Chatbots to the rescue?

Think chatbots, intelligent virtual assistants, and digital employees. These and related technologies provide another way for people to communicate with banks (sometimes without even realizing that they're not talking to a real person) – thanks to the power of conversational artificial intelligence (CAI).

CAI makes it possible for banks to respond to customers' questions more quickly, cost-effectively, and consistently than they could with a traditional workforce. Juniper research predicts that, by 2022, chatbots will reduce call duration by four minutes and interaction costs by $0.70 per call.

Banks' interest in CAI has surged recently due to customers' concerns about in-person banking, which has led to intense pressure on banks' call-center personnel. For example, since the pandemic hit, Citizens Bank has launched its first chatbot, and Connexus Credit Union began rolling out Alexa voice-banking capability. Meanwhile, Bank of America added one million Erica users per month during the first three months of the pandemic.

Many banks embark on the CAI journey by launching chatbots. Unfortunately, most chatbots fail to meet the objectives they were designed to achieve. This may explain why only 16% of the consumers we surveyed in our third annual AI 360 study expect that they'll prefer service by a chatbot over a human by the end of 2021. But COVID-19 has changed consumer behavior and elevated CAI from a nice-to-have feature to a must-have differentiator for banks.

So, why do most chatbots fail? Answer: the success of a chatbot program depends almost entirely on whether the people developing it have the experience and skills to tackle five important questions.

Not all chatbots are created equal. With dozens of providers, it is important for a bank to select a platform with the right level of sophistication to meet its needs and goals. Broadly speaking, banks can choose from three types of chatbots:

• Scripted bots: A scripted bot provides static responses to preset, keyword-based questions or statements. Developers must preprogram every possible question and syntax and its corresponding answer. For example, they must program “What's my account balance?", “What is the balance of my account?", and “Please give me my balance" for a user to get the answer they need. A scripted bot is the most simplistic of the chatbots and is not based on artificial intelligence (AI).
• Contextual bots: If you want to give your banking customers more than just canned answers to simple questions, contextual bots are probably your best bet. Contextual bots use natural language understanding (NLU) to determine what customers are saying or asking regardless of the syntax they use. These bots also keep track of the context of the interaction. So, if a customer asks, “What is that?", the chatbot knows what the customer discussed earlier in the conversation and understands what they mean.
• Learning bots: All chatbots must learn which questions to expect and which responses are appropriate. But platforms with machine learning (ML) capabilities automate that process. There are different levels of ML. For example, ML bots can learn different forms of the same question. Or they can discover entirely new topics and patterns. What's unique about learning bots is that you can teach them how to learn. ML developers can feed the bots vast amounts of data – such as previous responses to actual questions – so that they can learn from the data on their own.Of these three types of bots, scripted chatbots can implement most quickly. But their capabilities are limited. And they may end up leaving banking customers feeling underwhelmed. On the other hand, CAI platforms with NLU, ML, and contextual capabilities generally take longer to develop but can more effectively address customers' inquiries. CAI refers to these more advanced forms of chatbots.

A bank's policies and procedures form the basis for the responses that its CAI provides. But when banks begin to build CAI, they often realize that some of those policies are not documented anywhere. And if they are, the documents are often outdated or have conflicting versions.

Identifying and fine-tuning the source of knowledge for CAI is a task that can take many months, but it is one that banks must complete before they can teach their CAI anything. Without taking the time to feed the CAI platform the right knowledge, it can end up doing more harm than good, becoming a source of confusion and misinformation for customers.

Training a chatbot is not a one-and-done exercise. Banking services evolve, trends shift, customer preferences change, and new technologies emerge. So chatbots must constantly update their knowledge. Regardless of the method used to teach the CAI platform, your bank needs a clear plan for how the CAI will update its knowledge.

Just as your brain helps you interact with other people in a variety of ways, such as through speech, writing, or movement, a CAI platform helps banks interact with their customers through a number of channels such as web, mobile, text, interactive voice response, digital voice (including web-based and smart speakers), other internet-of-things devices, and social media. But not all channels work the same way.

Certain interactions are better suited for certain channels. For example, banks can easily summarize customers' account information via smart speakers. But it's best to provide detailed transaction information through a screen.

CAI that is a one-channel wonder doesn't help anyone. It's essential to be clear about the delivery channels planned for your bank so that the CAI engine can adequately support them all.

Effective CAI must interact with core banking systems to process customer inquiries and requests. Cloud-based contact centers enable financial institutions to seamlessly incorporate new capabilities, such as CAI, into their customer support ecosystems and make omnichannel integration a reality. However, banks often have systems that operate in silos or based on older technologies, primarily with on-premise environments instead of cloud architectures.

Banks must handle integration of the CAI with such systems carefully. They must gather and update information – ideally in real time – while maintaining the integrity and reliability of systems of record. And it's particularly important for banks to implement security measures to ensure that the chatbot doesn't open the door for fraudsters to sneak in.

CAI represents a new way of doing things for banking customers and customer service agents alike. As with any big change, it's important to have a plan in place to ensure smooth adoption. Banks need to set expectations regarding what CAI will and will not do and how customers should interact with it.

CAI allows banks to respond to customer inquiries:

• Immediately – rather than only when an agent is available
• Round the clock – instead of just within business hours
• Consistently – rather than based on agents' memories or interpretations of bank policies
• Cost-effectively – reducing call duration and interaction costs

Meanwhile, CAI frees up customer service agents, allowing them to shift from simpler tasks to more complex and critical ones (including conversations that drive sales) and from functional roles to more fluid, purpose-driven careers that deliver greater value to the organization. Working together, CAI and agents provide banks with a more adaptive workforce.

When it comes to available CAI solutions and platforms, there is no shortage of available options. What's critical is making the right decisions when answering these five strategic questions and executing on those decisions effectively. This requires specialized, human intelligence in CAI for banking, which usually comes from professionals with solid experience developing CAI and other technologies for the banking industry.

CAI is becoming a business imperative for banks. Are you thinking about how to best prepare your bank for it? If so, it's time to start the conversation.

This article was authored by David Vila, Omnichannel Leader, Consumer Banking. A version of it was originally published in FinExtra.