Supervised learning enhances Muthoot Group chatbot to achieve enhanced query resolution accuracy, Piyush Srivastav- AGM IT, Muthoot Finance

Muthoot Finance, India’s largest gold loan NBFC recently updated its virtual assistant platform to launch ‘Mattu’, an AI-powered virtual assistant. The company was using a chatbot since 2018 but was exploring a chatbot offering with good integration and interactive capabilities.

The conventional chatbots are more FAQ related however the AI enabled mattu will solve all queries that can be handled at the website and branch level (except for giving a loan because that requires the handling of physical gold).

While there were useful FAQ features well connected with other applications in the previous chatbot, the Muthoot Group was looking for a more interactive chatbot to ease the customer journeys, which could only have been fulfilled with an interactive chatbot that can be well integrated with WhatsApp, Facebook, Twitter and other channels. “Based on specific queries, we should be able to customise offers not only depending upon the information in the CRM, but also the customer information residing in the core banking software (CBS) thus giving real time information. This kind of integration is required for one specific workflow to be successful, which was present in the current chatbot,” says Piyush Srivastav- AGM IT, Muthoot Finance.

Conventionally, the chatbots claimed to be AI are structured and rule based however, “What we use is natural language processing (NLP) and not keyword matching,” says Manhardeep Kochhar, Head Marketing & Alliance, Senseforth.ai. Mattu will be able to handle the intent of what the customer is seeking. It will be able to handle a more comprehensive set of queries turning out to be a transactional AI chatbot and not just handling FAQs. The bot will be under supervised learning for training and retraining. “The problem with the chatbots build by the bigger players in the market is they are under an autonomous learning mode, which makes them to generate biases of their own,” says Kochhar.

During the course of designing mattu, it was realised that there are many security, integrations and workflows that need to be taken care of in order for the chatbot to operate as per the solution ideation. The IT team had to work vigorously on making sure about the robust security of the customer facing mattu because it had to be integrated with the CBS (having the customer information about the loan, interest, etc).

Mattu has resolved over one and a half lakh sessions (six lakh seventy two thousand questions asked) until mid September since it went live on May 7. Notably, with an overall 88 percent accuracy. The September accuracy has been 91 percent. Five lakh ninety one queries were answered directly without the need for the bot to transfer the session to a live agent.

From the Muthoot Group, there were about four people involved from the IT team, apart from the set of requirements put forth by the business team. The team from Senseforth.ai had about five people working on this launch, which took about three months.

Mattu is hosted on a virtual machine, which runs on an high availability (HA) mode. For security considerations, the API enabled integrations have a separate landing page and that they arent handled directly on the application API.

Recently Muthoot Finance had also launched a facility by which its gold loan customers can repay their loan or make interest payments through PayTM, Google Pay and PhonePe. There are also exciting cash back offers for online interest payments. Muthoot Finance Gold Loan customers can also avail loan top-ups in just a few clicks using WhatsApp. They can also get a gold loan from the convenience of their home through Muthoot Finance Gold Loan@Home service.

Senseforth’s patented technology powers enterprise-grade conversational AI solutions across the world. These virtual assistants enable large organizations to increase sales, improve customer engagement and reduce support costs.