Hosted by AWS superfan and CTO of ESW Capital, Rahul Subramaniam, the AWS Insiders podcast is a deep dive into all things AWS. Each episode delivers an insightful interview with one of Amazon’s leading product managers, and provides the best practices and insider secrets you need to keep up with AWS’s lightning pace.
In episode one, Rahul interviewed director and general manager for AWS AI, Ankur Mehrota, who provided an in-depth look at Amazon Personalize, a fully-managed machine learning service that delivers highly-customized recommendations to customers across different industries. You can listen to the full podcast episode now, or read on for an overview of this lively discussion.
What is AWS Personalize, and what problem does it solve?
Over recent years, businesses have been engaging with users more and more through digital channels – this was accelerated by the pandemic. Every digital touch point provides businesses an opportunity to create a tailored experience for users. This is where Amazon Personalize comes in. Delivering recommendations powered by machine learning, it helps businesses build high-performing systems that create more personalized experiences for their users.
As consumer habits shift, so does the way they engage with different businesses. This is particularly true in the retail world, but the demand for deeply personal user experiences has grown across a wide range of industries. For example, the meditation app, Calm, uses AWS Personalize to create mindfulness exercise recommendations to their users. By presenting exercises that align with their users’ individual needs and preferences, Calm has positively transformed its customer experience. Personalize is a powerful CX transformation tool that all customer-centric businesses should consider.
Getting started with AWS Personalize
AWS Personalize operates using these fundamental data sets:
- Interaction event data – the interaction between your users and your items, which can be products or media content in a catalog
- Item metadata – information such as product categories or brands, which Personalize can use to establish trends between items (e.g. different horror films or similar shoe styles)
- User metadata – information such as location, age group, and preferences
To start, you add these three types of datasets to Amazon Personalize, either through a set of APIs or by moving this data through an Amazon S3 bucket. Once you’ve done that, you can use a set of APIs to then select your use case or the type of recommendations that you’re trying to generate.
AWS also recently launched intelligent user segmentation, which allows you to specify an item, category, or genre and generate a segment of users interested in that item or genre. This can then be used for marketing campaigns such as emails or direct ads.
Once you have confirmed your desired recommendation type, Personalize runs in the background, using machine learning and training machine learning models to create recommendations that optimize for these use cases.
Improving recommendation quality
When commencing a personalization or recommendation system project, the key is to start out small by building a proof of concept. One problem that companies face when first using AWS Personalize is that they begin with a very small amount of data. This creates a challenge, because unless you have a sufficient data selection that accurately represents your overall data, the quality of recommendations will be much lower than if you used a larger data set.
For example if you are an eCommerce business, rather than picking data for just one or two categories, make sure your data set is representative of your entire catalog so that your recommendations have a higher level of accuracy.
Avoiding cost overruns
One of the benefits of Amazon Personalize is that it can scale automatically as your traffic increases or decreases, allowing you to provision the right amount of resources to serve your recommendations. In addition, this AWS service also allows you to reserve a certain number of recommendations per second – particularly useful when you expect a spike in traffic. To optimize your spending, use the auto scaling features so your traffic levels are consistent and predictable, and reserve your provisioning when you expect higher traffic to avoid increased costs. Setting up alarms and alerts is another great way to stay on top of unpredictable traffic and avoid expenses caused by continuous autoscaling.
AWS Personalize success story – Discovery+
When Discovery launched their first direct consumer streaming service, Discovery+, they had a very tight timeline to build their service from end to end and launch it. They chose Amazon Personalize to power their recommendations, and went from absolutely nothing to integrating video recommendations across their system and running Discovery+ for prime time at scale in only a few of months.
This is only an overview of the full podcast interview between Rahul Subramaniam and Ankur Mehrota, designed to introduce you to what is possible with AWS Personalize. Listen to their full interview here and discover how you can transform your own customer experiences with AWS Personalize.