Every person in this world is unique. Each has his or her own preferences, behaviors and attributes. So if no two people are exactly the same, why treat them the same in your marketing? Consumers and business buyers are also demanding to be treated as the unique individuals they are. With so much to choose from on the internet, they are giving their time and attention to those companies that provide unique experiences and tailored, relevant content that cuts through the noise.
Machine-learning algorithms allow you to provide such individually relevant experiences in a scalable way. The most common way we see marketers leverage algorithms today is with product or content recommendations, but they can be used to dynamically sort lists or arrange search results, modify navigation, and much more.
Essentially, machine-learning algorithms intelligently determine the most relevant experience for each individual based on all the data available. Yet putting that decision solely in the hands of a machine can be understandably difficult for many marketers, as they are giving up the ability to decide what each person experiences. A first-class personalization platform includes the capacity to control the machine learning strategies that drive customer experiences. In this new 3-part video series, Karl Wirth, Evergage Co-founder & CEO provides an overview of machine-learning personalization, explains how to roll out such a solution, and discusses innovative ways to apply machine learning beyond product and content recommendations.
As a take-away from these videos, I’ll elaborate on three components you need to exercise control over your machine learning-driven experiences so you can ensure that they are the right ones for your customers.
Build Your Own Strategies
First and foremost, you need the ability to create your own strategies (aka “recipes”). Many personalization vendors will provide you with standard algorithms, typically for product or content recommendations, but not tell you what those algorithms contain or even explain how they work. As a result, you may feel that your recommendations are generated by a “black box”: You don’t know how the experiences are selected or if they are even relevant to each individual that sees them.
If you want to leverage machine learning to determine multiple aspects of your customer experience, this lack of insight is not acceptable. You cannot control your experiences if you don’t know how those experiences are selected.
To build your own machine learning strategy, you need to pick a base algorithm (or multiple algorithms), opt to exclude or include certain criteria to give you more control over the categories, brands, price points, keywords, etc. that are shown, and then add boosters based on the affinities of each individual to completely personalize his or her experience.
For example, a retailer may choose to display trending products on its homepage. To do this, it would select a “trending” base algorithm. Then it could opt to include only higher priced items, excluding items of lower prices. Finally, it could boost the brands and categories that each individual prefers. With this approach, the retailer can control the types of products that are displayed in that area while also ensuring that each recommendation is individually relevant to each visitor. For more details on how to build customizable algorithms, download our eBook, The Marketer’s Guide to Machine-Learning vs. Rule-Based Personalization.
Check Experiences for Individuals
After you have defined a machine learning strategy, you need to make sure that it will work in the way you expect. Let’s say you have created the strategy described above (higher priced trending items relevant to each person’s preferred categories and brands). How can you ensure that it is actually working before you publish it? How do you know that the recommendations will make sense for each person? You don’t want to put it on your site completely untested.
To gut check the strategy, you need to be able to see the recommendations it would deliver for actual individuals on your site. You should be able to pull up a profile of a given shopper with her established preferences to see which recommendations she would see if this algorithm were live on your homepage. Then you could compare those recommendations to those that other shoppers would see. Do those recommendations make sense given who each person is and what they both like? If not, you would continue tweaking the strategy until you were happy with the experiences it produced.
A/B Test Algorithms to Find Optimal Choice
Finally, you need to test the effectiveness of your machine learning strategies after you have published them. That requires the ability to A/B test a strategy against other strategies and/or a control experience to determine which option works best. You want to be able to confirm that the machine learning strategies have a real impact on the overall experience and your business metrics. And even if a strategy is working, can it be improved? Regular testing allows you to continue to iterate and optimize performance.
You can also use testing to determine whether certain strategies work better for different audiences. Maybe that trending strategy you built works really well for new visitors, but not as well for returning visitors. You can use that insight to continue to target that strategy to new visitors but create and test a new one for returning visitors.
Many personalization vendors do not let you build your own machine-learning strategies, test them for specific individuals before publishing, and run A/B tests to optimize their effectiveness. But when you rely on machine learning to deliver experiences across your site, app, email campaigns, etc., you don’t want to just hope they are delivering the best experience and having a positive impact on your metrics. You want to know for sure.
In these videos, Karl provides great insights on using machine learning to create a truly personalized one-to-one experience for your customers across channels. I’m sure you’ll enjoy them!