Customer Stories / Software & Internet

Achieving Near-Zero Downtime and Powering Generative AI Using Amazon EKS with Ada
Learn how Ada boosted compute efficiency by 30 percent, cut costs by 15 percent, and runs AI workloads using Amazon EKS.
15% reduction
in compute costs
30% increase
in compute efficiency
20% increase
in cost efficiency of GPU usage
70% increase
in deployment velocity
5 days
to upgrade instead of up to 5 months
Overview
Software company Ada Support Inc. (Ada) specializes in artificial intelligence (AI) and manages seven clusters—with up to 700 nodes each—to support its solution for customer service automation. Ada wanted to enhance the customer experience by improving operational efficiency and reducing downtime, so it decided to migrate its self-managed clusters to a fully managed service from Amazon Web Services (AWS).
To run highly scalable, reliable, and secure Kubernetes environments, Ada used Amazon Elastic Kubernetes Service (Amazon EKS)—a managed service for starting, running, and scaling Kubernetes. Thus, the company reduces upgrade time, saves on costs, and empowers its engineering team to innovate.
%20(1).6441e944c5bd993513d958870e8437ff88253b8e.jpg)
Opportunity | Using Amazon EKS to Reduce Maintenance and Customer Disruption for Ada
Ada envisions a world where almost every customer interaction is resolved by AI so that businesses can automatically resolve more customer service conversations across channels and languages with less effort. Since 2016, Ada has powered more than 4 billion automated customer interactions.
Ada’s AI-powered solution has a reasoning engine that uses machine learning (ML) models and large language model calls to understand customer inquiries, assess appropriate responses, and resolve inquiries automatically. Given the complexity and size of Ada’s clusters, two or three full-time engineers used to take up to 5 months to upgrade Kubernetes versions. The company adheres to a service-level agreement with availability commitment of 99.9 percent, and version upgrades accounted for up to 30 percent of the yearly error budget.
Ada was committed to migrating with near-zero downtime to minimize the impact on customers. After weighing the options, it chose AWS Global Accelerator—a networking service that organizations use to improve the availability, performance, and security of public applications—because Ada’s engineers could incrementally route traffic to clusters and quickly roll it back if needed.

Using Amazon EKS, we have much more time to develop and improve product capabilities to deliver a high-quality customer experience.”
Mike Gozzo
Chief Product and Technology Officer, Ada Support Inc.
Solution | Increasing Deployment Velocity by 70 Percent While Reducing Costs by 15 Percent
In 4 months, with near-zero downtime, Ada completed the migration and several major version upgrades of Kubernetes in 2022. The vast majority of Ada’s workloads, ranging from generic compute to ML workloads, run on Amazon EKS. Because of the huge number of API changes in a multiple-version upgrade, Ada used blue/green deployments—a deployment strategy that creates two separate, identical environments—so that engineers could test changes in isolation. “Using AWS Global Accelerator, we could dial traffic incrementally from 0 to 100 percent with very good granularity,” says Kosta Djukic, principal engineer of cloud infrastructure, Ada. “We could canary-test the new cluster and gain confidence as we added more traffic to Amazon EKS.”
Amazon EKS automatically manages the availability and scalability of the Kubernetes control-plane nodes responsible for scheduling containers, managing application availability, storing cluster data, and other key tasks. By applying the blue/green strategy using Amazon EKS, Ada can upgrade its clusters in 5 days instead of up to 5 months. “With a larger error budget devoted to engineering, we can take more risks across our infrastructure and processes to improve our product,” says Djukic.
Ada Architecture Diagram

Ada saves time using managed services from AWS, so its engineering team has more time to optimize architecture and deployment pipelines. For example, Ada set up application sets from Argo CD—an open-source, continuous-delivery tool for Kubernetes—to keep copies of Ada’s applications running across different clusters in sync and efficiently manage differences when necessary. In the year and a half after the migration, Ada increased the deployment velocity by 70 percent. “Because the migration to Amazon EKS went smoothly, we had time to improve our automated processes and workflows,” says Djukic.
Ada now also slices the GPUs that are used by its solution’s reasoning engine to perform inference. Ada deploys its ML models as containers in Kubernetes environments. These models need to work together and make API calls across the infrastructure, but some models are too small to require an entire GPU system. Ada increases the cost efficiency of GPU usage by an estimated 20 percent across environments using GPU slicing, which would have been too time consuming and complex to implement with self-managed clusters.
Ada’s customers benefit from less downtime and more innovation. “As an infrastructure team, we can focus on improving engineering velocity and helping engineers deliver features quickly rather than being tied up with operational toil,” says Djukic. Since migrating to Amazon EKS, the company has had more time to devote to its cutting-edge generative AI agent, which is less rigid and requires less customer development than a declarative chatbot. Ada’s benchmark automated resolution rates increased from 20–30 percent with declarative agents before the migration to up to 77 percent using generative AI.
With new features, Ada has grown its overall compute footprint by 30 percent since the migration. Yet, the company reduced compute costs by 15 percent over the same period. With more resources for optimization projects, Ada implemented a scaling controller that effectively uses Amazon Elastic Compute Cloud (Amazon EC2) Spot Instances, which take advantage of unused Amazon EC2 capacity and are available at up to a 90 percent discount compared to Amazon EC2 On-Demand Pricing. Ada has also started running some of its infrastructure on AWS Graviton processors, a family of processors designed to deliver excellent price performance for cloud workloads running on Amazon EC2. To further reduce costs, Ada is considering migrating additional workloads to AWS Graviton processors.
Outcome | Investing in Generative AI and Expansion Using AWS Services
Ada plans to continue investing in innovative generative AI features and is considering expanding into other regions, such as Asia Pacific, because of its increased operational efficiency. “Using Amazon EKS, we have much more time to develop and improve product capabilities to deliver a high-quality customer experience,” says Mike Gozzo, chief product and technology officer at Ada.
About Ada Support Inc.
Ada is an AI-native company that provides solutions for automating customer service to make it extraordinary. Since 2016, the company has powered more than 4 billion automated customer interactions for several brands.
AWS Services Used
Amazon EKS
Amazon Elastic Kubernetes Service (Amazon EKS) is a managed Kubernetes service to run Kubernetes in the AWS cloud and on-premises data centers.
AWS Global Accelerator
AWS Global Accelerator is a service that improves the availability and performance of your applications with local or global users. It provides static IP addresses that act as a fixed entry point to your application endpoints in a single or multiple AWS Regions, such as your Application Load Balancers, Network Load Balancers or Amazon EC2 instances.
Amazon EC2 Spot Instances
Amazon EC2 Spot Instances let you take advantage of unused EC2 capacity in the AWS cloud.
AWS Graviton processor
AWS Graviton processors are designed by AWS to deliver the best price performance for your cloud workloads running in Amazon EC2.
Get Started
Organizations of all sizes across all industries are transforming their businesses and delivering on their missions every day using AWS. Contact our experts and start your own AWS journey today.