Google Cloud Retail Search and Browse Console deep dive

Key Takeaways

[Music] in addition to e-commerce search and browse functionality Discovery Solutions enables retailers to deliver AI driven product recommendations designed to increase conversion and customer engagement what models and customizations you choose depends on your business and user experience needs and where you plan to display the resulting recommendations for example the recommended for you model is typically used on the home page it predicts the next product that the user is most likely to engage with or purchase based on the shopping or viewing history of that user another example is the frequently brought together model it is useful when the user has already indicated an intent to purchase a particular product or a list of products and you're looking to recommend complementary items this recommendation is usually displayed on the add to cart or shopping cart Pages Discovery Solutions use machine learning models to optimize product recommendation predictions for a particular business objective which determines how the model is built with the impressive flexibility of Google AI you can combine different optimization objectives and models to support your business goals Discovery Solutions allow you to optimize AI models for the following objectives optimizing for click-through rate emphasizes engagement you should optimize for click-through rate when you want to maximize the likelihood that the user interacts with the product returned in the recommendation conversion rate optimizing for conversion rate maximizes the total purchase outcomes over all customer sessions if you want to increase the ratio of conversions per session optimized for conversion rate Revenue per visit optimizing for Revenue aims for products that users usually purchase together it is default only for the frequently bought together recommendation model that is useful when the user has already added a particular product to a shopping cart and you're looking to recommend complementing items for that product of course Discovery Solutions enable you to specify additional model configuration options to adjust the behavior of your model for example you may select a tuning preference that is most suitable to your website and business goals model tuning can be automatic performed every 90 days or you may manually tune your model only when there is a significant change in your catalog or events tuning keeps the model training optimal to automatically adapt the changes in your input data and customer shopping Behavior throughout the year besides model tuning Discovery Solutions exposes additional configuration options like Dynamic price re-ranking for predictions feature orders the recommended catalog items with a similar recommendation probability by price with the highest priced items listed first enabling price re-ranking help you to balance conversion rates and average order values because relevance is also used to order the search results shown to the user enabling price re-ranking is not the same as sorting by price to get a more detailed look at the retail admin interface Gooseberry controls faceting and attributes let's proceed with a console demo after you find yourself in the Google Cloud console go to the retail console using the navigation menu on the left the scrolling all the way down to the artificial intelligence action here select the retail option in the retail dashboard you can see the basic information about your retail project and its products for example the revenue reports also there is information about errors that correspond to data anomalies data input failures input issues or unjoined events here you can see that the retail API has been enabled as well as retail search and recommendations AI overall this retail default page is a high level information dashboard that provides you with essential reports and notifications first I would like to show you how to implement a boost rule in the retail console go to the evaluate page and select the search tab let's have a look at the results of a simple search query without any rules applied now I'm looking for dresses this is a website that sells old or vintage postcards so retail search processed my query and return highly relevant results of different postcards of women in dresses what I'm going to do is apply a boost rule to specific dresses so they appear at the top of the search results list in the panel on the left we have all attributes listed it's like a list of parameters we can use for boosting for example here's the colors attribute and there are three dresses that possess the purple color attribute next let's create a rule that will boost up purple dresses to implement that rule go back to the retail menu and select the controls option because the Boost barrier configuration is what we call a serving control click create control let's name it boost underscore purple underscore dresses and make it a search Rule and assign a boost Berry control type to it click continue here Define what will trigger the rule in this case I'm gonna say dress because it's a partial match that I configured it will work for all kinds of queries like dresses dressing and others starting with the word dress next let's set an exact match for dresses so these two things will trigger the rule when it's active I don't want to activate a Time range because it basically means when the rule applies what I'll do is just leave it blank for now and move on to the next step here to boost or bury a product I should provide some type of filter expression remember that I want to boost purple dresses and by that I mean I want to get the same number of dresses in the search results as I always get but I want purple dresses to be rated higher than the others I do that by using the color attribute so what I'm going to do is set attribute colors Now set the condition that would say any of purple it means that for the attribute colors if any of the following matches purple it triggers the rule meaning if any of the dresses has a value purple in the colors attribute it will be boosted now I'm setting a value to boost and I'm assigning it a high value you can pick any value between negative 1 and 1 with negative 1 burying a product down the list positive value will make retail treat the product as a special and elevated next select the serving config think of it as a specific place to contain your serving configuration for search we can only have one a default underscore search finally click continue and submit it worked now I have my Boost underscore purple underscore dresses rule go back to the devaluate tab and click search what I'm going to do now is search for address this should bring back the boosting Rule and hey look at that we now have the same dresses but the purple ones are listed first as top results this is the example of how we can use the purple attribute color to boost the applicable results and how to boost rule functions to implement the discovery Solutions you should Define the infrastructure and the source code and understand what must be requested and how to get working instances that can serve traffic the preliminary setup requires you to perform only managerial steps and no work on the development side approximately the whole deployment process starting with the design phase and resulting in a launch takes anywhere from 8 to 12 weeks depending on your solution complexity during this time you define and approve the architecture solution requirements and importing methods next you should Implement and operationalize the data import processes for product catalog and using event data and finally integrate your application workloads with the retail API for experimenting and testing you may use your existing or preferred Frameworks and tools to segment customers between experiment arms measuring differences in outcome quality for example leverage an AB experiment to validate how integrating Discovery Solutions improves your business objectives and identify what settings are more meaningful and effective for your desired user experience a b testing also allows you to compare results from different application versions and see what configuration better suits your goals in general it means splitting traffic into two groups and then comparing the results of Applied changes the experimental group receives a different treatment for example search and recommendation results from Discovery Solutions the control group does not customer outcomes known as user events are collected and measured for both groups to accurately determine the impact of incorporating Discovery Solutions with your website when you use Discovery Solutions you use the retail API to adjust user event and catalog data and to serve predictions or search results on your website the retail API uses the same data for both recommendations Ai and Retail search so if you use both you do not need to ingest the same data twice you can use the retail API to get personalized results for your website whether or not you are using additional Google tools this is how you can improve conversion and Order value by personalizing the shopping experience and drive engagement across channels through relevant recommendations that's what the future of retail is like and with Discovery Solutions being the essence of Google's enduring experience and innovation in search indexing retrieval and ranking you can end up winning by a mile [Music]