Amazon Rapidly Develops Toy Subscription Service
Amazon's well-known customer focus is more than an idea. It’s the company’s DNA, driving the development of new services from inspiration to execution. It was behind the creation of the company’s successful toy subscription service with a twist, which provides play experiences focused on science, technology, engineering, and math (STEM).
Jenn Lin, principal engineer at Amazon, describes it, starting with the customer being the path to discovery. “We have an annual review of potential business initiatives where we work backwards from customer problems. Our research showed that parents have a hard time finding high-quality STEM toys, so we decided to build a highly personalized delivery service to meet the need.”
The customer relevance also got the engineering team fired up and ready to execute fast, which ultimately led it to use Amazon Web Services (AWS). “Some of us have kids in school and were really inspired by the potential impact of the service,” Lin recalls. “Using AWS, we were able to get the solution live within seven months—and build a framework that has since been used by more than 20 additional programs.”
“AWS Lambda reduces the time we have to spend on operations so we can focus on improving the customer experience.”
— Jenn Lin, Principal Engineer, Amazon
AWS Services Used
Amazon is guided by four principles: customer obsession rather than competitor focus, passion for invention, commitment to operational excellence, and long-term thinking. Customer reviews, 1-Click shopping, personalized recommendations, Prime, Fulfillment by Amazon, AWS, Kindle Direct Publishing, Kindle, Fire tablets, Fire TV, Amazon Echo, and Alexa are some of the products and services pioneered by Amazon
- Earned millions in revenue
- Eliminated operations burden
- Scaled to serve more than 20 services
AWS Services Used
Intelligence on Call
This was not going to be a standard subscription service. While Amazon had offered item-specific subscriptions for many years, it had not yet created one that customized the selection to customers’ tastes. “This was a challenge that people at Amazon had talked about solving for many years: automatically choosing personalized items from a given category,” says Lin. “Creating it meant connecting product ratings, machine learning about customer preferences, fulfillment, and more.”
However, the services to discover and select products for customers on a recurring basis simply didn’t exist. The team had to build it.
The first prototype was not native to AWS. When it failed to deliver the low latency and straightforward implementation the team sought, the service was rebuilt on the AWS Cloud, using AWS serverless, database, and API capabilities.
The selection aspect of the service was built from scratch. “It does the work of choosing the toys by analyzing a wide range of data, including reviews, purchase history, and even whether the item has won awards,” says Lin. “As we evolve our data model, we can provide something more and more specialized to each customer.”
A Service Built on Serverless
Actions taken at the front end of the service begin their journey at Amazon API Gateway. AWS Lambda serverless functions normalize incoming data and send it to the Amazon DynamoDB database service. The database is also used to store configuration data.
When a customer is due a shipment, an internal workflow management tool calls the service through API Gateway and gets an optimized item in return. The workflow also triggers fulfillment using a second AWS Lambda service.
As usual, customers are at the center, with the ability to change their selection or skip a delivery at their discretion. This data feeds back into the machine-learning model to provide improved fulfillment of customers’ preferences.
The Power of Minimum Ops
While the team was excited about building an impactful solution, they were not interested in running a fleet of servers. That made AWS Lambda an ideal choice. “AWS Lambda reduces the time we have to spend on operations, so we can focus on improving the customer experience,” says Lin.
The instant scalability of serverless computing was a good fit for the complex usage patterns of a subscription service. “If there’s a lot of publicity around an item, we will get a lot of sign-ups. Sometimes the service is being used very little. In other words, we have both more idle and spikier traffic than average. With AWS Lambda, the service runs only when a customer needs it, which reduces costs, and then scales up very quickly whenever needed.”
The Amazon DynamoDB database service provides similar levels of operational ease. Lin says, “Amazon DynamoDB scales appropriately even to Amazon retail scale, and I recommend it to most of the teams I work with at Amazon as a principal engineer. It was also cost-effective, even compared to alternatives inside the company.”
The Proof Is in the Product
The solution has delivered on high expectations. “We’ve generated millions in revenue from the STEM subscription service alone, proving its value to customers, all with a system that features a low cost of ownership,” says Lin. “But the extensibility of the model has been even more transformative.”
The architecture has scaled without major changes to the underlying systems. “We have made only small optimizations to the configurations of Amazon API Gateway, AWS Lambda, or Amazon DynamoDB—in other words, it’s pretty much set and forget. This means the team can focus on the science behind the data model to improve predictions of customer needs.”
The solution has seamlessly handled high-velocity days—for example, when the collectibles subscription launched. “We were operationally prepared for any issues and closely monitored things,” recalls Lin. “It did just what it was supposed to do, even under high loads.”
By continually measuring the performance of the solution, Amazon can ensure it is delivering enhanced experiences for customers. “We look at metrics such as star ratings for the subscription, star ratings of the items selected, and renewal rates to analyze how the service is doing. It all feeds back into the machine learning model to improve those things over time.”
Even with all the technical sophistication, the team never loses sight of who all this work is for—the subscribers. “We send out surveys and get quantitative and qualitative feedback, which we use to prioritize features,” says Lin. “For example, the first week we heard customers requesting the ability to ship outside the US, and the team was able to implement their request within two weeks.” Using flexible, scalable AWS technology, teams like Lin’s can create customer experience innovation that wins long-term loyalty.