Finlex cuts cloud costs 50% and ships production AI with DoiT

As Finlex began exploring generative AI, the team quickly moved from curiosity to experimentation, building early prototypes of a chat-based “Finlex Assistant” and multi-agent workflows designed to support brokers and internal teams. While these early solutions demonstrated strong potential, scaling them into production introduced new complexity across performance, governance, and operational control.

As workloads grew beyond local testing environments, the team encountered challenges around latency, consistency, and observability. Operating in financial services also introduced strict requirements around auditability, data governance, and explainability, while increasing cloud usage made infrastructure spend harder to predict and optimize effectively.

At the same time, Finlex needed to define how AI could deliver meaningful business value within real insurance workflows, not just technical capability.

“The technical side is not the biggest challenge,” says Milad. “The real challenge is understanding how to bring AI into the business in a way that actually solves problems.”

Finlex partnered with DoiT through the AWS RAPID GenAI (AI Assessment) program to bridge the gap between experimentation and production. Working closely with Rajan Bhave, AI Architect at DoiT, the engagement combined technical deep dives, architecture workshops, model evaluation sessions, and hands-on support to help Finlex design a scalable, production-ready AI platform.

DoiT helped evolve Finlex’s initial LangGraph-based prototype into a more robust cloud-native architecture built on Amazon Bedrock AgentCore, improving scalability, observability, and operational control while reducing reliance on custom orchestration. The engagement also included a dedicated GenAI security and compliance workshop focused on secure agent design, threat modeling, GDPR, and EU AI Act considerations, helping Finlex build confidence in operating AI securely within a regulated financial environment.

Additional improvements included implementing secure multi-tenant identity propagation, introducing structured model evaluation frameworks, re-architecting document processing workflows using Bedrock Data Automation, and applying best-practice patterns for agent design and memory handling.

“We had sessions focused on very specific problem statements, as well as broader workshops,” says Milad. “Sometimes even code snippets were shared to help us move faster.”

Working with DoiT delivered measurable impact across Finlex’s platform, operations, and internal teams. Through improved visibility, governance, and optimization enabled by DoiT Cloud Intelligence, Finlex significantly increased cloud cost efficiency while building a stronger operational foundation for AI adoption.

_“In 2025, we reduced our costs by around 35–40%,” _says Milad. “And overall, from the end of 2024 until now, we’ve achieved more than 50% cost optimization.”

The engagement was designed to improve advisory accuracy, increase document extraction reliability, reduce repetitive support work, and accelerate contract onboarding through automated workflows and better document handling..

Beyond cost savings, the project accelerated development velocity and strengthened decision-making across the organization. Finlex successfully built its first end-to-end agentic AI solution, creating a reusable foundation for future AI-driven services and enabling teams to evaluate and launch new capabilities with greater confidence.

“We now have the confidence that if something new comes tomorrow, we can evaluate it, build it, and ship it to production,” says Milad.

Finlex is now focused on evolving from an AI-enabled platform to a fully AI-native architecture. This next phase will involve embedding intelligence across the entire user journey, moving beyond standalone assistants toward deeply integrated, automated workflows.

“We’re not adding AI as a feature on top,” Milad explains. “We're building it into how the platform works, step by step.”

Finlex is also investing in organizational transformation, reshaping its engineering structure to support this shift and introducing new roles centered around product-driven development.

At the same time, Finlex continues to refine how it approaches AI problem-solving, moving away from rigid, deterministic workflows toward more adaptive, AI-first thinking.

Looking ahead, DoiT will remain a key partner in this journey, supporting further productionization, scaling new use cases, and navigating the rapidly evolving AI landscape.

“There’s always uncertainty in AI,” Milad said. “But working with DoiT gave us the confidence to move forward.”