Using Unleash to experiment between AI models, ASAPP increased delivery velocity and improved key production KPIs by over 60 percent.
Unleash enabled ASAPP to run production experiments between AI models, reducing edit rates by roughly two-thirds while maintaining customer experience.
ASAPP uses Unleash to safely compare models, prompts, and routing logic in production without hard-coded decisions or risky rollouts.
By standardizing feature flags and experiments, ASAPP increased delivery velocity while keeping releases controlled and reversible.
“Full-stack experimentation isn’t just about tracking KPIs like agent augmentation rate or handle time. That’s important, but it’s not enough to drive behavior change alone — we also need to ensure the overall system behaves consistently in terms of latency, errors, and cost. Unleash makes this possible by integrating experiments directly into feature flags, and with its built-in schema for recording the lifetime of events, we can push everything downstream to our data warehouse. That way, every experiment ties business metrics to system performance, so we know with confidence when a model is truly ready to release.”
Tomas Piantanida
Lead Machine Learning Engineer, ASAPP
ASAPP builds GenAI native customer service solutions for enterprise call centers, spanning both chat and voice channels. Serving enterprise customers like Fortune 500 companies and global Airlines, ASAPP maintains complex operating environments, including single-tenant and multi-tenant deployment models.
Running on AWS, ASAPP’s platform engineering team bridges development and infrastructure to make building and shipping features fast, easy, and reliable across a microservice architecture.
Unleash is a core part of ASAPP’s delivery and experimentation toolbox, enabling controlled releases, A/B testing, and safe promotion of model changes in production.
ASAPP provides GenAI-native solutions to Fortune 500 companies, requiring a delicate balance between rapid model innovation and enterprise-grade stability. As they scaled, several hurdles emerged:
ASAPP adopted Unleash early, evolving it from a simple utility into a foundational platform capability. By making Unleash a core component of their “Golden Path,” ASAPP enabled their engineers to focus on AI innovation rather than deployment risks.
Instead of risky all-at-once releases, teams use feature flags for progressive rollouts, starting with a small percentage of traffic and scaling up as confidence grows. This safety net allows ASAPP to test new features in production with real users, knowing they can instantly revert to a stable state if performance degrades, without a full code redeploy. Unleash provides ASAPP with the “kill switch” necessary for high-stakes AI deployments.
ASAPP is constantly testing different GenAI models to determine the impact of AI on “True North” KPIs—like Agent Augmentation Rate—while monitoring backend health metrics like latency and error rates in a single, unified view. Unleash’s ability to correlate a feature flag with any metrics across the front or back end, enables ASAPP to ship feature that truly impact users in a positive way.
To scale efficiently, ASAPP’s platform team developed base frameworks—internal libraries for Go, Python, and TypeScript that provide a “Golden Path” for all microservices. These frameworks act as the connective tissue between development and infrastructure, with Unleash serving as the key enabler for this architecture.
By embedding Unleash directly into these frameworks, ASAPP eliminates manual configuration and ensures architectural consistency across every service. These libraries automatically inject standardized context fields like Tenant ID and Agent ID, ensuring a “sticky” experience where a user receives the same AI model variant across both the frontend UI and backend inference services.
ASAPP chose AWS to provide the resilient, enterprise-grade foundation required for GenAI at scale. By offloading infrastructure management to AWS, ASAPP redirected their engineering focus from “handcrafting” clusters to refining their flagship AI models.
ASAPP migrated to Amazon EKS to eliminate the burden of managing Kubernetes control planes. This move, paired with Amazon RDS for PostgreSQL to host Unleash data, replaced manual maintenance with automated, scalable services. EKS provides the flexibility needed to run both production and PCI-compliant clusters, ensuring that ASAPP’s voice and chat applications meet the highest availability and regulatory standards.
Security is non-negotiable for ASAPP’s Fortune 500 clients. Using AWS PrivateLink, ASAPP established secure, point-to-point networking between their distributed EKS clusters and the Unleash service. This ensures that sensitive data never traverses the public internet. Amazon Route 53 governs internal DNS, while AWS KMS automates “at-rest” encryption across the entire stack, providing a robust security posture that is easy to manage across multiple zones.
The integration with AWS Bedrock allows ASAPP to serve cutting-edge models like Anthropic and Mistral with high reliability. AWS provides the raw power and variety of LLMs, while the underlying infrastructure allows ASAPP to manage quota and reliability effectively. By running on AWS, ASAPP can shape traffic and manage model failover in real-time, ensuring their generative agents remain responsive even during provider-side volatility.
“EKS is worth the money to not have to manage your own cluster. We use the full AWS connectivity stack—PrivateLink, Route 53, and KMS—to follow best practices by default across our entire architecture.” — Ivan Lee, Principal Platform Engineer, ASAPP
Every feature flag change is automatically captured, traceable, and aligned with governance policies—removing the need for manual tickets or redundant workflows.
Unleash’s flexible deployment and pricing model allowed Prudential to scale feature management consistently across teams and technologies—supporting internal chargeback models and enabling shared infrastructure growth without operational overhead.
As AI-assisted development accelerates, Prudential relies on Unleash to safely manage experimentation, rollouts, and risk. Feature flags now serve as a control layer that ensures new AI-generated or AI-influenced code is introduced responsibly.
By integrating Unleash into their core workflow, ASAPP transformed feature management from a manual risk into a strategic asset.
Because Unleash is decoupled from code deployments, it serves as ASAPP’s operational safety valve. If an LLM served through AWS Bedrock begins to experience high error rates or latency, the team uses Unleash to instantly “kill” the failing model or shift traffic to a stable provider like Anthropic or Mistral. This ensures that ASAPP can maintain 24/7 reliability for their airline and enterprise customers without the need for an emergency code fix or redeploy.
ASAPP uses Unleash to give developers total control over their features without requiring constant help from the platform team. By using Unleash’s APIs, the platform team built a workflow where a developer can bootstrap a new flag via a configuration repository and then personally manage the rollout. This allows a developer to log into the Unleash dashboard and decide to scale a feature from 20% to 100% on their own timeline. This “self-service” model provides a perfect balance: the platform team ensures the infrastructure is secure on Amazon EKS, while individual developers have the agency to experiment and ship at their own pace.
The partnership between Unleash and AWS has transformed ASAPP’s ability to deliver data-driven AI solutions, turning experimentation into a competitive advantage.
Hosted, or self-hosted: it’s your call. It’s quick and easy to set up. Get started in 2 steps with the functionality you need to improve your software development workflow.
“Change control management is huge for us. Flipping a flag in production is considered a change that requires auditing, and as an enterprise, we have standardized on ServiceNow. But as a developer, I don’t like filling out tickets. With Unleash, developers never have to see ServiceNow. They simply interact with Unleash, and all changes and approvals are automatically synced in the background. Our auditors are happy, and our developers are more efficient. This is a win-win.”
Peter Ho
VP, DevOps Tech Lead, Prudential
