Customer service is one of the highest potential opportunities for bots, yet one of the hardest spaces to tackle. Inspired by a horrific support experience, Robert LoCascio founded LivePerson in 1995 to enable businesses to provide real-time customer engagement online. The company went public on the NASDAQ in 2000 and has since incorporated chatbot technology into their offerings.
We sat down with LoCascio to discuss the secrets behind successful customer service and how to strategically incorporate bots into the mix.
TOPBOTS: Thanks for being with us. LivePerson is one of the leaders in customer service platforms, and now you are augmenting your core offerings with bots. You have one of the front rows into this growing space. Can you start by describing how you see customer service bots?
RL: Customer engagement platforms have two bot flavors. The first is the one everybody is writing about. Customers talk with an automated bot from day one, and only with a bot. We however believe this is not as powerful as the second model — a bot assistant to the agent. We believe is that the integration between the agent and the bot is the best type of integration.
In the second agent and bot model, a lot happens behind the scenes. If an agent is messaging the consumer, having a bot there to assist makes the conversation more efficient. At the 2016 F8 Conference, Mark Zuckerberg showed the 1-800-Flowers bot on Facebook Messenger. That was us behind the scenes of the 1-800-Flowers bot. We are looking at ways to make human assistance more leverageable, we are building bots that can tell the agent how to respond best to customer inquiries.
These customer service agents are the front line of a company. If they do something bad, silly or annoying, then that reflects poorly on a company. Pure virtual agent bots, the first version we mentioned, see a very high error rate. Customer satisfactions is often below 70%, which is not what you want. Therefore, the best way to start today is the second hybrid model, and have the bot and the agent working in tandem.
TOPBOTS: Yes, definitely. We’ve all had experiences with automated systems that never seem to answer our questions. It’s super frustrated when customer service can’t answer basic questions and become broken records.
RL: That’s correct, and the biggest challenge with front-end pure virtual bots is that the people making these bots do not understand enterprise customers. They tend to script the bots so it’s not even AI. They also don’t hook the bot into back-end systems such as bill pay. Bill pay accounts to 30-40% of interactions of calls or chats to a brand. If a bot cannot resolve bill pay questions, then it’s half useless. These bots are like bad FAQs: when I say “I want to pay my bill” or something, they don’t do anything. It’s important for the bot to tie into back end enterprise databases and systems.
However, there are many functions corporate tech systems do not support today. Most enterprise systems don’t have a good APIs so we could not automate everything even if we wanted to. There are legacy systems. That’s why having the human there is helpful.
Another interesting fact is that in a recent study of US Consumers, 80% of respondents wanted to be told clearly up front if it’s a bot or human that they are speaking. Secondly, 60% assume that if they are talking to bots, it’s because the company is trying to save cost at their expense.
TOPBOTS: Very interesting statistic. Good to know to not make a bot pretend that they are human! If a company does want to build a bot that connects with the back end system, how do you do it?
RL: In our platform, we have a transfer feature. We’re hooked to Facebook as a back-end, we’re hooked to Google now, which is Google does messaging within surge. And we also do in-app of our customers, so we’re really proficient in understanding the enterprise and messaging when it comes to the agent, the agent has a way in their hand to transfer.
The agent notifies the consumer “Hey, I’m transferring you to this bot and they’re going to handle your bill pay.” Now, the agent is still there and can still see what’s happening, and if the consumer says, “I hate you. It’s a terrible… ” there’s something we do. We don’t need the customer to hit “0” a bunch of times before realizing that it was a poor transfer. We have built into the system we are looking at how the conversation is going. The live human agent can jump back in if needed instantly.
TOPBOTS: Do you have any metrics so far of how the agents plus bot model is performing versus a human agent alone?
RL: The agent and bot version sees about 30 to 35% more efficiency than a solo agent as of today. We also have this metric called “Meaningful Connection Score” (MCS) that monitors the bot and the health of the consumer, and we balance automation with live.
TOPBOTS: What are typical ranges for customer service MCS scores?
RL: In traditional voice contacts, MCSs are in the low 80%, in chat, they are in the mid 80s%, and in asynchronous messaging, we see high 80% to low 90%. Messaging is by far a much more preferred way to engage because it’s on the consumer time. So messaging in the next 5 years would have every enterprise opt-in, it will be a standard and I think you’ll see the death of voice.
TOPBOTS: In a typical agent augmented by bot model, what percentage of questions are answered by the agent vs the bot? Is there a balance you are trying to strike?
RL: The bot ratio starts out fairly low, and then it ramps over time. The 80-20 rule is a good way to start. Long term, I believe we can move 60% of interaction towards automation, but we need the back-end systems of companies to be upgraded to support the change.
TOPBOTS: That’s a lot of automation! Are there any examples of projects you’re working on currently you can share?
RL: Right now, we are working with Royal Bank of Scotland (RBS), we did integration with IBM Watson and with a company in Israel. What we are trying to do is really change the outcome to millions of interactions that are happening in the contact center. The staff and I are working on how do we change five or ten million interactions to something that can ride upon automation.
TOPBOTS: Thank you for your time. It was very informative to learn about about chat and customer service bot models. I know as a consumer, I may have been interacting with a human assisted bot agent and never have known!