Chatbots can be an effective way to reduce workloads across customer service teams. But not all chatbots are created equally.
Chatbots can greatly benefit the customer experience. When met with high workloads in customer service departments, chatbots can act as a first line, quickly answer questions from customers to get them on their way. These solutions can bring massive efficiency improvements and cost reductions. UK-based Breckland council was able to handle 80 percent of incoming customer questions and US-based insured Humana reduced operational costs by 30 percent.
Losing customers through poor customer service
A December 2021 survey published by Qualtrics, in collaboration with ServiceNow, showed that 80 percent of customers switched brands after experiencing a poor customer experience, with 43 percent saying they would consider switching to another brand when experiencing one negative customer experience.
The survey conducted among 1,094 consumers 18 and over from the United States who had interacted with a customer service department found that women were 15 percent more likely to observe that customer service experience had deteriorated over the course of the pandemic. Switching behavior differs per generation, with Gen Z being more likely to switch brands when encountering issues during the online shopping stage or problems with the website.
Boomers and Gen X were found to be more forgiving when it came down to poor digital experience. However, older generations were more likely to drop a provider when encountering defects with the product or poor customer service. The survey showed that respondents contributed worse customer service due to labor shortages, which had a negative ripple effect on customer service departments to adequately provide support to customers. Customers who indicated having a positive customer service experience had their issues quickly resolved, low hold times and the ability to speak to one agent.
It won’t come as a surprise that companies who failed to get customers’ issues resolved in a timely manner, whilst cycling through multiple agents, received the lowest scores. The Qualtrics survey revealed that only 34 percent of customers felt they were treated with empathy when interacting with a customer service representative. The Qualtrics survey showed how brittle the relationship between customers can be and the necessity to ensure steady operations at customer service departments. However, when short staffed this might prove to be challenging.
Consumers wary of chatbots
Chatbots can be a tempting alternative for organizations to deliver quick customer support when having trouble finding enough staff or trying to cut costs. However, many consumers will be wary when it comes down to receiving support from chatbots. Figures cited by Statista related to U.S consumer satisfaction in regards to chatbots during customer service interactions, paints an ambiguous picture.
The figures show that a majority of U.S. based consumers had a negative experience with a chatbot. While the majority, 32 percent, left the chatbot interaction with a neutral feeling, 22 percent were found to be unhappy with 21 percent being not happy at all. Only a minor share of consumers left with a positive experience, with 14 percent being happy and 11 percent being very happy.
These findings show that chatbots aren’t one-size-fits all and can do more harm than good. Considering consumer sensitivity to poor customer service experiences, a poorly executed, unsupervised chatbot can bring great damage to a brand’s reputation. This begs the question whether a company should consider a chatbot in the first place, as a low commitment to proper process integration can lead to disastrous outcomes.
To chatbot, or not to chatbot
Consumer facing companies will be tempted to consider chatbots when the going gets tough. This can occur when the customer service department is understaffed or sudden peaks in customer cases due to problems can put immense pressure on the existing workforce. These moments result in executives across the company to explore other avenues to maintain a semblance of good customer service.
However, the business case to opt for a chatbot might not be as strong as many vendors will make you believe. Co-founder and CEO of the San Jose based [24]7.ai, P.V. Kannan and author or coauthor of six business strategy books, Josh Bernoff, argued that chatbots aren’t always fit for the use case at hand. In May 2019 Harvard Business Review piece, both warned that executives might be taken up in the A.I. hype, but not all virtual agents, are created equal.
They point out that most chatbot integrations start from a cost reduction standpoint. A fairly limited use case for virtual agents, who can capture more incoming requests than initially anticipated. The piece highlights that virtual agents are available 24/7, with an uncanny ability to quickly solve incoming customer questions and requests. They cite American car rental Avis Budget, who was able to automate 68 percent of incoming service cars when rolling out virtual agents.
Dish Network, a U.S. satellite television operator, found great success with their chatbot, seeing customers rate their virtual agents on par with human customer service representatives. As time progressed Dish Network saw scores improve, as the chatbots became more efficient, handling incoming questions more effectively. However, as tempting as these results may be, not all virtual agents are created equal nor do they deliver the same results.
Kannan and Bernoff explained that the costs of implementing a chatbot might outweigh the costs of hiring agents. Implementations of virtual agents are the most cost-effective solutions at organizations that handle thousands of incoming customer requests that are in turn handled by hundreds of customer service representatives. Smaller organizations might find themselves with chatbot ill equipped to handle customer questions as they don’t have a great enough backlog to serve as training data. This is especially true for artificial intelligence solutions that don’t have predefined texts embedded.
Additionally, the development, implementation and optimization of virtual agents is a costly and arduous process. The costs don’t way up against the benefits a chatbot will bring to the company. Kannan and Bernoff argue that such lengthy projects only make sense when millions are allocated for daily customer service operations. Furthermore, as pointed out earlier, automated solutions are oftentimes a spur of the moment decision. As the rough seas at customer service settle down, a chat becomes less of an urgent need.
Common chatbot mistakes
In a 2017 opinion piece at CIO Magazine, Dave Smith, noted that companies seeking a chatbot enterprise solution should develop a framework to create a viable business case. A clear purpose needs to be defined for the chatbot, as without one, the return on investment will remain unclear for internal stakeholders. Measuring tools and their subsequent metrics should be put in place to quantify and optimize the contributions of the newly integrated system.
Enterprise planners, who are tasked to alight the business needs of all departments within an organization, will have to ask critical questions, such as how the chatbots will affect and improve the customer, partner and internal employee experience. What business processes will have to be aligned, replaced or improved? What systems will have to be connected to the chatbot and who will deliver and secure the maintenance?
These questions are critical to prevent the project from flatlining once it’s up and running. In March 2020, Managing director at Accenture Interactive, Rob Harles told Clint Boulton at CIO Magazine, that chatbot integrations can be treated as a rush job, solely because they are seen as the cool new thing, with the sole purpose to avoid dealing with customers. Instead, executives who consider rolling out a chatbot should first thoroughly understand customer pain points and find the appropriate solutions first.
Charles and his team have had to overcome many hurdles, co-developing a Facebook Messenger chatbot, Carla, for Colombia’s national airline Avianca. The chatbot enabled customers to check their flight status, view weather forecasts, receive luggage updates and other flight-related options. The chatbot is able to handle multiple requests at a time through support from Amazon Web Services, keeping the tool operational during peak hours. Since its launch in 2016, Carla has processed tens of thousands of incoming requests and handled millions of interactions.
Such complex processes aren’t easily implemented, requiring countless hours to design and build the conversations, having access to internal systems to provide relevant information. Customers are also unpredictable, jumping from one topic to another. The chatbot has to handle these unforeseen circumstances and learn from these interactions, whilst maintaining a reliable connection between its internal systems and the customer.
Monitoring dishonesty
We must also not forget that customers can lie to your chatbot, a phenomenon raised by Assistant Professor at the University of Michigan, School of Information, Alain Cohn, in May 2022. Cohn points out dishonesty isn’t a new phenomenon, but with the roll-out of chatbots, this becomes more prevalent across customer service interactions. The oftentimes unsupervised nature of many chatbot interactions can leave a blindspot in regards to unwanted customer behavior.
Through a coin-flip experiment, Cohn and her colleagues wanted to measure the amount of instants where customers would lie about the results. In one instance via video call or chat with a researcher, whilst another was unsupervised with results being confirmed by the participant to a voice assistant bot. The coins were flipped in private to prevent researchers from knowing whether the participant was lying.
The results showed that 54.4 percent of participants reported a successful coinflip to a human, resulting in a cheat rate of about 9 percent. The cheat rate increased substantially when the participants reported their results to the voice assistant bot, with the rate increasing to 22 percent. This translates to participants being twice as likely to report a false result to a virtual referee compared to a human.
This phenomenon could be attributed to reputational damage when reporting to a human, Cohn and the team found through a follow-up survey examining how they viewed the researcher versus the virtual assistant. In an attempt to reduce the cheat rate by anthropomorphizing the virtual chat. This however did not yield the desired result, with lying remaining just as prevalent.
The lying phenomenon brings an additional dynamic to chatbots that need to be overcome.If left unchecked, this could lead to undesirable outcomes. Overcoming this hurdle however, is challenging as there’s no easy fix, Cohn argues. Systems can be put in place to identify these cohorts, but once they have discovered the tricks, they might switch back to a human agent. Hence, Cohn commented, digital dishonesty will also be present.
Fraud prevention through chatbots
Despite the gloomy reminder, there are avenues that companies can take to prevent misuse of its virtual assistants. In a December 2023 blog, head of global business development at Nvidia, Kevin Levitt pointed out that generative AI can create chatbots that are more adept at detecting and improving fraud detection. Large language model (LLM) assistants who use Retrieval augmented generation (RAL) on the backend can use data detailing policies and base their decisions making on set rules.
Levitt highlights that LLMs are already used for customer transaction prediction, helping organizations detect whether a fraudulent transaction is bound to happen. As LLMs are becoming better at detecting such instances, they offload risk assessment teams and prevent compliance misuse. Financial services benefit greatly from these models, Levitt notes. The intelligent systems can omit rigid frameworks that harm the experience for regular customers, who can, unjustifiably, generate a false positive and come under investigation due to the rigid frameworks put in place by these financial providers.
AI powered systems meanwhile can tap into greater pools of data and streamline the customer experience. Admittedly these systems are far out of reach for the small to medium enterprise today, but show how chatbots and other language based models can prevent chatbot scams and prevent further financial damage. If chatbots and AI have been implemented correctly however, they can already bring greater efficiency to organizations.
Chatbots deliver impressive results
Chatbots can deliver impressive results when the business case is sound and the initial hurdles are overcome. Humana, one of the largest Medicare insurance providers in the United States, drastically improved its customer service experience whilst significantly reducing incoming calls by adopting IBM’s conversational AI solution, Watson. IBM notes that Humana delivers its services to about 13 million customers across the country.
These customers have complex questions related to healthcare, requiring providers to constantly adjust and innovate their offering. This time consuming endeavor puts immense pressure on Humana’s customer service agents, who have to deliver accurate solutions within an acceptable time frame. Humana relied on its IVR system, which failed to properly handle calls, resulting in high amounts of those being redirected to the customer service team.
Humana received an upward of one million IVR calls with the majority being rerouted to outsourced call centers, resulting in high overhead costs for the insurer. The company found that over 60 percent of those calls were recurring questions that could be mitigated through pre-service questions IBM noted. IBM roll-outed Watson in April 2019, with regular updates to optimize and expand the product’s functionalities. The virtual agent became more apt to answer customer questions as custom data was being added and the solution became more accustomed to healthcare terminology.
Humana was able to reduce service costs by around 30 percent, compared to its previous setup. Simultaneously, the company doubled the overall response rate. Director of Provider Experience and Connectivity at Humana, Sara Hines, commented that it took three years to optimize Watson, with the system becoming better over time, helping Humana to improve the provider communication. Watson now handles over 7,000 voice calls per day from 120 providers.
Camping World improves response times
In a candid customer case, Senior Vice President (SVP) of Sales and Customer Experience at Camping World, Brenda Wintrow, admitted that the response times at the company’s customer service were concerning, who decided to add IBM’s Watson solution to its customer care operations. Camping World relied heavily on its customer service department to deliver a strong experience, IBM noted. But as the world over the course of COVID-19 pandemic, this strong core started to display cracks, negatively impacting the customer experience.
The complex nature of the company’s business, which operates across three axes, namely retail, insurance and dealership, make it difficult for Camping World to assign agents to cover all three business units. Despite having a decently sized customer care team, Wintrow commented, it’s unrealistic to assign one agent to all three of the operations at the company. This makes staffing a difficult endeavor, she adds.
These shortcomings become especially apparent during seasonal surges, which put additional pressure on customer facing teams. Even in off-seasons, maintaining steady operations across its 24/7 call center, proves to be challenging. A solution had to be found to reduce response times and deliver relevant information to customers. In order to offload its customer service department, Camping World opted for a human-centered solution through IBM’s LivePerson technology, aptly named Arvee.
The virtual agent could be deployed across all the company’s web properties. As with many tools, human agents can take over if the conversation becomes too complex. The team started out with 75 to 100 intents and could be adjusted as time progressed to create a more tailored experience for the customers. SMS-functionalities were added to allow customers to switch from voice to SMS during calls.
As Arvee was starting to manage incoming customer requests, wait times decreased significantly, with agents’ efficiency improved by 33 percent as they couldn’t now handle multiple chats. Of the 13,999 chat conversations, 6,000 needed to be transferred to a live agent, Wintrow pointed out. Wait times dropped to 33 seconds with customer engagement increasing by 40 percent.
Reducing incoming calls
UK-based Breckland Council, managing 12 electoral divisions, opted for Ubisend’s solution to provide a speedy service to the population of Breckland. Only two staff members handled the over 9,000 incoming customer requests through chat and email, having trouble responding in time. This resulted in high workloads for the team, who helped citizens with their bills, ordering new bins and everything in between.
The council was in desperate need of a solution that could handle these recurring themes and offload its small team. Adele Newsome, Customer experience manager, set an ambitious target for the team, the efficiency needed to be improved by 90 percent over a period of four years. In order to achieve this goal, Newsome looked for a solution that could reduce the number of inbound inquiries.
Before building a full-fledged virtual agent, the team at Breckland first created an automated FAQ through Ubisend’s chat, which Breckland gave the humble name, Bobbie. The chatbot could handle easy, recurring questions, such as how to pay council tax, how to request a taxi license and hundreds of alternate variations. Through this simple integration, Bobbie was able to handle 80 percent of incoming questions, with only a minor 20 percent flowing to the web team.
Optimizing flights at KLM
In 2016, Dutch airline KLM started to explore new ways of delivering a conversational experience to its customers. KLM opted for Google’s Dialogflow solution after testing multiple platforms. In September 2017, KLM launched its chatbot, BB, short for BlueBot, representing the airline’s signature blue colorway. The booking bot started as a Facebook Messenger bot where customers could ask BB details about their destination, available flights and provide booking confirmations.
KLM director of social media, Martine van der Lee, explained Google’s Dialogflow solution housed strong natural language understanding, allowing for KLM to automate large sections of the conversation. In the press release, Senior Vice President Digital Air France – KLM, Peter Groeneveld commented that KLM handled 16,000 incoming requests a week, handled by its 250 human agents. As incoming requests were growing, the team at KLM sought for ways to deliver the same high level of service, without burdening its agents. BB was the next logical step for the airline.
In December 2017, KLM extended BB’s reach by integrating it with the Google Assistant. BB could inform customers about visa arrangements, medicine reminders, everything required for a safe and carefree journey. KLM was one of the pioneers of Google’s Dialogflow solution. Apart from it being an obvious brand awareness strategy for KLM who had been pushing guerilla social media strategies in years prior, from a customer service perspective, ensuring customers are well prepared reduces friction during security and boarding, helping KLM run on schedule and reduce incoming, last minute, requests to its customers service.
Paving the way to success
Poorly executed chatbots can bring more harm than. When not having the conversational abilities or systems access to properly reply and solve a customers’ needs, they can ruin the overall customer experience, resulting in negative spillover. May it be to public channels or to the human agents when available. Organizations will first and mostly benefit from identifying pain points and optimizing their internal systems and processes. A lesson United Airlines learned the hard way.
It took away the barriers that allowed problem solving for its customer facing employees, helping the troubled airline reclaim some of its deteriorated reputation. However, processes can only be smoothed out so much until customers get stuck somewhere along the way. This is where chatbots can become a nifty tool to offload the oftentimes understaffed and stressed customer service agents.
The process to build such a strong and independent tool requires a lot of upfront investments and therefore only make sense for corporations who have high customer service overhead costs. However, cost savings, whilst an important argument, should not be the primary driver to rollout a chatbot. A chatbot should act as an extension to the customer service to help the customer faster and accurately.
We are still a long way away from chatbots and AI-powered systems replacing human agents. However, the early signs for massive job displacement are already there. As these systems become more accurate through accumulated knowledge, we might see a tipping point where organizations greatly optimize their operations, whilst delivering stellar customer service.
