BACKGROUND IMAGE: Sergey Nivens/stock.adobe.com
When you got the email that your Lego Star Wars First Order Star Destroyer model had shipped, did you care who...
sent the message? Did you notice if it was a real guy named Brian or a chatbot who "talks" to hundreds of customers 24/7 and just happened to have adopted that name?
The companies turning to chatbots to enhance their customer service are betting the answer to both questions is "no." They say these computer programs save time, increase efficiency and reach their customers on more platforms than they could by using humans alone.
Better yet, the chatbots embedded with natural language processing (NLP) converse more casually, closer to what the guy named Brian would say -- the "natural language" piece of the tech puzzle. The NLP bots can detect the intent of what you type -- even if you misspell words -- use slang or speak Dutch (or one of the nine other languages that Facebook Messenger's 2.2 plug-in with built-in NLP can translate).
If you're one of the 1.3 billion monthly active users on Facebook Messenger, you've likely run across the 100,000 or so chatbots that share sports scores, the headlines and traffic updates and, increasingly, field simple queries from customers of B2B and B2C retailers and service companies who want to bypass the wait time on a telephone.
"Facebook Messenger is a massive trove of engaged users," said David Nelson, senior product manager for bots at HubSpot. "As trends change and users are turning to Messenger as a destination of choice to message a business or ask questions or seek resolution, we want them to be able to reach us in the way that's most convenient for them."
Facebook Messenger 2.2 opens the NLP bots floodgate
In addition to sheer scale, Messenger's 2.2 chat plug-in also supports payments via PayPal. Also, its conversation threads stick around and can travel between devices. A company's ubiquitous approach to improved CRM could also involve deploying an NLP bot on its own website, through email, maybe Slack, Twitter, Skype, WeChat -- wherever your customers tend to congregate.
Gartner predicted that, by 2020, 30% of our interactions with technology will be through "conversations" with smart machines. And with so many ways to interact, the investment in bots should focus on more than just the bottom line.
At HubSpot, chatbot technology and proven methodology drive "a delightful customer experience," Nelson said. "We're not putting a new shiny product in front of people to have to figure out."
HubSpot started using NLP bots on its own Facebook page about a year ago to nurture customers and gain new leads. The chatbots, Nelson said, helped promote e-books and other content by gathering information obtained in chat sessions. It then personalized its correspondence, updated contact records and followed up with suggestions and delivery.
Building on its internal success, HubSpot is preparing to launch Conversations in 2018. The new website-based tool will enable HubSpot's business customers to use live chat and tie all conversations back to the same page in the CRM, regardless of the channel it came in on.
What bots can do for business
One goal is to have NLP bots qualify leads before a salesperson jumps into the loop. A bot can be configured to know that a qualified lead is based in a certain geographical area, has 50 to 100 employees and falls within a range of revenue.
"The bot moves the threshold so a sales rep spends time closing a deal instead of sifting through information to see if the person is sales-certified," Nelson said.
Nelson said a common mistake is believing ROI should only calculate efficiencies. HubSpot focuses instead on how a bot can help its customers optimize their knowledge base and increase the efficiency of their growth stack. "Data is paramount," he said, "and so is being able to train our bots to allow customers to leverage their own data."
Natural language tech as proxy for human contact
Joe Lobo, botmaster with Inbenta, said your chatbot should convey the feelings people value when they interact with a company: a sense of worth and that they are individuals. That's where new natural language capabilities can help. Encounters should be casual in tone and able to handle colloquial phrases, context clues and keywords with different meanings.
For example, a greeting card retailer will encounter conversations with many instances of the word card -- some that refer to a credit card, debit card, birthday card or sympathy card. If the bot's NLP isn't worth its salt, you'll log off straight away and never look back, Lobo said.
Most first- and second-generation chatbots are static. They perform based on the scripts they are given. Scripts can take anywhere from an hour to five months to create -- or as long as it takes to brainstorm all the ways a customer can ask a question for all the possible scenarios.
But for certain things, people today prefer to feel like they're engaging with someone, especially when a contract or large-volume sale is on the line. Gary Gerber, head of product marketing for Conversica, said his company's clients "trust the AI to engage with their most precious resource: prospective customers." The work an NLP bot can do upfront is important but not as important as the creative work a human can do with the freed-up time to discuss product features or broaden and close deals.
Is it real, or is it an NLP bot?
Gary GerberHead of Product Marketing, Conversica
Conversica's AI sales assistant (there's also one for customer success) mines contacts for leads that are handed off to a human sales representative. Using NLP and, in some cases, natural language generation, the assistant emails according to a proven business cadence.
"Tons of leads remained undertouched, and opportunities were left on the table," Gerber said. "But it wasn't a good use of a person's time. Now, the AI assistant is able to have conversations at scale and find the ones who are interested in talking to a sales assistant. The beauty is the assistant is like [a] human. It feels like you're interacting with a human."
Joanne DeLangie agreed. She engages one of Conversica's AI sales assistants for Peak 10, where DeLangie is the VP of marketing and customer engagement. "Our assistant never gets that tone in her voice," she said. "And her persistence is incomparable." DeLangie noted: How many sales reps who get an out-of-office email would note the return date and schedule to reach out again after that? Or ask which phone number is the best to use?
Peak 10's assistant originally followed up on people who came in through the company website or by reading its syndicated content. Now, her duties have expanded to include people who attend events.
"She is personal, friendly, prompt, responsive and persistent," DeLangie said. "She re-engages in a way that no one else can. She follows up to see if someone got back to a contact, and we will know if the deal is at risk."
All the information is captured in Salesforce, DeLangie said, so trends are easy to establish. "We've had a 52% engagement rate on some campaigns if it's someone she follows." And since the assistant has a name, customers recognize her. If emails come from her, the open rate is significantly higher. "Some people will come to an event and ask if she's here because they want to speak to her," DeLangie said.
Getting a basic chatbot up and running can take a few hours to a few months, depending on the complexity of the scripting involved. With a machine learning model, the bot is trained on the job, with an existing history of calls and industry-specific lexicon. Once it's installed, it needs time to learn and start making predictions.
Peak 10's implementation took only hours, with a Salesforce administrator, a marketing rep and someone from customer success. No developers sat at the table.
Value vs. worth
"The investment decision wasn't a hard sell," DeLangie said. "The cost is nothing compared to what it would cost to do this with humans." She added that she can see with Salesforce how many opportunities her assistant engages -- and the status of those deals -- and it adds up to real dollars. "She's influenced more pipeline than it [has] cost for the tool."
There was some initial concern that some people would find out the assistant isn't real. But, DeLangie said, she's treated like a person and is "absolutely a part of the success with customers." They just have to look at their response rate. And they've started making other communications shorter and similar in tone and format.
Define the use case first, then program the chatbot
The best chatbots are those with the most clearly defined and narrowest use cases, said Mikhail Naumov, co-founder and chief strategy officer for DigitalGenius. The chatbot for a florist will understand if you want to buy a corsage or ship a large arrangement to an anniversary party. An airline, for example, will alert you upon check-in if your gate has changed.
With some slightly more complex applications, you can add more scripting and build a bot to plug in to your CRM to track packages. Big telecoms with thousands of agents around the world, for example, can create more efficiency in their contact centers by using chatbots, Naumov said. They can outsource the repetitive questions to a bot to reduce the amount of manual work a human does and make the business run more effectively.
Where AI, machine learning and deep learning are becoming important is by enabling companies to create projects that are trained on historical customer service data and chat logs. The bot can then suggest answers to a human agent once it's passed off the call. Over time, the algorithm learns from the human agent and transcripts, and these fourth-generation bots can automate responses.
"All conversations are supercritical, but it doesn't make sense for a person to do them," Gerber said. "The vast majority of times people interact with a bot they know it, and that's not necessarily a bad thing. It's like using an ATM instead of a teller."
ROI stories can be simple to gauge with a chatbot that tracks packages. A human would have to call the mailing company, and the phone call costs money. You can assign a dollar figure to the percentage of inquiries that are automated by an NLP bot instead of resolved with a phone call. Chatbots can also be configured to ask if they solved a problem, so success can be measured as well.
A bot with AI doesn't have to fit in to such a narrow box, said Naumov. Value can be measured in the increased efficiency of agents and the amount of time it takes to handle a job. Consider, he said, how you would measure the ROI of a calculator in an accounting firm or Excel at bank. "We are witnessing a transition where AI tools are table stakes for professionals in contact centers," Naumov said.