How to get Google quality search and recommendations for your online store

More and more customers are looking online to browse and purchase the products they are interested in. In the EU, e-sales accounted for 17.6% of turnover on average. However, only 23% of enterprises are leveraging this opportunity. For large enterprises, e-sales account for an even larger share of turnover with an average of 23.1%. It is clear that this is an essential part of business for many companies.

Ensuring a good customer experience on your company’s digital home should be top of mind. Forbes states that 71% of customers feel frustrated when a shopping experience is impersonal. I am sure we have all had the experience of looking for a product at a specific store, but getting so frustrated with the search that we go back to Google and type [Product Name Here] [Retailer name here] after which we are finally presented with the exact article we are looking for.

In this article, we will introduce Discovery AI for Retail. Google’s product is aimed at ensuring that your customers have a good experience with searching on your website or application and helping you provide them with personalised recommendations.

In this blog post, we will discuss: 

• What the Discovery AI product suite can offer you
• The different components it is made up of
• How you can leverage it in your company

 

Introduction to Discovery AI for Retail

Discovery AI for Retail is a managed service offered by Google, which is built according to the principle of Bring Your Own Data (BYOD). This means that you only need to present your data in a specific format and Google Cloud will take care of creating models personalised to your business. There are currently two main offerings provided in the platform:

• Retail Search: providing Google quality search algorithms to use directly on your website. Making it easy for users to find what they are looking for and making their checkout process quick. This is extremely important to prevent customers from getting frustrated and either stopping their browsing session or moving on to a different retailer where the product is more easily found.
• Recommendations AI: enabling you to present your customers with personalised recommendations of products they might be interested in. Different models are provided that optimise for different objectives. Examples are “Others you may like”, “Frequently bought together”, “Recommended for you”, etc. A full list can be found here.

Getting your data in

There are two main types of data to be considered. There is the catalogue, which represents the products available in the store and there are user events, which represent users interacting with the catalogue.

The Catalog

Discovery AI for Retail represents products in a catalog. This catalog stores products in a hierarchy which enables it to reason over these relationships when training models. The following important concepts make up the catalog:

• Products: a product represents one entity in the catalog
• Product attributes: product attributes provide information about a specific product. These attributes are divided into predefined and custom attributes.
  • Predefined attributes are set up by Google and it is highly recommended to populate them for each product. Examples are brand, colour and size.
  • Custom attributes allow you to add additional properties to a product which can then be used to customise the behaviour of Discovery AI.
• Product levels: product levels represent the hierarchies in the catalog. There are three different types:
  • Primary:  this is what is returned in search or recommendation. It can be either one item or a group of similar items. An example could be “V-neck shirt”
  • Variant: a variant is one version of a primary product. An example could be “Red V-neck shirt, Size L”
  • Collection: a collection is a bundle of products or variants that have some relation. An example could be an entire outfit or a jewellery collection.

In order to load data into the catalog, a business either needs to transform its existing data into the correct format OR if you are already a Merchant Center customer, you can use the built-in integration to make this process quick.

User events

User events represent interactions that a user makes with the website. Not only actual purchases are important here. There are seven types of user events:

• Add-to-cart: Adds product to cart.
• Category-page-view: Views special pages, such as sale or promotion pages.
• Detail-page-view: Views product detail page.
• Home-page-view: Views homepage.
• Purchase-complete: Completes checkout.
• Search: Searches the catalog.
• Shopping-cart-page-view: Views shopping cart.

All of these events play an important role in accurately representing user interest. Of course, they are not equally important. A purchase is a much stronger indicator than merely viewing a product page.

In order to import user events, they need to be transformed into the proper input format. They can then be loaded in from Cloud Storage or BigQuery. If you are already a Google Analytics 360 or 4 customer, this can again be done very easily without any mapping.

Getting value out

Once data has been loaded into the platform, it is time to set up the models so we can call them from the website. This is done by defining the serving options for each model. This allows you to customise behaviour to for example boost specific products so they are displayed more (in case of excess stock, promotions, etc.) We will not go into detail on that in this article, but the interested reader can find more here.

Once models are available, the website needs to call the Discovery AI API to fetch predictions. This is a simple REST call, where the structure depends on the model being called. Some examples of different points where the website might make an API call are:

• Search Autocomplete: makes suggestions while customers are typing
• Text search: presents the most relevant products when a customer types something in the search bar
• Browse Search: provides the most relevant products when a customer is browsing the website
• Fetching recommendations: getting recommendations from different types of models for example:
  • Displaying items frequently bought together on a product page
  • Displaying items recommended specifically for this user on their homepage

These calls are where Discovery AI is providing value to your business by enhancing the customer experience.

4. Monitoring performance

In order to be able to monitor the impact that Discovery AI is having, it is important to use attribution tokens. These tokens are attached to predictions made by Discovery AI and should then be attached to the customer’s session. 

By including this token in the user events that follow, Discovery AI will be able to assess the impact that its predictions are having and learn over time. This important feedback loop will allow it to adapt to user behaviour and improve performance over time.

Another important subject on the monitoring topic is that of A/B testing. Where website performance is tracked across different experiments. Different users should be divided into two groups, where the experimentIds field indicates which group a user belongs to.

By adding the experiment id, Discovery AI is able to compare metrics of the versions of your application and give you insight into performance. This will allow you to properly assess the impact Discovery AI is having on your e-sales.

More revenue and an improved customer experience

With economic forecasts turning out more favourably than expected, consumer spending is expected to follow. This means more people will be looking at your website, trying to buy the products they are interested in. This means that you need to be ready to provide them with the best experience possible or lose out on their business.

Discovery AI for Retail can help you drive more revenue and improve your customer experience and perception. As shown in this article, getting it set up is a matter of providing your existing data to the platform. If you are unsure of how to get started or would like an expert opinion, we are here to help!