The Evolution of Chatbots in the US

Online experiences – mobile or desktop – have a similar set of shortcomings originating from a limited set of UI structures. Here are some of the most common issues with online experiences: difficulty to navigate sites, finding answers to simple questions, finding basic details about the business, time-consuming efforts to find the services, poorly designed smartphone apps, useless search options, inaccessible functions on mobile devices, etc.

Nonetheless, there are ~4.1 billion users on messaging apps, and intelligent conversational chatbots are the new interfaces for these apps. They are changing the way businesses and customers interact. Powered by artificial intelligence, those computer programs, when just introduced, held a great promise of enabling close to natural conversations with people.

While the amount of intelligence and abilities to communicate at a human-like level have been largely overestimated across industries, intelligent bots do represent a shift towards conversational interfaces, through which businesses can serve their customers 24/7 in the most efficient manner and at a high rate of customer success. Oracle claims that chatbots could save $174 billion across insurance, financial services, sales, and customer service.

Chatbots may not be the only definitive answer for improving customer service, but they can go far to improve the responsiveness and efficiency of a company’s customer service function. Chatbots are increasingly being implemented in B2B, B2C, and internal communications in the financial services industry. Customer service bots are enabling customers to check balance, transfer money, pay bills, and more. All of this, however, is still the starting (or at most, the transitional) point of the evolution of chatbots in the US, where the vast majority of examples are cases of a rushed jump into the chatbot bandwagon – a fear to become irreparably irrelevant for customers.

The evolution of chatbots in the US can be illustrated with a continuum on which every existing example can be placed: from informational bots at the beginning, to advisors at the most sophisticated stage of development. With every step from an informational stage to advisor, chatbots are enriched with functionality. The pivotal point in this evolution is the ability to not only understand the request and provide information but to act on that information.

Let’s walk through three major stages in the evolution of chatbots in the US with particular examples:

Informational Chatbots

For a chatbot to create a meaningful and successful interaction with the customer, the experience has to be at least 10 times better than the one with the human assistant or without the assistant at all. Informational chatbots, however, in their most incarnations today, can’t deliver even an experience that is three times better. Having a highly disappointing experience with banks, network operators, airlines, telcos, retailers, and social media platforms because of their assistants in various forms, I could distill the issue with informational assistants to three main points:

  1. Inability to understand the intent. Most informational assistants understand separate words from a request – whether oral or written – without understanding the context and intent. In any request, they look for key phrases or words and give scripted responses.

  2. Companies are too fast in implementing chatbots; they force users to find the necessary scripted option from the list of queries that their bots can process (the user, however, doesn’t know that list, and is just pointing into the sky with requests). In almost all cases of my interactions with chatbots – voice or text – the difficulty to reach a human assistant is increasingly frustrating and highly disappointing.

  3. Informational chatbots repeat the scripted answer from the FAQ section, but cannot execute on the information provided. My attempt to receive a refund for a service that has been discontinued, but for which I have been still billed has been infuriating to the least – the chatbot does not understand the query + no human assistant enters the discussion + the chatbot cannot perform the transaction upon my request.

One example is American Express: on April 2017, at F8, Facebook’s annual global developer conference, American Express announced an updated Amex bot for Facebook Messenger enhanced by AI and servicing technology. The updated Amex bot for Messenger enables eligible US consumer and OPEN card members to get on-demand answers from the Amex bot for Messenger to answer certain queries related to their account and card information.

To be fair, the experience with informational chatbots and customer success depend highly on the industry that we are talking about. The examples mentioned earlier are not the ones where this stage of development of the chatbot is sufficient for effective customer service. Areas such as travel, e-commerce, and general info search, on the other hand, are probably better-suited for informational chatbots.

Matt Gillin, CEO of Relay Network, shared that for chatbots to be successful, companies should first determine the specific use cases that could benefit from this technology. Gillin’s recommendation is that bots are best for scripted transactions or tasks that don’t require a lot back and forth. Chatbots are most effective in situations where a customer is trying to resolve routine issues, complete specific tasks like placing an order, or guiding a user through a multi-step process, Forbes reports. The benefit is the ability to close the loop with the customer along a process, efficiently and in a delightful way, shared Gillin. The ROI is in cost reduction, efficiency, and improved customer satisfaction.

The financial services industry, however, is not the one where this stage of development is sufficient, and, in most cases, modern banking chatbots are on the next stage of development, incorporating a transactional capability.

With consumer brands, informational chatbots are more likely to be developed for and deployed in third-party environments (most often on social media platforms). One of the reasons is the level of risk involved in enabling a third-party platform to perform a transaction for a service/product of a brand.

This isn’t always the rule though. In 2016, Mastercard announced its plans to launch an AI bot platform that allows consumers to transact, manage finances, and shop via messaging platforms. The Mastercard bot for banks promised to seamlessly extend Mastercard services to customers on messaging platforms and make financial information and decisions part of consumers’ everyday lives. In the pilot phase, Mastercard partnered with Kasisto. The Mastercard bot was powered by KAI Banking and could fulfill customer requests and solve problems on messaging platforms such as Facebook Messenger and SMS.

Information + (Trans)action

The natural development of informational chatbots is the one with the ability to act on the information – namely, to perform a transaction. These chatbots are likely to be developed and deployed in proprietary environments, performing actions related to the primary business. Fortunately, a number of chatbots from the largest institutions in the US have already reached this stage.

Bank of America’s chatbot Erica, launched in 2016, for example, was already at this stage of development. Erica uses artificial intelligence, predictive analytics and cognitive messaging to help customers do things like make payments, check balances, save money, and pay down debt. It can also direct people to look up their FICO score and check out educational videos and other content.

BNY Mellon has a different, more inward-facing application. The 233-year-old custodian bank deployed an army of 220 bots – software created to carry out an often repetitive task that would normally be performed by humans, range from automated programs that respond to data requests from external auditors, to systems that correct formatting and data mistakes in requests for dollar fund transfers.

In 2017, the bank estimated that its fund transfer bots alone were saving it $300,000 annually by cutting down the time its employees needed to spend on identifying and dealing with data mistakes and accelerating payments processing.

Capital One Financial has also developed a chatbot Eno – that can communicate with the bank’s customers via text message to give them information on their accounts and help them make credit-card payments from their smartphone.


This is the stage where the 10X experience is possible; it’s the most advanced embodiment of a chatbot. The difference is in the level of sophistication – both in the level of AI advancement and functionality.

While there are no examples of chatbots being at this stage in retail banking (with most stuck on the second stage in their evolution), robo-advisors can arguably be considered the closest this end of the continuum and the most advanced form of an AI-powered program – a bot. Encompassing automated investment/trading platforms (apps) and WealthTech, representatives are mostly either a tech startup or an investment bank with a reinvented/complementary wealth management offering.

With automated investment apps, advice itself is hidden in the algorithmic trading/investments – AI gathers information and takes action, leveraging changes in stock prices with the goal to make a dollar return. Robinhood could be one of the examples.

There is also another type of advice with an emphasis on wealth management and personal goals achievement. Schwab Intelligent Portfolios, for example, asks customers about personal goals and funds intended for investment, then determining the most appropriate structure of the investment portfolio. Advice, in this case, is hidden in optimizing the portfolio structure, guided by personal goals.