Live Innovative

How DBS Bank's KAI banking chatbot is setting a new benchmark in customer service

In the last two years, the KAI virtual banking assistant has been able to handle 80% of the queries received by the bank with close to 80% accuracy. This has led to a reduction in the call centre hiring "We not only see Private or Public Banks as our competitors but also see our competitor landscape as expanding to include many other fintech companies. And in fact, many of them have adopted UPI payment interfaces into their platforms which is a direct service that is in competition to ours. We see their biggest advantage as their knowledge on how to achieve mega-scale, i.e getting to a billion customers or more at a rapid pace, and they are religious about customer experience," says Alfian Sharifuddin, MD- Head of Technology and Operations, DBS Bank.

As one can see from Sharifuddin's quote above, about an year ago, technology platform scalability and customer experience was the focus for DBS Bank, while launching Digi Banking in India. In other words, as the customer count grew, the bank did not want to see itself scaling up with manual, non Straight Through Processing (STP) processes. For example, there should not be any corresponding increase in the staff irrespective of the rate at which the customer base is increasing.

The bank was looking for tools, processes and capabilities available in the market that would help to meet this need across all the functions. KAI Banking fitted DBS Bank's expectations. The platform is AI enabled and can do voice and text communication with the customer. All the major FAQs like balance check; branch opening hours; how to open a bank account; queries regarding upgrades; what to do if the debit card is lost etc from the customer can be answered with relative ease by the platform without going into the chain of multiple clicks or standardised short code formats. The platform is intelligent to understand what's being asked. There is no need for asking in specified formats. Certain keywords does the work for the AI enabled chatbot.

Chatbot has delivered satisfactory numbers

This also takes away the customer's pain of answering the Bank's Interactive Voice response (IVR) or interacting with the call centre agent. Out of all the responses from the bank, on an average more than 80% queries are answered by the KAI platform and the rest 20% responses are from the call centre. This year till date (1-Jan-2017 till 14-Nov-2017), the digibot virtual assistant has handled over 1 million conversations (each conversation averages 3 questions) and 87% of the conversations are handled by the bot without the need to hand off to a human agent. The average response time by the chatbot is in subseconds i.e in less than 0.5 seconds. "We are so happy with the performance of the KAI platform, that we have made an investment in the startup company, that provides the solution viz., Kasisto," says Sharifuddin.

Kasisto is also an important player in the conversational AI industry because it was born in the Stanford Research Institute (SRI), which was the original company that designed Apple's Siri platform.The company has been in the business of Natural Language Processing (NLP) for years. "One of the core branches of Siri has been branched into Kasisto," informs Sharifuddin.

The platform thus far has reduced the cost to the tune of less hiring at the call centre. In the absence of the KAI platform, close to 300 agents would have been working in the call centre, which is just 55 now. The bank, when it kicked off operations (DigiBank India) about an year ago, started off with 15 call centre agents. This was basically done for queries which are advanced in nature and could not be solved by the bot. The call centre currently, handles 20% of the calls. The KAI platform resolves 80% of the queries. This has resulted in tremendous amount of savings, which, when compared to the recurring savings on an annual basis would be a huge amount. In the first six months, the bot was like a baby and was only trained for the conventional queries. Over a period of time, as the nature of the queries became more complex, the bot was made to understand the transcripts. It helped the bot to solve similar queries, when it was asked the next time.

The bank is now keen to train the device to keep reducing the number of calls handled by the call centre. "We are focussing on training the KAI platform. Any AI device is like a blank child. The more it learns the better is becomes," says Sharifuddin. It's like DeepMind, the Google AI that defeated the number one player in the Game of 'Go'. It simply happened by making the DeepMind AI play millions of games to learn more about the game. Similarly, DBS Bank is training the KAI platform to solve more customer queries. The unsolved requests and queries by the KAI platform are identified and then it is trained on how could it have been solved. "This makes the bot evolve and become smarter, week-after-week. In the initial months, the percentage accuracy of the responses from KAI was about 55% - 60%. However, in the last six to nine months, it has been improved to 87 and the target is 95%,"

Whenever the call centre head comes for an additional hire, Sharifuddin asks for the number of annotations that KAI has been trained for in the last month. This way, the call centre hiring has tapered off in the last few months and the focus is now on making the Bot more smarter.

KAI platform: Well connected with other channels

The KAI platform has been integrated with the web and the app channel. "We are actively assessing the voice channel, wherein the voice accuracy, language and accent tuning is being worked upon" states Sharifuddin. The bank has been consistently ranked number 4 in the Google Playstore and the KAI platform has contributed immensely to the ranking. "Instantaneous query resolution is one constant feedback, we get. This is because people hate to wait for the call centre agent."

DBS Bank has adopted a hybrid model for hosting the KAI platform. A part of the KAI platform is on premise, because of security considerations. The rest is hooked onto the AWS platform. The bank makes sure that all the customer data and the personally identifiable information is stripped off before the information is sent for processing.

Future plans

DBS bank has launched banking through Facebook messenger, Whatsapp in different geographies globally. The bank is actively considering launching the same for the India market too. This feature will be launched when the market will be ready for it because, the customers in India think that banking on Facebook might not be as safe as banking on a bank's parent platform. The best risk based approach is alo being considered to launch such services in India. In terms of the type of transactions, interbank transactions can be done. However transactions where the bank has to interact with any other non financial entities, the KAI platform has the functionality ready, but it is yet to be launched.