Improve Oversight of Machine Learning Recommendations with Enhanced “Recipe” QA Feature

Evergage Blog

Ideas and Strategies for Real-Time Personalization
Improve Oversight of Machine Learning Recommendations with Enhanced “Recipe” QA Feature

September 21, 2018 by

For many marketers and business users, machine learning is still somewhat foreign. Can you trust an algorithm to make business decisions or control the customer experience? Conventional digital marketing practices tell you that you should incrementally test campaigns before launching them to the masses, which seems to contradict the approach with AI or machine learning. But it doesn’t have to.

Various forms of machine learning have been used in digital marketing for years — most notably in presenting product recommendations based on what's known about a particular shopper or visitor. We launched our own recommendations solution, Evergage Recommend, back in 2015 and have seen great adoption by our customers across industries. This is in part because Evergage can also recommend content, brands, categories and departments – elements that enable business users to leverage machine learning to dynamically change a visitor's experiences in unique ways. Evergage's machine learning also draws upon unprecedented analytics data to make the most informed decisions possible.

But the true power of Evergage Recommend has always stemmed from its customizable and transparent nature. Within the Evergage platform, business users can create recommendation strategies – or what we refer to as "recipes" – by using one or more algorithms, adjusting weighting parameters of the algorithms, adding filters (inclusions and exclusions), boosters and variations. Within a few minutes, anyone – even those without a master's degree in computer science – can build a recipe that can deliver a unique experience to each and every website visitor. Pretty cool, right?

Testing Your Recipes

Once you've created a recipe, however, how do you know if it will actually work the way it's intended? Well, we've actually thought of a solution for that too. From its launch more than three years ago, Evergage Recommend has included the ability for business users to preview what experience a recipe would show to a specific individual based on all it knows about him or her — before publishing the recipe. In other words, Evergage allows a marketer to QA a recipe before publishing it on a website, in an email campaign or a mobile app.

Here's what this has historically looked like in the Evergage platform:

preview machine learning recommendations

Improving the Testing Experience

While incredibly useful and powerful, the preview capability had some limitations. A business user would have to find a specific shopper or visitor to test, for instance, and only the results for one shopper/visitor at time would be returned. As such, ensuring that the recipe would deliver an experience that made sense for multiple visitors was a highly manual process.

The good news is that we have completely overhauled the recipe preview capabilities within the Evergage platform. Now, instead of having to remember a specific shopper or visitor (though that option is still available), business users can find select groups of people by filtering based on certain parameters. For example, a retailer can create an audience based on visitors who have been active in the past six months, who have made a previous purchase and who have a high lifetime value. Or, rather than creating a new audience, a business user can select any existing segment.

Once the audience or segment has been selected, an Evergage end user can simply select the recipe they want to preview and, if applicable, an anchor item. Upon doing so, the updated preview screen will display a list of shoppers and the recommendations they would see based on the recipe and anchor item (i.e., product purchased, viewed, etc.).

preview machine learning recommendations

What's more, within the preview screen, business users can find additional details by hovering over a specific product or a shopper's affinity details. You can even quickly swap out which shoppers appear in the preview screen.

preview machine learning recommendations

This screen works for companies focused on content recommendations too:

preview machine learning recommendations

Final Thoughts

Whether you're a computer scientist familiar with machine learning, or a marketer who's never used machine learning before, Evergage offers a powerful solution for delivering great customer experiences. But the best part is that Evergage’s transparent approach enables incredible oversight to ensure the recipe you've built is designed to achieve the desired business outcomes.

Share This
testingpage2