Design and Testing of a Chatbot for Customer Service at a Microfinance Bank in Lagos State

Customer service bottlenecks at microfinance banks in Lagos State, particularly during peak hours, result in long queues, delayed enquiry resolution, and poor customer experience. This study describes the design and testing of an AI-powered chatbot for automating routine customer service interactions at a microfinance bank operating in Lagos Island and Surulere branches. The chatbot was developed using Dialogflow as the natural language understanding engine, integrated with a Flask-based webhook and connected to the bank's core banking API for balance enquiries and transaction history retrieval. Training data comprised 1,200 labelled customer enquiry samples categorised into eight intent classes, including account balance, loan status, PIN reset, and branch location queries. A knowledge graph of 150 frequently asked questions supplemented intent-based responses. Testing involved 50 volunteer customers submitting 200 live queries during a two-week evaluation period. Intent classification accuracy reached 89 percent on the test set. Customer satisfaction scores averaged 3.9 out of 5.0. Escalation to human agents was required for 18 percent of queries. Average response time was 1.2 seconds, compared to an average human agent response time of 6.8 minutes during peak periods. The study concludes that chatbot-assisted customer service can meaningfully relieve pressure on human agents at Nigerian microfinance institutions and recommends extension of the system to WhatsApp and USSD channels for broader customer reach.

Keywords: chatbot, customer service, microfinance bank, Lagos State, natural language processing