How AI Is Changing Customer Experience

By Bob Meara, Senior Analyst, Celent

Artificial intelligence (AI) has quickly moved from science fiction to the mainstream across many industries, including retail and commercial banking. AI is proving itself to be useful in a variety of ways in back-, middle- and front-office applications. Not all of these applications tangibly impact the customer experience (CX). Those that do are of significant and growing interest among institutions of all sizes across the globe. This article looks at the ways in which AI is changing the customer experience, how this is happening and the business case for doing so.

AI is influencing CX in two primary ways. The first is through personalised data insights and advice generation (also known as next-best-action or next-best-conversation)—both direct to the consumer via digital channels and/or via interactions with a business banker, contact centre or branch staff. More recently, conversational AI is increasingly influencing CX, taking on a variety of forms. Specific technologies include natural language understanding, generation and processing (NLU, NLG and NLP) and predictive or propensity modelling of various types. These technologies are not used singly but are combined to support a large and growing number of use cases.

Data insight and advice generation

Many banks have a general understanding of their customers and/or customer segments. Fewer have a deep understanding at the individual customer level. Some have built the capabilities to understand individual customers’ profitability, lifetime value and/or share of wallet. These are useful perspectives, but they are rarely used to inform how banks engage with individual customers. Operationalising the use of customer data to form actionable insights that inform the customer conversation in real-time, across all touchpoints, is a rarity. But, thanks to AI, it won’t be for long.

A small but growing number of banks are creating “decision hubs” powered by a portfolio of statistical models and fed by readily available customer data to create next-best-conversations that are customer, context and channel aware. Initially built to maximise sales effectiveness at many banks, these hubs are increasingly designed to support a diverse set of business goals such as acquisition, retention, cross-sell and compliance but are stringently prioritised against each other—to ensure banks engage customers with the most relevant next-best-conversation possible—at each point of interaction, from branch and contact-centre conversations to mobile apps and chatbots.

Chatbots, conversational AI and virtual assistants

Chatbots (and other variations on the theme) have been a white-hot area of development over the past few years. All have two important elements in common:

1. Conversational interface. Rather than a user navigating through an app or website-user interface to meet his or her needs, conversational interfaces can offer a convenient and often faster user experience. Text chat dominates this realm now, but voice is quickly gaining ground for a subset of use cases.
2. Automation. Conversational interfaces are not new. Firms have offered live chat for years and have invited person-to-person conversations in contact centres and branches. Doing so requires considerable human resources, however. The key benefit of automation is in its ability to meet routine customer needs conveniently for customers and at low cost for the institution, freeing resources for more complex tasks.

Chatbots broadly refer to natural language text interfaces used as an alternative to navigate an app or browser to accomplish certain tasks. Celent refers to all types of automated conversational interfaces as chatbots and AI-powered bots as virtual assistants. First-generation chatbots were constructed using rules that encouraged linear interactions navigated by pre-defined flows. Chatbots were designed for a specific and narrowly defined number of tasks. If a given user’s need could be met using predetermined flows within the bot, then the CX was favourable. But if the user’s intent veered outside the bot’s predetermined-scenario portfolio, then the CX could quickly deteriorate. This limitation is why first-generation chatbots were generally a disappointment. The good news: more capable alternatives are now available, thanks to conversational AI.

Conversational AI utilises natural language processing (NLP) and natural language understanding (NLU) to enable dynamic dialogues between a customer and a machine; that is, one with multiple turns as opposed to a single-turn static conversation (for example, “What is my balance?” “What was my last credit-card charge?”). The AI model interprets context and sentiment, going beyond searching for an answer and delivering it. The dialogue can be by voice or text. Text is more mature than voice technology. Both require a seamless transition to a person when the conversation exceeds the ability of the technology to provide a satisfactory outcome. Of course, AI is used to make that determination.

Today, most virtual assistants use a chat interface, but a limited number of avatars are available. Avatars and virtual assistants are also used for internal needs such as the IT (information technology) helpdesk.

What AI is doing

In Celent’s research, we see AI profoundly influencing CX both in customer-facing applications through digital channels as well as in employee-facing applications. In retail banking, customer-facing use cases dominate, while in corporate banking, employee-facing use cases have the edge.

Customer-facing applications

A useful way to distinguish use cases is to map AI applications by the pyramid of customer needs. At the base of the pyramid is “Tell me—basic queries”—that is, the customer has a basic question (for example, “How do I change my password?” “What is my checking account balance?”). A notch up is “Do it for me”, which includes facilitating account onboarding and basic-task optimisation (“You can send an electronic request for payment to this buyer”). The next level is “Tell me–data insights”, which includes descriptive (such as report generation) and predictive analytics (cash-flow forecasts). At the top is “Advise me”, which involves bespoke recommendations to address a specific need or resolve a specific issue (for example, “A cash shortfall is expected; here are three options to cover it”). Celent’s research suggests that most banks striving to scale are focused on the bottom two rows of the pyramid, representing roughly 70 percent of all use cases in production among leading AI vendors.

While not as sophisticated as next-best-conversations, these use cases can improve CX while providing a compelling return on investment for banks.

Employee-facing applications

Similar to customer-facing applications, AI is being used to improve the tools available to relationship managers (RMs), contact-centre and branch staff. Since employee-facing use cases are more prevalent in commercial banking, we’ll focus there. For example, an AI platform could detect customer dissatisfaction during her last call to the service centre, which could trigger an outreach by her RM. Or it could perceive a negative outlook for a customer’s business made by an analyst, which could trigger a review of the customer’s loans outstanding.

By increasing RMs’ productivity while making their jobs more interesting, AI will be increasingly embraced as a partner instead of a competitor. RMs will share their desktops with their own virtual assistants, enabling them to delegate basic tasks. The concept of humans and machines working side by side has already been realised in compliance and fraud operations in which humans can review the work done by AI and focus on the cases that AI cannot resolve—and through their actions, train the machine-learning model. A useful way to assess RM-facing AI applications is across the customer lifecycle, beginning with marketing and prospecting to engagement, cross-selling and monitoring. Celent’s research finds the top use cases are: customer engagement (includes inbound and outbound interactions, 36 percent), relationship building (includes cross-selling, 22 percent) and client monitoring (19 percent). Like customer-facing use cases, there are relatively fewer implementations upstream.

The business case for AI

No matter how appealing or headline-grabbing a use case is, it needs a business case to garner resources. Recent Celent research asked AI vendors, based on their banking experiences, to rank three business cases: cost savings, revenue generation and customer-experience improvements. The results suggest cost savings is the primary basis for justifying front-office AI investments (see next figure). But AI investments are rarely tied exclusively to cost savings. This signals that banks are increasingly confident in AI’s ability to generate indirect return on investment (ROI)—that is, a return that is difficult to tie casually to a hard-dollar impact. For example, improvements in customer experience should drive an increase in revenue growth, but so could other factors (for example, prices and rewards).

There are a variety of reasons why cost savings leads. First, banking has all the attributes of an industry for which AI shows the potential to deliver significant cost savings. It is a relatively process-heavy, paper-intensive business with many repetitive tasks. Moreover, its labour costs tend to be above average. Second, advancements in conversational AI are building banks’ confidence in experimenting and then implementing AI to augment call centres, and they are witnessing impressive call deflection. Third, customers have basic “Tell me” and “Do it for me” needs that a virtual assistant can fulfil faster than a customer service representative (CSR), which leads to a secondary positive impact on customer experience. Success at realising cost savings while fully supporting a customer, however, is predicated on a seamless hand-off to a CSR when necessary.

COVID-19 has introduced another business-case justification for AI: scale. In the United States, Congress passed the CARES (Coronavirus Aid, Relief, and Economic Security) Act, designed to assist small businesses through the pandemic. It resulted in two decades of small-business loan applications to be processed in a single month. The massive demand transient brought banks’ loan-application operations to their knees because they are not typically highly automated. Similarly, consumers flocked to digital-banking channels and contact centres as banks closed branches, resulting in multi-hour wait times to speak with a contact-centre agent. All of these scenarios have destroyed CX during the pandemic while increasing costs for the banks. Judicial use of AI in each of these examples would have produced a win-win for banks and their customers.

Hopefully, we won’t be facing another pandemic soon, but demand transients, albeit less severe, will happen again. Banks equipped with virtual assistants were able to train them with COVID-19-related scenarios and deflect significant majorities of common queries away from their contact centres. With the current state of vendor readiness, a bank can implement a high-functioning virtual assistant and have it fully tested in a matter of a few months. Over the last two years, front-office AI implementations grew at nearly 20 percent CAGR (compound annual growth rate) globally. We expect COVID-19 to catalyse rapid growth again in the coming year.

Celent, a division of Oliver Wyman, is a research and advisory firm specialising in financial-services technology. For more information, see www.celent.com.