loader image

Naya —

Vastly trained AI-powered RAG chatbot that automates customer support, saving 87% of customer care agent time.

Naya is an highly-available AI-powered chatbot designed to simplify and personalize banking experiences for customers across the world. Developed for Sterling Bank and available everywhere on the website, Naya offers real-time assistance on a wide range of services, including providing up-to-date foreign exchange rates, detailed product information, financial tutorials, and personalized advice to individuals.

With multilingual support, Naya communicates fluently in both English and Nigerian Pidgin, ensuring accessibility and relatability for the bank's diverse customer base.

Naya was designed to deliver instant, reliable, and human-like assistance, she handles over 56,000 monthly conversations with an average response time of 6 seconds, cutting down human agent workload by 87%.

With her multilingual capabilities and upcoming voice interaction features, Naya represents a scalable, intelligent solution for modern banking customer support.

The Problem

Naya (4.84)

Human Agent (3.93)

Approximate Analysis of how Naya compares with a Live Customer Agent. (Supported by sources from Zendesk and Tidio)

Since joining the bank in November 2022 as an Executive Creative Web Designer, I have seen a year-on-year increase in the user acquisition trend of the website, going by analytical reports. 

While this is a good thing, it comes with growing customer service demands.

To access some vital information or make enquiries, customers either have to turn to the various social media platforms, place phone calls, or search Google with hopes of getting answers via one of the web landing pages.

Available support channels weren’t always reachable round the clock, leading to delayed responses, reduced satisfaction, and missed opportunities to deliver timely, helpful assistance.

As a UX-focused designer, these limitations started to become a personal concern for me. Spurred by nudges from my line manager, I embarked on the mission to permanently fix that problem. 

This is how Naya was born.

The Solution

Naya was developed as a Retrieval-Augmented Generation (RAG) chatbot designed to add an additional touch point that radically improves how customers interact with the bank.

Powered by Open AI and Pinecone, Naya is available 24/7 to assist users with their inquiries, product information, financial guidance, and real-time support.

By combining natural language understanding with extensive training datasets, Naya not only automates repetitive queries but also personalizes conversations in a way that feels human, reliable, and fast. Responses are usually returned in under 6 seconds, but this can vary slightly depending on the complexity of the question asked.

Naya was launched in July, 2024, and based on user feedback, Naya had a 94% satisfaction rate in the second month of its launch.

Click on image to expand.

My Role

As the lead developer behind Naya, I was responsible for designing and implementing the entire conversational system from AI integration and training data pipelines (technically called embeddings) to deployment and performance optimization.

Naya was built on top of another open-source software, so spend was stripped to the barest minimum. Naya is self-hosted and operates independently of third-party subscription services.  

The concept was originally inspired by my line manager, whose persistent encouragement acted as a critical nudge toward building something innovative and scalable.

Their shared vision helped shape Naya not just as a chatbot, but as a pioneering product in Nigeria. 

Features

+12%

Satisfaction Rate
0 %

+52%

New Users
0

+28%

Avg. Enquiries Handled Daily
0

-20%

Average Response Time
0 secs

Naya is always online, providing around-the-clock support to customers without downtime. This ensures users can access help any time of the day, even on weekends and holidays, dramatically improving accessibility and convenience.

Naya uses Retrieval-Augmented Generation (RAG), which combines the power of large language models with real-time document retrieval. This enables it to provide accurate, contextual answers even for complex or previously unseen queries by pulling from up-to-date internal knowledge bases.

Currently fluent in English and Nigerian Pidgin, Naya is being expanded to support local languages like Yoruba, Hausa, and Igbo. This cultural adaptability helps foster trust and relatability among diverse user groups across Nigeria.

With an average response time of just 6 seconds, Naya delivers near-instant answers, greatly enhancing the user experience when compared to the 46-second average for live agents globally.

Naya can intelligently draft personalized emails on behalf of users, recommend relevant service channels, and even help them take the next step in complex workflows like document submission, account issues, or loan applications.

Over 87% of customer inquiries are handled end-to-end without needing to escalate to a human agent. This not only reduces support costs but also frees up human agents to focus on high-priority cases.

A future update will enable customers to engage in live, natural voice conversations with Naya, delivering a seamless experience that mimics speaking to a real human, with zero delay.

Naya is designed to handle thousands of conversations concurrently, up to 200 real-time voice sessions at once, making it highly scalable for enterprise use.

Summary

Naya has redefined customer interaction at Sterling. Averaging over 56,000 conversations in a single month and a projected reach of 8.75 million engagements in 2025, Naya is a living proof to how AI can meaningfully reduce operational load while improving customer satisfaction.

As the developer, this project deepened my understanding of conversational AI and large language models (LLMs), while also teaching me resilience and patience in the face of challenges and the temptation to give up.

Ready to create something huge?