Fiverr Relies on SageMaker to Streamline and Simplify Machine Learning Models

When it came time for Fiverr to analyze the data around user behavior on its platform, the freelance services website did what many modern businesses do in terms of recruiting the right tool for the task: they engaged Amazon Web Services’s SageMaker. Amazon SageMaker is an API that enables developers and data scientists to readily build, train, and deploy machine learning models to serve any need and at any scale.

“This was a very important milestone for us because, moving forward, we want this to be the standard in the data science deployment and all our predictive modeling,” says Eran Abikhzer-Agam, head of data at Fiverr. “We want our data scientists to be completely independent as much as possible, and SageMaker allowed us to do that.”

Like any online business, Fiverr was interested in streamlining the user experience on its platform, tracking user behavior, and predicting when a customer might need a bit of help. “We wanted to be able to monitor and intervene at specific points where we felt the user is either lost, not finding what he/she needs, and needs a bit of assistance,” Abikhzer-Agam says. “So the business aspect was to minimize the number of users we are losing due to this fact.”

Building and training machine learning models to analyze a data array can be a complex and arduous task. And even with a fully-trained model, there are still the sizable tasks of integrating it with your applications and optimizing performance, which puts a great amount of strain on developers, not to mention computing and storage. SageMaker greatly streamlines and simplifies this process.

“One of the most impressive things we saw with SageMaker is a mechanism for exploring multiple variations of hyperparameters for models,” Yuval Ben-Zion, data science team leader at Fiverr, says about utilizing the platform. “The ability to compare between variations in a rather approachable way made it easy to decide which one we’re going to continue.”

The end result for Fiverr was that SageMaker allowed them to minimize efforts in building and placing models, quickly run experiments, and deliver better service to their users. “With SageMaker, we are able to run real-time predictions based on users’ behavior on our website and trigger product interventions to try and help them find what they are looking for,” Ben-Zion says. “I think this is the milestone that we wanted to achieve.”

Ben-Zion adds that they believe this is the right way to make a real-time impact and provide a better experience for their customers overall. “To help […] make it easier for them and get the services that they need, this is our goal.”