TAGS: AI CX Merchandising Online Shopping Retail
Netflix has a lot to offer when it comes to giving customers what they want.
In the world of ecommerce, it seems like retailers only compare themselves to other retailers. Step by step, web shops are becoming more and more alike. However, few retailers look at what we can learn from other industries. In this blog, I am going to cover one example of all those great ideas out there that we can use to make shoppers happier and balance sheets better.
In this example, let’s explore some ways Netflix helps users sift through thousands of mediocre films and discover new favourites.
Just like in fashion, liking or disliking films is an absolutely personal judgement. There are so many different genres and we all have our personal favourites. However, just because we have favourites does not mean we are always going to watch the same style of film. In the end, the decision very much depends on the mood we are in and who we are with.
Building a complex statistical model that can accurately model all these possibilities is practically impossible. When you log in to Netflix on a Friday night, such a model could not possibly infer whether you are with your date or your best friends. In fact, they don’t need to. Instead, they built their user interface around the idea of removing ambiguity. This way, Netflix enable their users to fill the blanks in the statistical model in a much more economical way (and without appearing creepy).
Aside from promoted content and some top picks at the top, Netflix’s homepage is structured by interests. Each row of films shows another genre. Moreover, the order of interests is personalised for every user. The higher up a row, the more certain is Netflix’s recommendations algorithm that it would be interesting for the user.
For each interest, users can pick one of the suggested films or click to view everything on offer in this category. Of course, Netflix’s recommender system is also used to rank film suggestions within the genre. Just like genres themselves, they are based on global popularity and personal signals.
Personalisation ensures only relevant content is shown to users. The grouping by genre makes it easy for users to browse content for different situations and skip everything they are not in the mood for.
How would this look like in a retail environment?
In ecommerce, we face exactly the same challenges as Netflix. With hundreds or thousands of products per category, we need to be careful not to overload users with choice. The extent to which this ‘choice overload’ influences users is easily visible when looking at click rates per position in a typical product list.
Users quickly lose attention when scrolling down a page. The further a user scrolls, the less chance the product has of being spotted and bought.
Using an approach similar to Netflix’s, one can increase attention for relevant products even if they are not shown at the top position, effectively making the distribution less steep.
In the simplest form, one can place banners at the top of a category page that allow users to filter the results. Essentially, it is little more than a glorified facet filter that would also have been available in the left-hand navigation. However, highlighting the different interpretations to the user makes it more obvious to the user that a selection will help improve results and increases the probability the user will actually click on a filter. That helps users help themselves.
We can further improve the interface by using actual products. For example, Kurt Geiger uses rows of products to not only highlight the ambiguity in the search ‘boots’, but also to promote some of their bestselling shoe trends.
The approach can be taken further if we already know something about the user. For example, if he is likely male, we can automate these rules further and make sure men’s boots are shown first. We need not stop at the target group, which, with its strict preferences, is relatively simple anyway and could be implemented with re-ranking rules. Why not create groups for styles like workwear, casual, party, etc. based on predictions of the user’s affinity for these styles; or affinities for brands and colours?
Once all groups of products are available automatic and personalised, re-ranking can be applied to ensure the most likely content groups are shown first, leading to a personalised experience that is similar to Netflix.
You won't likely find your personalisation solution by copying other retailers. But by taking cues from other industries, you might find the answers you’re looking for. Netflix keep customers happy by reducing ‘choice overload’. In retail, the same type of interface will improve the customer experience. And that, in turn, will lead to better conversion rates and improved customer loyalty.