Ever wondered about the volume of calls to banks? It is probably inevitable that banks would turn to chatbots sooner or later. Chatbots offer huge opportunities for banks across three broad dimensions.
Efficiency and opportunity
The most expensive forms of interaction with customers are the personal ones: face-to-face, and by telephone. This explains why banks have encouraged customers to adopt internet banking. But it also explains why banks are looking so hard at chatbots: reduce the price of telephone interaction by using bots, and you have made some serious efficiencies. But of course to succeed, the chatbots have to be effective.
A number of chatbots are already in the pipeline for specific purposes. MasterCard and Bank of America both have plans for chatbots that can provide answers to specific (and relatively simple) questions, such as providing details of the last transaction, and even giving some basic financial advice. Both will use Facebook Messenger, and Western Union has also announced plans for a chatbot to allow customers to send money via Messenger.
Banking apps are not always the most user-friendly, and chatbots offer the opportunity to improve user experience, and make banking easier. Getting chatbots right could turn out to be the key to attracting and retaining customers. Investment in data science input is likely to be very worthwhile. Chatbots could indeed be the new competitive frontier in banking.
Chatbots that are really useful for customer services are quite hard to create. There are two main types of chatbots: those that have a limited number of options for response, and so control the conversation, and those that do not control the conversation. The first is essentially a talking web form, like those described before. The second is required to take customer service interactions with bots to the next level. Unfortunately, it is also much harder to create, because it is impossible to code for every option.
The list of requirements for chatbots is deceptively simple, Chatbots need natural language processing, ability to engage with a particular context, real-time data and ability to cite written materials when appropriate, and ability to work across channels. They also need deep knowledge that can only be supplied by correct use of data. Finally, they must be able to learn over time, and anticipate customer needs.
Machine learning, may be the way to create chatbots that can interact with customers without controlling the conversation, learn from experience, and anticipate needs. These chatbots would either divert questions to a human operator and ‘watch’ the human response, learning from it for next time, or be taught by human operators based on selected real interactions.
But careful design is crucial to getting it right. We probably all know the cautionary tale of Alexa ordering dolls-houses. When chatbots go wrong, it can be spectacular and messy. Building in safeguards is essential, such as allowing people to divert from the chatbot to a human when they wish, and flagging phrases that might expose the bank to fraud. Limiting the options and keeping it simple are also both good ways to improve customer interactions.
Humans as the limiting factor
Research suggests humans are interested and willing to engage with chatbots. Huge numbers of people asked for more suggested commands, but very few actually tried any commands outside those suggested. This means that developers need to think about how to encourage users to experiment with commands.
All these issues add up to data scientists being essential in building deep learning algorithms and creating better banking chatbots. They play a vital role supplying the right information, including real-time data to enable the chatbot to understand, for example, why customers may have called, or when the bank is at risk of losing a customer. Data scientists are essential to chatbot development.
The relationship between chatbots and data scientists can be two-way. Data scientists support chatbots by supplying suitable sources of information, and making sure that the bot is calling on useful data. But data from chatbot interactions can also support other data science and analytics projects, meaning that there is a real incentive for data scientists to engage.
Have your say
As is often the case with emerging and embryonic capabilities, many options exist. What have your experiences been? We want to hear from you, whether you design, deploy, work with or use chatbots.
Join our conversation on Twitter on 22nd June, from 14hrs CET/ 13hrs UKI/ 0800 hrs EST. We will be posing these questions to our kickoff panel, and the discussion will be on the twin hashtags of #chatbots #AI stream.
- What are the best chatbot example in the market today?
- How are banking chatbots different?
- What are the key metrics used by banks to assess effectiveness of their chatbots?
- What are the main drivers of customer adoption?
- What do you see in the future of chatbots?