Envoy Media Uses Amazon SageMaker to Increase Machine Learning

Envoy Media needed a machine learning platform to optimize the quality and quantity of the leads it generates. Envoy wanted to predict the lifetime value of a potential customer to net a higher return on the advertising dollars it invested. With Amazon SageMaker, the company improved its machine learning prediction accuracy from 85 percent to 98 percent.

Envoy works with a select group of financial services companies to help consumers with their personal finance needs. The industries it operates in are heavily regulated, and various financial products are tailored to highly specific needs. Envoy uses sophisticated algorithms to accurately match potential customers with a company that will meet their needs.

The company relied on advertising networks' algorithms to find people interested in filling out a form to request more information. Yet, as it analyzed the results, it recognized that optimizing for this event would not accurately reflect long-term value for its clients. The advertising networks would optimize to get forms filled out but failed to target true long-term value generated from those forms.

That made it incredibly difficult for Envoy to tailor its advertising spend to lifetime value. It needed a way to predict the long-term value of each of the leads it generated and send those predictions to the advertising networks. This prediction model also needed to align with Envoy’s best-in-class DevOps practices and seamlessly integrate with its existing datasets.

“With AI bidding, you want the quality of your predictions to be as accurate as possible and the quantity to be as high as it can be,” says Michael Taggart, Co-founder and Chief Technology Officer at Envoy Media. “By figuring out higher quality leads and a greater quantity of them, we could leverage the automated bidding algorithms to our advantage.”

While Envoy had built machine learning models with other platforms, it quickly realized the value that Amazon SageMaker could bring in improving the accuracy and lowering the operational overhead of deploying those models. To take advantage of all that’s possible in SageMaker, Envoy turned to Toptal, which boasts a readily available, on-demand network of AWS certified developers, engineers, architects, and consultants.

As an AWS Advanced Consulting Partner, Toptal matches AWS customers with pre-vetted experts to power its teams on an hourly, part-time, or full-time basis. Customers can work with experts entirely on demand, paying only for what they use, and scaling up or down as needed, just like they can with AWS. Toptal also provides customers with entire project teams that turn concepts into reality by building cloud-native applications, from ideation to production. As an AWS Service Delivery Partner and AWS Well-Architected Partner, Toptal specializes in machine learning, data science, and DevOps projects for AWS customers.

Toptal’s machine learning experts worked with Envoy to automate a data model that created a new framework for Envoy’s marketing department. Envoy chose SageMaker because it allows users to independently train, test, and deploy machine learning models.

Toptal incorporated the clean data Envoy collected into SageMaker, which married up with the information already stored in the AWS stack. Using SageMaker enabled the team to fine-tune the parameters of each machine learning model it created. Those slight alterations allow Envoy to bypass cumbersome and time-consuming steps. Instead, it can bring generalized data into SageMaker and then tune the model, deploy it, and monitor its performance in minutes. SageMaker gave Toptal and Envoy all the tools to automate the process while maintaining full control in case theyneeded to change any part of it. Using multiple models trained with and deployed on Amazon SageMaker, Envoy dramatically narrowed the gap between its predictions and reality, increasing the accuracy from 85 percent to 98 percent with its largest advertising network.

“By using machine learning in Amazon SageMaker, we were able to see how the leads we generated would pan out in one week, one month, and one year from now,” says Taggart. “That’s unbelievably valuable data in our sphere. Right now, machine learning bidding is a luxury for a digital marketing firm. In two or three years, it will be a necessity.”

The reliability and performance of integrating the AWS Cloud data made Amazon SageMaker’s response times extremely fast.

“Amazon SageMaker allows us to focus on core machine learning tasks and business value,” says Andranik Khachatryan, an AWS Certified Machine Learning Specialist from Toptal’s DevOps practice, who steered Envoy’s SageMaker project. “It takes care of hyperparameter search, scaling/load balancing, logging, monitoring, debugging, version control to name a few. This is lots of heavy lifting we don’t have to do.”

By leveraging the seamless integration of Amazon SageMaker with existing datasets, Envoy is able to deploy new models into production within minutes.

Amazon SageMaker enabled Envoy to obtain immediate feedback on various parameters and different models for lead generation. Taggart explains that Envoy is able to analyze the efficacy of multiple models in minutes when it had required weeks and months before deploying Amazon SageMaker.

“Amazon SageMaker made everything more cost-efficient, instantly more reliable and optimized for machine learning deployments,” says Taggart. “It hit all those buttons and had the versatility that let us tailor it for our specific needs.”

Additionally, the machine learning capabilities make the communication patterns a win-win for Envoy and for the companies it buys ads from.

"With this design, our ML is able to send strong signals to advertising networks' ML which helps us optimize both front-end conversion and back-end value," says Ryan Marlow, Envoy's Director of Technology. "We can test and deploy many different models for different needs to create a healthier AI ecosystem."