Yuriy Oparenko, Product Designer at Intercom, developed his first chatbot in 2015. It was for a UK bank that wanted to jump on the conversational AI bandwagon, as did many companies at the time. Live chat support paired with AI was expected to be the next big thing in customer service. But it wasn’t. Oparenko explains:
They were meant to be the new websites. They were meant to kill 99% of the apps. [But] the conversational revolution didn’t happen after all. Conversational UIs haven’t replaced websites or killed apps. Most chatbots out there are not even distantly intelligent.
Companies looked at chat as the quick fix, the cheapest way to solve customer interactions. They didn’t think about how to do it properly, what they could layer on top of live chat, what bots were or weren’t good for. Instead, they immediately began implementing chatbots, skipping the necessary step of mastering chat with a human agent first.
The AI available at the moment isn’t sufficiently powerful to implement good chatbots for all types of interactions with customers yet. Emphasis on yet. We’ll get there, just not now.
Expectation vs. reality
I think one of the biggest pitfalls with companies’ using chatbots was only seeing one side of the coin. They read statistics that showed CSAT was higher for chat than for emails. Having a chatbot in place was also cheaper than having agents replying to emails. It looked like chat was going to solve a lot of companies’ problems. But moving to chat doesn’t mean moving immediately to chatbots.
When you engage with a company through chat, you don’t want to come across an algorithm trying to guess what it is that you want from them. It’s like going on a date and, instead of having light, intelligent conversation, it’s just someone trying to guess what you really want to engage with. It’s never going to work. As a human, you’re able to figure out if you’re interacting with a bot by testing whether it provides the same answer when you ask the same question over and over again. It’s frustrating and it’s not tailored to a specific person with a specific problem. As a customer, you feel like you’re just one a cog in the quarterly earnings machine, beholden to a company that doesn’t care enough to provide remotely personal service.
What we do expect from live chat is to have someone on the other side who’s able to sustain a conversation. We are social beings, so we’re all about social reward. We get that in chat when we see that digital caterpillar inching along, three dots looping until the other person’s sent their reply. It’s this social dimension that makes us really connect with a chat tool and expect the interaction to be fluid, like an actual conversation. If, as a customer, I get the sense that there’s an agent on the other side who is paying attention to me, my trust levels in this company or brand increase.
The chat experience becomes even better if I can communicate in my native language, without having to worry whether I asked the right questions or if the person on the other side understood exactly what my problem was. This can become a challenge for companies, since hiring agents covering chat channels in all languages is very difficult, both from a personnel and financial perspective.
Real time chat translation could be the answer, but is it possible?
Instant chat translation
Real time chat translation is a challenge. We know that to be successful, agents have a 50 second mark to reply to a customer. That means that, in those 50 seconds, the agents need to write something, Unbabel needs to translate it and send it back to the agent who sends it to the customer. If it’s a longer reply we have more time, and customers understand that. Even as a human I tend to spend more time typing if I need to construct a longer answer, which might take more than a minute to write. That’s understandable.
But for a normal chat interaction, we’re talking about an average of 50-55 words. 70 words is already considered a long message. What we have managed to do is to translate the average sized texts in near real time. But, that also means that we have made a huge effort on our side to make sure that we have the right training algorithms, the right interactions, and the right feedback loops to get learnings for our chat models. This way, they are able to learn and react as quickly as possible. This is important because there’s no time for someone on the other side to be correcting the texts or even translating them in real time. So we need to be able to trust our AI to enable that near real time translation of chat.
This is where we stand right now, and where we are seeing some great results. How do we know that this is working? We can see it on the success we are having with the clients with whom we have been using this approach. For me, the biggest success metric is the CSAT score of our clients. What we’ve observed is that our customers who have been using our chat product have registered CSAT scores that are consistently either equal or even a little bit better than the native speakers’, which is really cool.
When we get to this level, it means that we are definitely doing something right, because, for me, measuring quality in chat is always subjective. Having these hight CSAT scores tells you something about how customers felt. Were they engaged in the conversation? Did they feel that the agent provided the right answer? And, on top of that, did they love being able to interact in their own language?
Localization is key
In this day and age, companies cannot afford to offer support that is not localized, especially in parts of the World where so many people speak a certain language that it doesn’t even make sense for them to speak English. In India, for example, even though people speak English, if we take a closer look we’ll see that Hindu or Parsh are the most spoken languages. If you go to China, good luck getting your way with speaking anything other than Chinese.
When you want to reach out to customers and you want to be global, you need to be ready to provide a localized experience. Either you hire people left and right in every single country who speak every single language and are able to provide 24/7 support, or you can have that same thing being done by a multilingual support solution like ours.
At Unbabel, we have what we call the easier languages and the hardest languages to translate. There are some languages that are a bit more challenging to do than others. What we are trying to do is to mimic exactly the work we’ve done for our more successful languages, to try to copy and leverage it into the more difficult languages. We are working on things such as understanding how we can improve the feedback loop to make it quicker and more effective. This way, we can open new languages or model existing ones to be adapted to a specific customer. Eventually we will get to the point where the model knows how to speak that customer’s language without having to go through too many learning steps in between.
For us, chat is definitely a big part of the future of customer service. Even though predictions thought chat would be fully up and running by 2019, I think the decrease in its usage is due mostly to companies that tried to force the use of chatbots instead of looking at chat as an opportunity to get the load off other channels such as voice channels, but with the help of a human. But I also think that businesses will revisit this soon enough and understand that, while chatbots are not ready yet to provide a seamless experience, we can trust humans with AI-enhanced language skills to be on the other side ready to reply.
The future lies in live chat
For some things, customers already expect to have a chatbot in place. For example, if they’re asking a standard question like, “What’s the address for your office?” That is information that can be easily located, so for a question like that, mechanical answers are perfectly fine. For other issues, especially if customers are complaining about something, they don’t want to encounter a machine on the other side of the conversation.
Others will never interact with a chat tool altogether. More than a personal thing, it’s a generational thing. My parents will never engage via chat, human or bot, because it’s something so foreign to them that they will always opt for voice assistance.
In younger generations we have a mix. There are people who prefer to call a company because they feel it’s the channel that will provide the quickest and most final solution to their problem. Other people, especially millennials or even younger than millennials, feel more comfortable having these interactions through a digital channel because it’s what they use to communicate in their personal lives. They don’t seem to care if it’s a person on the other side or not, as long as they find a solution for their problem.
If we look at the people who are more engaged with brands, however, my two cents are that chat is going to be one of the channels that, if companies play it right, will be the future for most businesses out there. We are no longer engaging with brands unilaterally, through the commercials we see on TV. As customers, we are now able to reach out to a brand through social media and chat channels and be the ones to establish that first contact, which didn’t happen before. Even the notions of time and expectation are completely different. No one wants to wait anymore. If I wait for two days, or even a couple of hours, for a reply, it’s way too long.
Chat will be the sweet spot for most interactions, because we actually see the difference in how people interact and reach out to brands. Customers get a channel where they feel they don’t expose themselves too much, which might happen over the phone, but it’s interactive enough to make them feel like they can say what they have to say and get an answer right away. It represents the perfect balance between voice and email.
The next time you think about chat, don’t just think about chatbots, think about your customers on the other side of the screen. Engaging with them in the right way can save you dollars on gaining new customers as you will keep the ones you have by avoiding churn. Companies need to think about the people and how to make their experience better. Bots may help, but they need to be part of a bigger, bolder plan.