How Smartocto Deployed Their Editorial Analytics Solution in 3 Months Using Amazon SageMaker

Smartocto is a smart editorial analytics system that brings actionable data to newsrooms, storytellers and online publishers. They provide content analytics to 350 newsrooms and media companies around the world through its smartocto system, which features both near-real-time and predictive data features. In order to provide the analytical capabilities smartocto used a combination of open-source products and AWS services, including Amazon Elastic Compute Cloud (Amazon EC2), which provides secure and resizable compute capacity for virtually any workload.

When running their workload on resizable compute instances smartocto was facing a number of challenges. Their architecture made it difficult to onboard new customers quickly and deploy new Machine Learning (ML) models. This was because the teams had to build a new secure environment for each of its customers. To complete the onboarding process several manual tasks had to be executed, which is both time-consuming and prone to human error. Smartocto started to look for an ML solution that would help improve this and lower the compute costs, reduce the amount of time spend on managing their infrastructure. The solution would allow them to free their teams to focus on the tasks that mattered, such as fine-tuning the accuracy of their algorithms.

Technical details of solution

Smartocto worked alongside the Amazon Web Services (AWS) team to build a proof of concept and test Amazon SageMaker. Amazon SageMaker gives customers large and small in every Industry the ability to build, train, and deploy ML models for virtually any use case with fully managed infrastructure, tools, and workflows. Training an ML model has different requirements than using the model to perform inference. Amazon SageMaker helped smartocto decouple different parts of their ML workload to right size the compute used for different steps. This is done by running more compute heavy tasks, such a model training, on different compute instances than lighter tasks, such as model inference.

Being able to pick the right tool for the right job enabled smartocto to achieve a 10x lower resource usage per ML model, since they do not over-provision and scale according to their needs, to meet customer demand. Here, SageMaker takes away the undifferentiated heavy lifting, this enables smartocto to focus on the tasks that matter to them.

After the proof of concept smartocto started to migrate several of its existing ML models to the cloud. Here they leveraged Amazon SageMaker endpoints to support predictions in near real time. Due to this new capability the company decided to develop Smartify, a predictive editorial analytics solution that uses ML to forecast the expected engagement, such as click rates, likes, and shares, of a news post on a particular channel.

Summary

It took smartocto less than 3 months to finish developing Smartify. They were able to quickly deploy the solution to production – an estimated 6 months ahead of schedule.  “It was amazing how fast we were able to release Smartify using Amazon SageMaker,” says Ilija Susa, cofounder and chief data officer at smartocto. Using Amazon SageMaker, smartocto has achieved a 10 times lower resource usage per ML model while delivering better predictions.

Besides improving their ML workflow smartocto also automated the process for onboarding new customers to Smartify. Previously, it could take up to a few weeks to onboard a new customer in their manual process. Now, the onboarding can be completed in a matter of days. “Adding new customers is much faster and simpler for us to do,” says Susa. “We can spend our time focusing on training our ML models for accuracy instead.”

If you want to learn more how you can spend your time on the tasks that add the most value you can take a look at other ML related case studies. For more information, please reach out to Felipe Chies, Senior Specialist AI/ML at AWS Benelux. We’re here to help you achieve and accelerate your goals, ML related or otherwise.

1200 600 Partner Content
My name is HAL 9000, how can I assist you?
This website uses cookies to ensure the best possible experience. By clicking accept, you agree to our use of cookies and similar technologies.
Privacy Policy