Applying Big Data to online merchandising

Online merchandising, also known as e-merchandising, follows the same rules as traditional, in-store merchandising insofar as it aims at delivering the right products at the right price, at the right time and in the right place for each prospect based on business insights or data-driven solutions.
The main difference is that e-merchandising can rely on big data, which is a tremendous source of information. Through data collection, analysis and activation, Big Data allows us to predict customer behaviour and to optimise merchandising by gathering insight from various sources:

• Digital signals: how do users interact with a retailer’s website or app, with its products & offers? How many of them have recently shared content or viewed a branded video?
• CRM databases: what are these users’ socio-demographic profiles and purchase histories? Do they belong to specific customer clusters?
• Product characteristics: what is the impact of each product’s availability, pricing and seasonal evolutions?

Despite all this, business issues remain essential in defining a merchandising strategy: what are the main objectives of our merchandising? Which KPIs are relevant to assess its performance? What approach should we take regarding our offer: should we push best sellers, products that have the highest margin, products with a large stock? And what about new products?
Indeed, all these questions should drive business objectives but the human brain can prove to be limited when it comes to computing and predicting the best strategy in real-time for optimised merchandising – especially as compared to algorithms and machine learning solutions.

So here is what a data-driven approach can allow e-retailers to achieve concretely.

Gather insights to inform merchandising rules

Catalogue & assortment

While marketers can form insightful intuitions about trends based on basic KPIs – what are my best sellers? What are their main characteristics? How can I find new products that are similar to them in order to expand my catalogue and ensure sales? –, big data can compute and predict trends or future best sellers according to purchase history. It can also cluster products according to observed data-driven patterns (such as price range, colour, size, style, brand…). While taking more than 3 dimensions into account is already a complex process for the human brain, big data can process multiple input dimensions to cluster products. This way, you can avoid making assortment mistakes because of someone’s hunch about customer habits.

Pricing & special offers

Just as yield management is used by many travel companies, special offers can also be computed and displayed on an ecommerce website: instead of a manual discount of 30% on all products you want to destock, big data can help you predict which discount will be most effective for which product, and thus optimise your margin, revenue or stocks depending on the business objectives that lie behind the special offer in question.

Localised offers

For e-retailers, implementing a global vision while also maintaining local specificities is key. Big data can help them find clusters of countries with similar purchase habits so they can adapt their sales strategy. It can also detect trends that are not global but local, thus allowing them to modify their catalogue accordingly and to recommend products for specific regions or cities. Such insights or analyses can be taken into account much faster than if they had to be processed by a human analyst and then to be cleared with the board before moving forward to implementation. Moreover, the trends could already have changed 4 times since data extraction. The following Netflix story speaks for itself: the media leader set up a contest in order to find the best way to organise its movie list. Thanks to his knowledge of movies, their ratings and their genres, a human being managed to win against algorithms! But the human brain hit its limit when the winning selection was to be refreshed regularly. Indeed, given how quickly trends evolved, he would have had to recompute everything every day, which would have taken a lot of time and left a lot of room for mistakes…

User segmentation

Knowing who their customers or prospects are is another key issue for e-retailers. With a basic CRM database, you can already segment your audience based on characteristics such as location, socio-demographic profile… but Big Data enables you to enrich your analysis with multiple dimensions and new data such as on-site (or in-app) behaviour and navigation paths, customers’ purchase histories or even third-party data in order to identify their “big moments” (moving house, having a baby, starting a new job, etc.).

Ensure a real-time optimised shopping experience

Below are the 2 main practical areas on which we generally focus at 55 when taking advantage of big data to enhance the way e-merchandising is handled by our clients:

Optimise product ordering

On an ecommerce website or app, product ordering is essential to allow visitors to discover the retailer’s catalogue and then land on a product page (and hopefully make a purchase). In the same way as a Google search result ranking is vital, optimised product ranking is also key to success, since users tend to focus their clicks and attention on the first results appearing on top of a list.
When deciding how to order products, you should always at the very least comply with basic business rules, as you don’t want to be pushing out-of-stock products or winter collections during the summer sales. But in addition to collecting variables such as product characteristics, browsing history or purchase data, it is also possible to combine all soft signals so as to predict user interactions. Of course, all these signals and input data for machine-learning algorithms must be defined and selected by a human brain – preferably one working at 55 – in accordance with specific business objectives. But then algorithms weigh each variable in their model in order to deliver the best prediction, and therefore the best user experience. Last but not least, Big Data also enables us to test hypotheses, such as whether a brand new product will perform well. By pushing the product in question to the top of lists, we can gain quick insight into its inherent performance in order to validate the hypothesis. At fifty-five, thanks to our methodology and algorithms, we helped a retailer achieve a 4.5% uplift in its revenue per visitor in the 1st quarter of 2017.

Customise product recommendation

Let’s start with a simple figure: 35% of Amazon sales are generated through their recommendation engine, which suggests popular products to new visitors. Product recommendation is fundamental because, given how extensive ecommerce catalogues tend to be, there is little chance that a visitor will browse through it all when visiting a website or an app. This is precisely why it is important for retailers to display relevant products when a visitor lands on their website in order to customise the customer’s shopping experience in real-time, for instance by suggesting products related to the visitor’s current or previous browsing, last products viewed, wishlist, add-to-cart selection and past purchases. Hopefully this does not mean that everyone will buy the same products as, thanks to big data, we can deliver personalised recommendations – which means unique lists of recommendations for each visitor or for clusters of visitors.

As you can see, not only does Big Data enable retailers to analyse and gain insight into their own offer or product strategy, it also helps them automate their e-merchandising in a smart way that brings value to their customers. Machine learning can compensate for the lack of human interaction with a vendor on a web interface, and it can go even further thanks to its computing capacity. So it might not be a coincidence that Early Birds – a French start-up specialising in e-merchandising – named one of its predictive solutions “Personal Shopper”.

Want to get started with machine learning? Here are a few tips and key tried-and-true success factors from 55 for every machine learning project:

• Start from business objectives to build the project
• Build a strong architecture that enables automation
• Always test the model to measure uplift & speed up the learning curve
• Set up alerts in case an issue should arise
• Anticipate special events that can impact the model: private sales, social buzz (whether good or bad)