Google Vertex AI Search for Retail is a comprehensive solution for e-commerce platforms that uses the latest artificial intelligence (AI) and machine learning technology to deliver highly personalised search results and product recommendations. The technology is designed to improve the customer experience, increase conversion rates and, in the best case, support revenue growth. It is ideal for companies that want to benefit from artificial intelligence without having to develop complex models themselves.
In this article we are happy to introduce the key features and benefits of the tool, the available models and the target group the tool is suited for. So let us dive right in!
Key Features of Search for Retail
1. Personalised Search:
Google Vertex AI Search for Retail lets you show search results adapted to the individual behaviour and preferences of users. For example, if a customer often searches for a specific type of product, the system adapts to this behaviour and offers more relevant results with every subsequent search.
This feature also includes:
- Search expansion – increases the number of relevant results returned for queries that would normally return fewer results. For example, for searches that use very specific keywords.
- Relevance thresholding – lets you set the precision (the relevance of the returned search results) and the quantity of results (returning more results for a given query) of the search.
- Filtering – lets you use search expression syntax for filtering that refines the search results on your website.
- Ordering – lets you set the order of search results by several priority categories.
- Boosting results – controls the ranking of search results by promoting or demoting certain types of results.
2. Advanced Product Recommendations:
The tool lets you train and use various recommendation models to offer the products that are most relevant to the user. For example, if a user is browsing sports equipment, Google Vertex AI Search can recommend related products such as sports shoes or clothing, which increases the likelihood of a purchase. The whole system combines information from your product catalogue (e.g. Google Merchant Center) with information about the behaviour of your customers.
You can currently choose from seven models:
- Recommended for you – The model predicts which product the user is likely to buy or focus on, based on their previous purchase or browsing history. It is often shown on the home page.
- Others you may like – This model offers products that might appeal to the user, based on their previous activity and its relation to the product currently being viewed. It is used mainly on product pages.
- Frequently bought together – Suggests products that are often bought together with the currently selected product. This model is used after an item is added to the cart, on product detail pages or directly in the shopping cart.
- Similar items – This model recommends other products that have similar properties to the one the user is looking at. It is used on product pages or when a product is unavailable.
- Buy it again – Suggests products that the user has bought in the past and might want to buy again. It is used on product detail pages, when adding to the cart, in the shopping cart, in categories or on the home page.
- Page-level optimization – Automatically adjusts the recommendations on a page using several recommendation panels. It is used on product detail pages, when adding to the cart, in the shopping cart, in categories or on the home page.
- On sale – Recommends products that are currently on sale. This model is used on the home page, when adding to the cart, in the shopping cart, in categories or on product detail pages.
The recommendation models allow:
- Personalised results – all recommendations are tailored to the customer. The algorithm works with each customer's individual online behaviour.
- Real-time predictions – every prediction takes into account all the actions the user has previously taken on the site (e.g. views, add to cart, purchase, etc.).
- Automatic training and improvement of the model – the model can, for example, be fine-tuned every month so that it also accounts for the most recent data.
- Choosing the optimisation objective – For the Recommended for you, Others you may like and On sale models you can decide what the model will be optimised for during training. These three models can be optimised either for maximum click-through rate (CTR) or for conversion rate (CVR). For Recommended for you and Others you may like you can also optimise for maximum revenue per session.
- Omnichannel recommendations – thanks to the model's API you can optimise the whole customer journey – personalised recommendations in email, mobile apps, …
3. Generative AI for Complex Queries:
Thanks to the integration of generative AI, the system can handle even more complex, multi-step queries. This means that if a user searches for, say, "black running shoes for rain in size 42", Vertex AI can accurately identify and display the relevant products, which traditional search engines might not handle as effectively.
4. Integration with the Google Ecosystem:
Vertex AI Search works seamlessly with other Google Cloud tools such as Google Merchant Center and Google Analytics 4, which makes implementation and optimisation easy.
Why Use Search for Retail?
The fact is that artificial intelligence has long ceased to be a thing of the distant future. In some areas it already significantly outperforms humans. This is especially visible in the online world, where we have an extreme amount of data – a person is simply no longer able to take absolutely all the variables into account.
Search for Retail lets you take advantage of artificial intelligence and machine learning algorithms without having to be an expert in the field. The first evaluation of Search for Retail directly from Google shows a significant improvement in search results, click-through rate and turnover.
Which Companies Is Search for Retail Suitable For?
Search for Retail is a great option for companies that are considering implementing personalisation and personalised search on their websites using third-party services, but at the same time want to keep maximum control over the whole process.
Search for Retail is also suitable for companies whose data is already well anchored in the Google ecosystem – they use Google Analytics to measure traffic, use Google Ads for campaigns and have their product portfolio in Google Merchant Center.
We recommend testing this solution above all to companies that have already entrusted their data to Google Cloud.
At Archetix we have learned to work with the tool and to provide its configuration and management to our clients as well. Our data team cannot praise this service enough and we look forward to implementing it for another client.
How to Get Started?
Thanks to the integrations mentioned above, launching a pilot can be a matter of a few days. You can import historical user interactions on the website, which lets the models learn immediately, without having to wait, as with other solutions, for example half a year before the model learns to provide correct results. The solution offers integration via Google Tag Manager, and with the Enhanced Ecommerce layer you can collect the required data by modifying the relevant objects.
To start using the Search for Retail models in your business, you will need a few essential things – among others a product catalogue (Google Merchant Center or a special feed) and historical information about customer behaviour (which can be taken from GA4, or you can prepare your own dataset).
The easiest way to get started is to contact us, Archetix! We will help you set up data collection from your e-shop and the subsequent evaluation. We will recommend the most suitable type of model for your business and take care of the integration into your project.