Devoteam Revolve Helps Botify Industrialize its Data Science Activities with Amazon SageMaker Pipelines

Botify, a New York-headquartered search engine optimization (SEO) specialty company founded in 2012, wanted to scale up its data science activities. Working with Devoteam Revolve, an AWS Premier Consulting Partner, Botify developed a solution based on Amazon SageMaker Pipelines, which significantly reduced the development and production time of its models, resulting in greater agility, more innovation, and the ability to grow on all fronts: more products, more customers, and more employees.

Botify is a SaaS software company that enables brands to increase their online traffic and sales. Botify uses machine learning technologies to offer an automated SEO analysis, decision support, and implementation solution to its customers to help them realize their growth potential. “In 2020, we launched Botify Intelligence, which uses data science and model building to improve our products and services,” says Yanal Wazaefi, head of data science at Botify.  

In order to accelerate this activity, which was already running on Amazon Web Services (AWS), Botify decided to launch an MLOps project to reduce time to market for its models, to improve scalability, and to allow data scientists to develop and put their models directly into production. Previously, they had to pass the work to a separate production team, slowing deployment.

This acceleration was driven by market forces. “The web moves a lot, and we have to launch models frequently,” says Wazaefi. “We needed to be able to constantly evolve our models and compare them.” Additionally, Botify needed a way to manage the volume of data it was generating—the analysis of its clients' pages, according to several thousand factors and dimensions, generated more than 2 petabytes of data.  

Industrializing MLOps activities required Botify to innovate. To do this, it worked with Devoteam Revolve, an AWS Premier Consulting Partner. The deliverable is not a model, but the tool to generate models. “Getting the model to run, with deployment and monitoring constraints, and provide insights produced on a daily basis was made possible by using AWS,” says Sylvain Graceffa, data and analytics practice leader at Devoteam Revolve. “The cloud is an enormous lever to help solve these problems. And we have benefited from the power of innovation of AWS.”

To continue to innovate, Botify looked at Amazon SageMaker Pipelines—the first purpose-built, easy-to-use continuous integration and continuous delivery (CI/CD) service for machine learning (ML)—which piqued its interest. “It offered everything we wanted to build our ML pipelines, including tracking,” says Wazaefi. This meant Botify could keep what it had developed and stay in a familiar environment, but still add new capabilities.  

After consulting with AWS experts to confirm that SageMaker Pipelines could meet its expectations, Botify tasked Devoteam Revolve to lead the implementation of SageMaker Pipelines into Botify’s workflow. After a few weeks of consultation, the project was implemented in 2 months. The technical teams began by configuring the working environment of the data scientists using Amazon SageMaker Studio—the first fully integrated development environment (IDE) for machine learning—to prototype in collaborative notebooks and to monitor and evaluate machine learning experiments conducted using the Amazon SageMaker Experiments functionality.  

The result of this activity was an easy integration of the developed models into the production environment. The integration of the various Botify systems allowed the ML pipelines to be defined and activated from the compute and analysis workflows built with Amazon Simple Workflow Service (Amazon SWF).

Once the implementation was up and running, Botify quickly saw results. “The improvements were beyond our expectations. In less than 6 months, we were able to put as many models into production as we had in the previous 2 years,” says Wazaefi. “It takes us 1-2 weeks to prepare the pipelines. We are putting more models into production, faster, and we can fix bugs quickly. Fixing a problem used to take weeks. Now it's done with a ticket in a few hours.” 

The success of the project allowed Botify to accomplish another goal—tripling the size of the data science team thanks to how easy the system is. “Using the Pipeline Sagemaker is quite simple, even for someone who is not well-versed in development,” says Wazaefi.

Previously, recruiting data scientists with the level of experience needed was difficult. Now Botify’s data scientists have the power to put their models into production simply and quickly, which has made recruitment easier. “The fact that we use a framework gives a lot of reassurance, and data scientists can concentrate on the research part,” says Wazaefi. “I can now recruit the best researchers rather than the best coders.”  

And that's not all. In the past, prototypes used real customer data, but on a small scale. Integrating the new model took months to ensure that the code worked for everyone. “Today, the experimental environment that has been developed with Devoteam Revolve allows the new solution to be activated for all customers in the ML Pipeline,” explains Wazaefi. “And before production is launched, we have the opportunity to see how it works across all scales and customers.”

This gain in speed and fluidity directly impacts Botify's offerings. Showing the new models developed by the data scientists to the product teams is easy. And the latter, knowing that activating the model for their client is just as easy, can go to them with a demonstration in hand and propose these new smart features that call on ML—sending SEO alerts, simulating various modification and seeing the effects, or estimating the potential of an SEO recommendation for a site. “Today, we are the only ones able to launch models on this scale and at this rate,” says Wazaefi. “We can go further and faster on our products. We deliver stronger results at scale and at speed. And that saves our customers time and money.” 

Thanks to the new talents recruited, Botify will continue to develop its solution, but not alone. “The AWS teams that developed SageMaker Pipelines want to exchange with Botify,” says Graceffa. “The work we’ve done and what’s been learned working with customers can help the product evolve. It’s a win for everyone.”