We all love a fortune teller, even if we think it’s just a bit of fun. We like the idea of a complete stranger intuitively knowing something about us. It makes us feel special; and if they hit the nail on the head, we’re hooked! What else do you know about me? Tell me more. Why? Because we’ve been seen – validated in some way. It creates a connection and we feel important, if only for a few minutes.
There’s something about the personal acknowledgement of our customers that increases their willingness to spend as well as their loyalty. In store, that’s easier to create: good product training plus polite, friendly staff and a healthy dose of quality service. Customers will return time and again for that feel good factor, even just to browse. Online, in front of a cold piece of tech, that personal connection can get lost – and your customers with it.
You all know the solution, right? Recommender systems are all the rage but, like the iPhone, it’s the newer versions that provide better functionality and increased capabilities. We’ll look at why that’s important in a minute. First let’s get a feel for how they work.
Some simple tech basics
Recommendation engines help you get to know your customers through implicit (e.g. device used, clicks on a link, location, time, etc.) and explicit (e.g. user profile, history, purchases, ratings, etc.) interactions with your interface. All of this information is collected, stored, analysed and then filtered to provide a recommendation.
Usually, they run on either a collaborative or a content-based algorithm. A collaborative system uses the relationship between users and items (obtained through purchases, ratings and so on) as well as similarities between users. The content-based algorithm only uses characteristics about the item itself e.g. price, features, colour, etc. and/or the customer’s history to make its recommendation. This system doesn’t suffer from the cold start issue (sparsity of data) but is less personal and nuanced than the collaborative version. However, the latest engines use a hybrid approach to maximise the benefits and minimise the drawbacks of both.
How it happens
User interactions are recorded and the data is collected. This is stored until needed for analysis. Information is analysed in real-time for an immediate recommendation, near real-time for a same session one or periodically using batch analysis for delayed feedback, such as a targeted email. Filtering determines what recommendations are made. Content-based filtering offers similar products to what the user likes, views or has purchased. The cluster option tells you what goes well together: how about a cell phone cover or ring stand with that? Collaborative filtering makes recommendations based on similarities with other items or users: you like this book, customers who bought this also liked…
Filtering is a significant element of the personalisation process and is vital in determining the accuracy and coverage of your recommendations. Accuracy is the fraction of correct recommendations from the possible total and coverage is the number of item or user recommendations that your system can provide.
Both are important: you need to show the appropriate items to the relevant customer data subset (accuracy), whilst ensuring that all those who would benefit from the recommendation actually receive it (coverage). An accurate recommendation sent to the appropriate customer segment = personalisation. And that’s really what enhances customer retention with the added benefits of upselling and cross-selling opportunities.
However, everything has to work well together. For example, the data collection and storage need to work with your current architecture and the recommender system software. With the ever increasing quantity of information, you also need agility and scalability built in to your system. If you’re running on legacy tech that’s not as straight forward as it sounds – it might not support the functionality you seek.
Why bother with extra functionality?
Let’s return to our fortune teller. For ease, we’ll call her Mary. She can derive information simply by observing you: body language, attire, tone of voice, etc. Some people buy into that but it’s too general for most because anyone can do that. This is like a basic recommendation engine, using your interactions to predict what you’d like.
If she’s perceptive, Mary can tell whether your happy face is a front, your relationship is a sham and your job is a filler while you look for something better. Now, you’re interested but you can still walk away. After all, it could just be lucky guesswork. Does she really know you? This is the realm of the hybrid system – enough similarities to get your attention but not necessarily make you buy.
However, if Mary’s intuitive, she can take all of that information, interpret your history and infer your future behaviour. That’s when you truly feel acknowledged: she really knows me! So you’ll listen to her predictions, trust her insight and overlook any minor inaccuracies. Mary has made a personal connection beyond the average interaction and then taken it to another level. That’s something you’ll go back for again and again. Welcome to the next gen recommender systems.
Let’s up the personalisation game
A bespoke online shopping experience equals happy customers who feel acknowledged. A company who shows its clientele that they’re more than just a number will increase their levels of customer satisfaction and retention. Why? Because personalisation is a substitute for the human touch and shows your customers that you care.
Locstat takes personalisation to the next level. We combine our recommender engines with complex event processing (CEP) and graph tech analysis. Think of it as a personal shopper, psychologist and life coach all in one! This unique fusion provides a holistic approach for a connected business. The CEP interprets contextualised data and triggers rules to implement other processes such as contacting a customer who’s about to churn.
Graph based analysis enriches the whole environment and is the most efficient way of connecting data points. It is more accurate and powerful than stats based systems due to the complexity of data it can process and interpret. Essentially, a graph database unlocks the value contained in your data’s relationships.
Furthermore, our recommendation engine has customer segmentation built in: metadata tagging on items and user behaviour data provide item recommendations for specific customers. This enhanced functionality means that you can precisely identify who to target and what to recommend. With all of this at your fingertips it’s difficult not to get personal!
