Einstein Recipes: Transparent, Trainable, and Zero-Code Machine Learning Algorithms
Business users can now themselves build, test and train machine learning algorithms - Einstein Recipes, to promote personalized content (articles, blogs, infographics, products etc.) using a point and click user interface.
Think of being able to create algorithms on the fly for scenarios like co-buy, co-browse, smart product bundle, look-alike audiences and many more.
Sounds super interesting, doesn't it? Let's see how it works.
Interaction studio is the integrated personalization and RTIM (real-time interaction management) platform that allows you to collect data natively across channels, enrich it based on known data and then perform real-time decision & analysis thus affecting the Experience that the customer / visitor / prospect has across channels.
The below high level flow shows how Interaction Studio ingests events, processes them, and delivers amazing personalized experiences to the target audiences.
As mentioned above Einstein Recipes allow you to create machine learning algorithms to boost your content. Every recipe has 4 primary components - ingredients, filters, boosters and variations. Let's take a look at each one of them.
Ingredients are the base algorithms that make up a recipe (content / products) and they have a weight and a lookback window associated with them. Below are a few examples:
Co-Buy: This ingredient surfaces items that are purchased or downloaded by previous buyers along with the anchor item. Example: because shin guards are bought in conjunction with soccer cleats (the anchor item) during the lookback period
Co-Browse: This ingredient surfaces items that are viewed along with the anchor item by previous visitors. Example: because soccer cleats are viewed along with goalkeeper gloves during the lookback period
SmartBundle: This ingredient looks at products which are bought together in the same cart and returns recommendations based on explicitly defined categories.
Collaborative Filtering: a visitor sees products or content recommendations based on a "people like me" algorithm
Similar Items: This ingredient uses content or items similar in name or description, is included in the same category, or has been tagged the same as the anchor item being viewed or in the visitor's cart
Trending: This ingredient will show trending products or content across your site. You can define this as items/content based on the number of purchases/downloads, the amount of revenue generated, the number of views, or the amount of time spent viewing.
Filters - With filters you can do additional narrowing of the recommendations and avoid common pitfalls. Example:
avoid recommending items already in the cart
avoid recommending items already purchased
Boosters - This pull affinity data from the users profiles. Example: if categories include "Mens | Running | Shoes", then items in categories "Mens" and "Running" would get a boost.
Variation - force variations into recommendations so that the visitor sees a broad range of items.
The combination of these components helps drive powerful 1-1 recommendations. These recipes can be tested and simulated for test groups and for various anchor items to see how the recommendations are surfaced. A business user can themself modify the recipes and quickly calibrate the experience depending on the broader trend.
At JourneyBlazers we are working on delivering some amazing Experiences (personalized hero image, product recommendations, custom infobar, exit intent pop-up, next best action for financial advisors etc.) to our clients via Interaction Studio.
To learn more - reach out to us at email@example.com.