Approving 70% more loans for underserved customers using AWS with Nequi

In Colombia, many people manage their finances without the borrowing histories that traditional credit models rely on. For these individuals, obtaining loans and other financial products through conventional banking models can be difficult or impossible.

As more customers joined Nequi, customer support volumes increased, response times became harder to maintain, and the cost of managing interactions continued to rise. Operating without physical branches, the financial services provider needed to handle interactions digitally, balancing customer experience and operational efficiency.

Nequi believes that understanding customers is essential for expanding financial inclusion. “More than educating people on how to use financial services, we place a greater emphasis on learning from them,” says Ruben Vargas, chief data and AI officer at Nequi. “We listen first, so that we can accompany each customer in the best possible way.”

To extend credit and handle millions of queries, the bank decided to use AI on AWS. This would help Nequi assess customers with limited credit histories, deliver digital customer service at scale, and create personalized experiences.

Nequi wanted to support a production-ready multiagent architecture to serve millions of customer interactions in a regulated environment. So the fintech used Strands Agents, powered by Amazon Bedrock, a service for building generative AI applications and agents at production scale. “Everything at Nequi grows fast, and the AWS team has been a great ally in helping us build systems that run at scale,” says Vargas.

By building a supervisor AI agent that orchestrates more than 10 specialized agents, Nequi created a coordinated AI system that directs customer interactions to the appropriate specialized capability. Together, the agents support customer service, transaction history, fraud-related requests, and advisory handoffs while helping create more natural customer experiences. “We’re using AI to understand people through a conversation, not only a click,” says Vargas.

The multiagent architecture also serves as the foundation for Nequi’s broader AI operating model, extending AI capabilities across lending, digital collections, personalization, and internal business operations. Because AI is embedded at the core of Nequi’s business, the company built Nequi Agents, an internal platform that’s used by all employees. Nequi Agents is powered by Amazon Bedrock AgentCore, an AWS service that helps companies get agents to market faster, scale without rearchitecting, and focus engineering efforts on differentiation. Beyond answering questions, these agents retain context across interactions and connect with external systems using capabilities such as retrieval-augmented generation and model context protocol.

To improve lending decisions, Nequi chose Amazon SageMaker AI, a service for building, training, and deploying AI models for virtually any use case. Using graph ML and synthetic data in Amazon SageMaker AI, the bank developed an alternative credit-scoring method that analyzes customer relationships and financial behaviors beyond traditional credit histories. This broader view of customer behavior helped identify creditworthy customers who were previously invisible to traditional financial models. Nequi further applied AI to digital collections, using Amazon SageMaker AI and Amazon Bedrock to personalize notifications and conversational interactions that help customers manage overdue accounts.

Nequi began its personalization journey by using Amazon Personalize—a service that helps elevate the customer experience with AI-powered personalization—and has since evolved toward Amazon Bedrock to power personalized campaigns with generative AI. These capabilities help create relevant customer experiences and recommendations while increasing experimentation across teams.

To help teams incorporate AI into everyday work, Nequi rolled out “University GenAI,” an internal AI training program that was developed with Nequi's AWS account team. The bank is also advancing an agentic AI development strategy powered by Kiro and Claude Code on Amazon Bedrock, adopting spec-driven agentic workflows across its engineering teams. "This effort is centered on a company culture that supports the use of technology to change the way people interact with financial systems," says Vargas.

The graph ML–powered credit models that Nequi developed in Amazon SageMaker AI helped the bank double its preapproved credit customer base within 1 year and increase loan approvals for customers with limited or no credit history by 70 percent. Now, by combining its multiagent system with deterministic flows and complementary collection strategies, Nequi manages 80 percent of its overdue portfolio internally and digitally, cutting credit management costs by 50 percent while preserving customer trust.

The system resolves 80 percent of customer inquiries through self-service while achieving 90 percent FAQ accuracy. With a 10-second average response time across the bank’s daily message volume, it helps customers receive answers faster while reducing pressure on support teams.

Nequi also increased marketing conversion rates by 30 percent and sped up experimentation by 10 times. Together, these improvements help the bank continue expanding its AI capabilities while lifting barriers to financial access.

Looking ahead, Nequi will continue evolving conversational financial experiences while responsibly expanding AI under Colombia’s banking regulations. “The strategy is to use AI to help us give credit to people who have never had access to it,” says Vargas.