Bots Are Good, but They Can’t Do It Alone
Driven by accelerating adoption in the customer service arena, the worldwide chatbot market is expected to more than triple, from $2.9 billion in 2020 to more than $10.5 billion in 2006, expanding at a compound annual growth rate of 23.5 percent, according to research firm MarketsandMarkets.
A big reason for this surging popularity is that customer service chatbots continue to get better due to the introduction of machine learning and artificial intelligence, says Fergal Reid, director of machine learning at Intercom, a provider of customer communications platforms.
“They’re getting better at understanding the question that they’re being asked, and they’re getting that wrong less often,” Reid says. “We’re also seeing an increase in their multilingual capabilities. There’s also very specific ways that they’re improving in terms of functionality. People are building bots to handle more and more complex flows and more and more complex questions.”
At Pegasystems, for example, chatbots have evolved to where they have the intelligence to use context to more accurately answer a query in real time, says Fortune Alexander, Pegasystems’ senior director of product marketing for customer service and sales automation. “That evolution continues. This whole notion of conversational AI is coming to life and to bots.”
Another recent advancement in customer service chatbots is the increasing use of natural language processing to understand the same queries asked in a variety of ways, according to Alexander. “That reduces the agent’s interaction time and makes it more fluid when chatting back and forth, providing a faster resolution. That saves time, and customers are happier because they have gotten their business taken care of and they can be on their way.”
Raghu Ravinutala, CEO and cofounder of Yellow.ai, says the most impactful recent evolution of customer service chatbots is their ability to automate workflows from digital conversations.
“The customer is not just getting automated responses, but also settlement,” Ravinutala says. Today, when a customer calls about rescheduling a flight, the airline’s customer service bot fulfills the request, changing from the original flight to the rescheduled one and issues an electronic boarding pass, all without human intervention.
Experts largely agree that customer service chatbots are popular for a couple of reasons, including the following:
- They can alleviate contact center staff from answering the mundane, oft-asked questions, like “What’s my balance,” allowing them to move on to more complex customer service issues.
- Contact centers can’t effectively scale human agents to handle peak times, an issue that became evident during the early days of the COVID-19 pandemic.
- In regulated industries, they can ensure that the customer service agent adheres to the script, delivering the required payment and interest rate disclosures for banks, for example.
Companies’ successful use of customer service chatbots during the pandemic has prompted more interest among organizations that were slow to adopt the technology initially, according to Alexander. “We see a whole lot of interest in how you can use bots to be more efficient in the contact center,” she says.
“Chatbots should be used to automate resolutions to repeatable issues,” says Puneet Mehta, founder and CEO of Netomi, a provider of artificial intelligence for customer service. “Typically, anywhere from 40 percent to 80 percent of a company’s customer service tickets span the same use cases. These use cases typically have a lot of historical data that can be used to train the AI and ensure it can accurately understand the user’s intent and provide the correct response.”
This is important, Mehta says, because people can ask the same question in many different words. Comcast, for example, found that its customers asked the simple question “I want to see my bill” in 7,500 different ways. By leveraging past tickets to train the chatbot, it will have much higher confidence when classifying the intent of an interaction.
“Trained to answer hundreds of frequently asked questions, like baggage policies, upgrade availability, or how to fly with a wedding dress, WestJet’s AI-powered chatbot Juliet has automated up to 87 percent of tickets,” Mehta adds.
But that doesn’t happen automatically. Companies need to provide excellent support to derive the greatest benefits from customer service chatbots, Reid says. “A lot of our customers are staffing specialist roles to design and manage their bots. There’s also an emerging skill set in customer support to understand what the bots do well, what they actually understand, and what they do poorly.”
Intercom’s more advanced customers often employ people just to monitor chatbots’ incorrect answers to train them to improve responses, Reid adds. “There are some domains where it requires deep human judgment to actually understand what a question is about, and bots are not yet able to do that. We’ve seen a lot of advances in the technology in machine learning. We’re getting better at understanding context, but we’re not quite there yet.”
Reid, therefore, recommends that individual businesses make realistic assessments of what chatbots can do for them by examining recent customer support conversations to determine which ones are most frequent, easily answered, moderately complex, and require human intervention.
Alexander also notes that the situation with customer service chatbots vs. customer service agents isn’t either-or. “We see these chatbots becoming co-pilots. When an agent is involved, the chatbot is right there being the note taker or transcriber, offering suggestions [to the agent] in real time about what to do next.”
Such a tandem relationship enables the live agent to talk to customers and understand their issues while the chatbot fills in any necessary details, speeding the entire interaction along to resolution.
Ravinutala agrees. “Companies have realized that great customer experience needs a combination of automation and humans. The companies providing the seamless combination of humans and AI are the ones providing great customer experience.”
CURRENT CHATBOT USES
This is true for many chatbot users. “We see customers using [Intercom chatbots] for very real use cases and getting a lot of value out of these products,” Reid adds.
One such company is European logistics company Stuart, which uses its Intercom chatbots for courier interactions. Every week, those bots handle more than 55,000 chats, reducing response times to less than 30 seconds, reaching resolutions four times faster, and saving the company more than 2,500 support team hours, according to Reid.
A Pegasystems customer, a large insurance company, has deployed chatbots to handle the queries with the highest volume and lowest complexity, according to Alexander. “You don’t have [callers] being told about long wait times, you are able to resolve issues right there. In that case, it’s using bots connected to a smaller, network-based IVR.”
Another Pegasystems customer, a bank in France, uses chatbots to answer more than 3 million emails each year, Alexander adds.
CURRENT CHATBOT LIMITATIONS
But Sean Drummy, vice president of product management at Vee24, a provider of customer service platforms that combine humans and automation, points out that customer service chatbots don’t work well for certain types of businesses, like high-end luxury goods. “I spend $10,000 on a watch and you’re going to give me to a chatbot? There is no way that is going to be palatable.”
Another area where customer service chatbots have yet to live up to their promise is personalization, according to Drummy. “Even in customer service, chatbots are highly transactional. In the retail space, personalization is the key. While you can build that in with chatbots today, it’s pretty expensive. People are still figuring out how to aggregate all of the data [for personalization].”
While some more sophisticated chatbots include sentiment analysis, they still have trouble providing emotionally sensitive responses, Ravinutala adds.
As chatbot technology continues to develop, Drummy expects that over the next few years one of the biggest areas of concentration will be on interaction continuity, with more seamless handoffs between the chatbots and live agents when needed. While technology can pass on information from the chatbot to the live agent, all too often customers need to repeat the information they already gave to the chatbot when the live agent comes on the line.
“Nobody wants to have that dropped handoff,” Drummy says, adding that integration of customer service channels is low-hanging fruit.
Drummy also expects the chatbots of the near future to be able to better handle customer fulfillment data, like order status and location within the supply chain.
The underlying machine learning and AI will continue to advance, Reid predicts, noting that this will, in turn, advance the variety and complexity of customer service queries that chatbots can answer accurately.
“You’re going to see deeper and deeper integration of these customer service bots with business platforms,” Reid says. “That will enable them to perform more tasks in response to the customer query. Today the focus is primarily on information retrieval.”
As chatbots evolve, expect them to also gain more dynamic functionality that will allow them to understand context and alter their responses so they can accomplish more without human intervention, Reid continues.
That evolution will include better and more advanced use of conversational AI, Alexander says. “That will let agents have that hands-free experience so that they can have the best dialogue with the customer.”
Ravinutala expects marketing as well as e-commerce capabilities to also be incorporated into the customer service chatbots of the future.
“Companies need to choose their solutions wisely. Chatbots can do this today; this isn’t all just in the future,” Alexander concludes.
The Six Biggest Mistakes with Customer Service Chatbots
Chatbots have been largely successful where they have been used so far, but there have been some missteps along the way. Below are some of the biggest mistakes that companies need to avoid:
- Forced chatbot usage.
Companies shouldn’t automatically push every interaction to the chatbot.
Sean Drummy, vice president of product management at Vee24, recommends that companies quickly push any queries not in the top tier handled by chatbots to human agents. If, for example, a furniture company’s delivery truck runs into a customer’s garage door during a delivery, resolving the issue will be beyond a chatbot’s capabilities. The company’s insurance carrier might use a chatbot to handle the initial basics (name of driver, type of accident, where it happened, etc.) before handing the call off to a human agent.
“For customer service chatbots to be successful, they have to unapologetically admit when they don’t have the information and immediately get you the help you need from a human agent,” Drummy says. “If someone wants to talk to an agent, give them an early out, don’t force them to use the chatbot, and then satisfaction levels will increase. Don’t offer a chatbot-only solution, always have agents to back up the chatbot. If you don’t have agents behind it, it’s never going to work.”
- Buying a stand-alone point solution.
“You have to connect [your chatbots] to a central brain and artificial intelligence,” advises Fortune Alexander, Pegasystems’ senior director of product marketing for customer service and sales automation. “At Pega, all of the customer service information and intelligence is in one place, regardless of which channel you’re coming in from. It’s all connected to the same central logic.”
- Trying to be all things to all people.
To optimize the performance and benefits derived from a customer service chatbot, use it to answer the top handful of questions that will make up as much as 80 percent of the queries, Drummy says. “Focus all of your energy on going deep into those use cases, not trying to be everything to everybody. Don’t try to handle 200 cases in a shallow fashion.”
Doing so would be nothing more than a glorified FAQ answer sheet, Drummy adds.
- Including sales pitches.
While chatbots can sell by offering different product choices, options, etc., this is beyond the scope of many customer service chatbots and is better left to chatbots designed strictly for this purpose, according to Drummy. Such chatbots currently require heavy investments in AI and machine learning, he explains, noting that chatbots of the future will likely be better able to handle both sales and customer service transactions.
- Ignoring support/maintenance.
Chatbots are not something that you can just set up and then leave alone. “It’s not unlike any complex piece of technology. You have to continue to review it, not only the data that’s coming in, but your approach to the chatbot. You have to be reactive to what you’re seeing out there,” Drummy says.
Failure to continuously train a chatbot will mean performance won’t improve, notes Fergal Reid, director of machine learning at Intercom, adding that a trained, seasoned chatbot should perform better than one recently deployed.
“We always encourage people to make bots that will do the top things very well and be smart enough to realize when they cannot help the user with a particular question they’re asking about,” Reid says, echoing the recommendation for a quick handoff to a human agent when that occurs.
Phillip Britt is a freelance writer based in the Chicago area. He can be reached at firstname.lastname@example.org.