preference scores. The structure of the data can vary widely, from simple tags to audio recordings to a personal assistant like Siri.

Best Uses of Recommendation Engines

Most recommendation engines are hybrids between content-based and collaborative filtering, meaning that they take as input what the client is looking for, but also take into consideration the background information they can find about the customer to create a consistent response.

Everyone is familiar with Amazon’s “Customers who bought this, also bought…” and “Frequently bought together” sections. These include complementary or alternative products and can be useful if the initial recommendation is out of stock or too expensive or if you want to create a package that includes a primary piece and accessories.

The algorithms also create product bundles at a price that comes at a discount when compared to the individual items. Most of the times such packages include a long-tail item. It is the recommendation engine’s way of boosting popularity for new products or those that could be of interest to a particular niche of customers and which otherwise would never get a chance to become known.

The power of recommendation engines guides clients along the sales funnel by creating a dynamic call to action (usually derived from a CRM), smart, customized pop-ups based on location and cart content, retargeting banners and even customizing the items displayed on the web-site based on your previous searches.

Possible Drawbacks & Future Directions

As the name suggests, for these algorithms to be efficient, you need large quantities of data relevant to your clientele. If you are just starting in a new market, it will take some time for the machine to learn what your clients like and the first sets of recommendations could range from irrelevant to hilarious. This problem is known as a “cold start.” Even the giant Netflix faced this issue when they tried a localized approach to their marketing and will probably have to deal with it again when they launch content in different languages. Paradoxically, the biggest pitfall of a recommendations engine could be its fine-tuning. Searching for a particular scale on Amazon comes with suggestions of drug-related items.

Another possible problem with the recommendation engines is the snowballing effect of popular articles. Once a product gains some traction, it can be presented more times and, thus become even more liked, overshadowing similar articles which are good. A possible solution to this is the long-tail bundle approach previously mentioned, which helps less popular items get some visibility by association with star products.

The general direction of recommendation engines is creating better predictive marketing solutions, customizing whole baskets based on user personas. We can imagine that the algorithms will become so performant that they will come up with a list of proposed items and instead of checking in the cart what you want, you will just uncheck the items you do not need.

Print Friendly, PDF & Email

Recent Editor's Picks:

Categories: BlogCustomer AnalyticsDigital MarketingEditor's Pick


No responses yet, why not leave yours?

Add Your Comment (All comments are reviewed by moderator, no spam permitted!)