Chatbots have been a fixture of customer service life for the past few years and in the public eye for nearly a decade. Unfortunately, chatbots have garnered a reputation in that time as a generally mediocre technology, or otherwise the apex of merely a passing fad.
Yet a B2B market for chatbots continues to thrive. According to a recent press release by Gartner, chatbots are predicted to become the leading customer service channel for about 25 percent of organizations by 2027.
Though historically overhyped, chatbots have proven useful for handling some simple and repeatable customer responses, and in routing customer inquiries to the right person to handle them. While the technology holds the promise of handling a vast array of customer questions within high-context conversations, we’re still relatively far from that vision today.
Distinctions Between Chatbots and Conversational Agents
At various points throughout this article, you will hear us mention the term “conversational agents.” Generally speaking, a “chatbot” refers to an AI application that responds to a customer via text, whether that be via e-mail or some other type of chat interface.
Chatbots today are generally one manifestation of a broader suite of customer service and experience capabilities. These might include:
- Medium of interaction: A conversational agent may interact with a person via a voice interface, whereas chatbots are limited to text.
- Proactive agent assistance: A conversational agent may listen to a customer call or read the agent-customer chat interaction and prompt the human agent with next steps or actions to better service the customer.
For these reasons, in some of the use cases below, the term “conversational agent” is a more apt description.
Why are Banks Interested In Conversational AI?
Enterprise conversational AI must include more than simply chat interfaces. Customers expect fast service via phone, chat, email, etc. — and they expect the record of previous conversations to be visible to the company agent when they call on the phone the next day.
Before exploring individual use cases for customer service and customer experience, we’ll unpack the importance of omnichannel and personalization.
Customers today are more available than ever across digital platforms, especially mobile. According to a report by Pew Research:
- The percentage of adults who say they own a smartphone has grown from 35% in 2011 to 85% in 2021.
- The number climbs to 96% and 95% for people aged 18-29 and 30-49, respectively.
Another critical aspect of CX for banks is personalization.
In an article reporting the results of a study by UK-based management consulting firm Capco, 72% of customers rated personalized banking experiences as “highly important,” with just 8% responding that it’s not.
The demand for personalized banking experiences cuts across generational divides. Here are the reported percentages of customers who report wanting personalized banking (1,008 U.S. customers):
- Millennials: 79%
- Gen Z: 75%
- Gen X: 74%
- Boomers: 58%
Traditional or NLP-only chatbots simply cannot deliver the level of personalization – and customer engagement that banking customers demand. The limitations of chatbots in providing personalized customer experiences are one reason for the sunsetting of overall chatbot hype.
One way that banks leverage customer demand for personalization is through AI-driven applications that provide personalized financial insights. One of the vendors we discuss in this report, abe.ai, uses AI for this purpose. Another is by delivering an immersive digital-yet-personal banking experience.
We start our evaluation of conversational AI in banking by focusing first on how personalized AI solutions can lead deliver enterprise ROI.
Conversational AI for Customer Experience
Banks’ use of conversational AI to deliver personalized financial insights is a potentially powerful use case. While banks are sitting on troves of free customer data that can be converted into personalized solutions, many have failed. Per a 2021 report by Forrester:
“Customers seek convenience and value through real-time, predictive, and proactive financial experiences that serve their needs. Alas, while customers’ expectations for value and relevance have evolved rapidly, many banks have failed to change their approach to personalization to meet – let alone exceed – those expectations.”
The ability to deliver personalized solutions directly impacts money coming in. Per a McKinsey report on the importance of banks implementing a personalization strategy, banks that plan for and execute a well-defined campaign earn 5 to 15 percent more revenue.
Conversational agents can serve as the underpinning of several personalization-focused banking use cases. Some examples:
- Financial insights – e.g. spending changes: a model detects that spending on entertainment is higher this month than the previous three months and notifies the customer with a bar chart highlighting the difference.
- Savings and Investments – e.g., automated savings: an algorithm calculates a customer’s disposable income using historical cash flows, recommends a monthly savings amount and offers to transfer said amount into a new or existing account.
- Financial Predictions– e.g., expense forecasting: a model employs predictive analytics to provide a rolling forecast of customer expenses.
We’ll first explore an example of our first bullet point above in abe.ai, a company that focuses on delivering financial insights to customers using conversational AI.
Vendor Example: abe.ai
Abe.ai (since acquired by Yodlee) is an example of a vendor that sells AI-powered, omnichannel financial insight software. The “about us – what we do” page of the company’s website purports that “the most prolific consumer services utilize data to make their services easier to use and difficult to replace.”
Abe.ai states on its homepage that its Virtual Financial Assistant (VFA) solution benefits both the bank and the customer in that it supports the “financial wellness” of the latter while assisting the former in its operational efficiency and data-gathering efforts.
The financial insight functions for abe.ai’s platform appear to use natural language processing (NLP), machine learning, intent recognition, and speech-to-text conversion.
One of the features of the software is offering “balance-driven” credit card offers. In this case, the software identifies a high balance on a credit card and matches the user to a qualified offer. The AI then notifies the user of both the high balance and offers to transfer the balance to a zero- or low-interest card.
Another feature is a “high spending alert,” where the customer is notified of an increase in spending in a particular category, in this case, groceries. Anomaly detection pinpoints the outlying financial data, and the model performs the proper calculations before alerting the customer. Finally, the software offers to update the user’s monthly budgeted amount.
High spending alert
According to the features page for VFA, enterprises can customize the platform using different tools:
- SDKs & Toolkits: A suite of tools, from business logic integration to IoT, mobile, and web channel integrations.
- Support & Deployment tools: Enterprise functions for reviewing APIs and triaging programmed business rules and AI for better engagement.
In one case study, abe.ai reports that a 50,000-customer bank needed to implement a user interface while meeting “very high user experience (UX) expectations.” The bank had previously attempted to build a CRM-based chatbot, but it did not meet set standards.
The case study lists two challenges faced by the abe.ai client:
- Delivering a high-quality, intuitive solution to a customer base with high expectations
- Reducing contact center staffing costs related to an after-hours, third-party call center
To solve the above problems, abe.ai recommended its VFA solution be positioned as the initial point of contact, and primary user interface for customers. The company states that VFA’s positioning as “a first line of service” made the AI layer responsible for capturing and understanding user inputs more effective.
Abe.ai states in the case study that the model was trained using a combination of the in-built advanced machine learning algorithms and its own professional data curation services.
The company states that its VFA solution was integrated into the existing mobile application and online banking platform. In this way, the software had access to the requisite customer data.
Another feature of abe.ai’s solution is the ability of enterprises to access “Success Metrics” via a dashboard. These appear to be KPIs specified by the client at project onset. The company states that it delivers continuous feedback to assist enterprises in improving its models.
According to the company, the solution was integrated across multiple channels, including mobile, web, SMS, and Facebook Messenger.
Regarding business outcomes, abe.ai reports the following results:
- 24% user adoption in the first 6 months
- 87% chat deflection
- 7,400 call center deflections in the first 90 days
- $166,000 annualize recurring savings (calculated using the average cost per call)
Vendor Example: kore.ai
Kore.ai is a banking and financial services solution vendor that claims it uses advanced natural language understanding (NLU) to deliver “next generation virtual assistance” platforms.
Its BankAssist product is a comprehensive financial assistant solution designed to help customers understand, improve, and manage their finances across channels.
Kore.ai states that BankAssist is omnichannel – accessible via IVR, web, mobile, SMS, and social media – and uses pre-trained data for out-of-the-box capabilities. The company claims the solution includes over 150 of the most common banking requests.
Kore.ai states that BankAssist integrates across multiple enterprise applications in its product literature.
Among the purported features of BankAssist are:
- The ability to produce outputs for multiple user intents (e.g., transferring money and locating an ATM).
- The identification and storage of relevant customer information.
- The ability to answer an unrelated query mid-conversation and return to the original request.
On its product page, kore.ai reports that customers who have integrated BankAssist into their operations have achieved the following results:
- 90% containment rate
- 18% reduction in contact center call/chat volume
- 85% automation of transaction-based use cases
Conversational AI for Customer Service
Successful deployment of conversational AI in customer service offers potential benefits to banks, such as:
- Better call prioritization, triaging and call routing
- Faster problem resolution
- Increased customer satisfaction (often due to lower hold times)
- Lower labor and utility costs
- Retasking human agents with more value-adding, complex customer issues
Intent Detection and Call Routing
Conversational AI can determine the reason(s) for a customer’s call and then route the call to the correct recipient. The use of natural language understanding (NLU), allowing customers to use their natural voice instead of keywords, has been a rising trend in recent years for this purpose.
In a recent podcast here at Emerj, we spoke with Dr. Tanushree Luke, the head of AI at U.S. Bank. We spoke with Dr. Luke at length about how natural language systems have advanced to the point where, in conjunction with machine learning, conversational agents can route calls according to urgency:
“So fraud, for example, there’s an urgency involved in it, as opposed to somebody who’s just calling in to ask a question about mortgage rates in the future…So how does an agent prioritize this [against] the 10 calls that they have? Which ones should they be answering immediately? Which one is on fire? That’s the way to think about it.”Dr. Tanushree Luke, Head of AI at U.S. Bank
AI can be used to trigger an automated conversation to verify customer credentials from digital channels such as chat, email, messenger, or webform. For example, Cognigy.ai offers a conversational AI solution that guides a user through a two-factor authentication process using SMS.
Conversational AI can also be used for user authentication using the customer’s voice. Advanced AI makes it possible for chatbots or voice assistants to recognize who they are talking to and allow the user to proceed.
Voice and Chat Automation
Despite increasing digital engagement through messaging and other mediums, banking customers will continue to call, especially in times of crisis.
Per a report published by Accenture in March 2020 – the onset of pandemic uncertainty in the U.S. – 57% of customers ranked call support as their preferred communication channel. In the same report, 58% stated they prefer to solve urgent issues through phone calls.
Advances in AI technologies, especially the three “natural languages”: processing (NLP), understanding (NLU), and generation (NLG), mean a more human-sounding AI agent at the other end.
Vendor Example: boost.ai
Boost.ai is a Norwegian customer service-focused AI vendor that offers no-code conversational AI solutions for banking and other industries.
The company claims on the product page of its website various platform features, including:
- A no-code platform that makes for a more accessible design and deployment experience.
- A multi-layered “intent interface” allowing users to design complex workflows.
- Analysis of conversation data for continuous platform improvement.
The solution appears to be a simple drag-and-drop interface with customer intents as the only option in the sidebar. After creating a new intent, the user enters a name for the new intent (e.g., “Status of claim”). From there, the user is presented with a drop-down menu consisting of seven options:
- Entity extraction
- API connector
- Search existing
Here is a helpful, short video demoing Boost’s user interface.
Boost reports that each of its markets – Denmark, Finland, Norway, and Sweden – received thousands of daily customer service requests and had problems managing call volume. Before implementing Boost’s AI solution, Nordea was receiving up to 2 million contacts annually across three contact center locations with a full-time staff of 150.
Boost recommended a conversational agent as the best solution for decreasing call center volume. The bank had previously discovered that chat was the preferred communication medium.
The model was trained using Boost’s pre-built “module of common banking-related topics” that Nordea could then customize with their brand-specific content. The AI model took the form of 12 “virtual agents” across the four markets.
According to the case study, the AI can answer questions on more than 2,300 banking-related topics for over 50,000 monthly conversations. About 25,000 of these conversations are handled solely by the AI with no human agent intervention.
Other results include:
- 91% in-scope resolution rate for private banking customers and 95% for corporate customers.
- Successful automation of hundreds of inquiries daily