On Nov 30th, I was invited to CXxAI — an interactive roundtable for Customer Experience leaders in enterprise organizations. During the panel, we focused on the most widely used form of artificial intelligence in customer experience — chatbots and virtual assistants, enhanced with natural language processing capabilities. We discussed how leading organizations are using this technology. What are the challenges and opportunities? And what are future trends?
The questions at the panel, as well as subsequent conversations, research, and reflection, inspired me to write this blog to share my thoughts on the business opportunities of these technologies, as well as some dos and don’ts of bot design.
Let us level set on some terms before we begin. Broadly defined, Artificial Intelligence (AI) is a computer program that attempts to mimic human intelligence, and that can sense, reason, act and adapt. It can respond to various inputs such as text, voice, computer vision, geo-location and sensors that detect physical characteristics such as temperature, weight, volume, humidity, motion, etc. In the future, even more inputs may be possible, such as digital smell, and human emotions.
“Bots are the new apps” declared Microsoft CEO Satya Nadella in 2016, and they are certainly on the rise. In fact, Gartner predicts that by 2019, 20% of brands will abandon their mobile apps for chatbots, which will power 85% of all customer service interactions.
The advantages of chatbots are that they are available 24/7 and customers today prefer to use messaging over other forms of communication. According to BI research, messaging apps have surpassed social media, and this trend will continue.
Good bots, bad bots
While there is a lot of interest in bots, we are still in the early days in learning how to design them well for business impact.
“Bots have the illusion of simplicity on the front end but there are many hurdles to overcome to create a great experience….We have to unlearn everything we learned the past 20 years to create an amazing experience in this new browser.” — Shane Mac, CEO of Assist
Here are ten tips I gathered from the roundtable, as well as from independent research and reflection. I want to capture how it feels to use bots, and thereby develop some design principles. I have included personal bot-related anecdotes that friends and colleagues shared with me to illustrate these principles.
1. Start with high-frequency use cases
The most successful bots are those that serve high-frequency use cases well. Businesses can identify the most frequently asked questions through call center logs or web search analytics.
One example is Capital One’s Eno bot, which is designed for five use cases: Tracking account balance, checking recent transactions, viewing available credit, confirming payment due dates, and accepting payments. For most other tasks, the customer must interact with a human.
Bots that track packages and airline check-in bots are other good examples.
The advantage of starting with high-frequency use cases, is that it frees the internal support team from answering the same questions over and over again: a low-value activity. So, it is win-win for both the customer and the business.
2. Avoid rigid bots
While it is great to start small, if the bot capabilities are too narrow, the customer experience will feel rigid and unnatural. Human beings will always behave in non-standard ways, so if the bot does not anticipate that or learn over time, the experience will feel less than delightful.
In general, chatbots based on rules tend to be restrictive, and users have to be very specific and precise to use them. However, chatbots built on an AI platform can understand natural language, so the user does not need to be ridiculously specific.
The Duolingo Conversation Chatbot is a good example of a bot that does not come across as rigid. It allows the user to practice a language by responding to its questions. It can process what the user says and respond, but it can also suggest more effective ways of saying the same thing. For example, If the question was “What do you want to drink?”, someone might respond with “coffee.” The bot would then suggest a better answer, “one coffee please.”
Several users I spoke to loved the DuoLingo app for its natural, conversational feel. They specifically mentioned how they appreciated the “better answer” feature.
3. Focus on efficiency
One of the key value propositions of chatbots is efficiency over other forms of interactions. So make the bot interaction as streamlined as possible. One rule of thumb is to compare the number of steps it takes to interact with your bot than your app.
For example, Skyscanner has a bot and an app to book flights and manage itineraries. The bot interaction takes a number of clicks to see the trip, flight, fees etc., while the app displays all of the information on a single screen.
4. Remember, garbage in garbage out
It is important to consider the quality of the company’s knowledge base when creating a chatbot, as this is the information from which chatbots pull answers. If the information is not current and of high quality, it will affect customer satisfaction and, ultimately, the success of the chatbot project. This is crucial to address before chatbot implementation, especially as companies move beyond simple rule-based systems to AI systems, powered by machine learning.
Who can forget the lessons learned from Microsoft Tay, an AI-enabled chatbot that started posting racially insensitive messages because the quality of the training data was poor?
5. Consider the customer journey
Most organizations start small to gain experience and over time expand their scope. However, it is important to think about how the bot fits into various touchpoints along a customer’s journey.
A typical customer journey has the following stages:
· Awareness — How does the potential customer become aware of product?
· Engagement — How will you nurture their interest and engage them?
· Transaction — How can they buy the product? Can they customize the product? How will they pay for it?
· Service — How can they access customer support? How might they return the product if not satisfied?
· Advocacy — How can a satisfied customer refer other customers?
A bot could be deployed in any one of these stages. However, the customer may have a disjointed experience if their end-to-end journey is not taken into consideration.
For example, my friend used a chatbot to purchase sunglasses. The company handled all of the shipping and order confirmation communication via Facebook’s messenger bot. Overall, he liked the interaction, especially because it made access to shipping information easy, without searching through all of his emails. However, he received the incorrect product and had to return it. He tried to communicate the issue via the chatbot, but it could not handle the query. He then had to figure out a way to connect with a customer service representative, and he ultimately found the experience disjointed and challenging.
6. Design for human and bot teamwork
At this point, most businesses do not have enough high-quality data for robust machine learning algorithms, so don’t try to pass the Turing test! Instead, design hybrid bots that work alongside human agents to automate routine queries. This is a great way to ensure a high-quality experience for the customer.
The bot could gracefully hand off to a live agent in either of these scenarios:
· Chatbot may offer an explicit option to transfer to a live agent
· Or, implicitly transfer the customer, if the query is too complex
Similarly, if the query is routine, an agent could hand it off to the chatbot, once the customer is notified and agrees to it.
It is important to anticipate and design these hand-offs, since they can degrade the quality of the customer experience. Sometimes customers want to know if they are speaking to a bot or a human, and not knowing could make them feel uncomfortable.
My friend is a fan of the outdoors and he uses both the Backcountry website and their embedded app for specific product questions. He does not like that the bot pretends to be a live agent, however. He would prefer if the bot revealed that it is a bot, as he would feel more forgiving towards it.
7. Extend your brand experience
While most companies are initially looking for cost savings in their customer support, bots offer the opportunity to extend a business’s brand. So, it’s important to speak in an authentic voice that resonates with your audience. Research shows that bot users skew younger, and they are an opportunity to expand your target demographic, while ensuring that the bot still represents your brand.
The American eagle app matches the brand well and makes shopping look like fun.
Yet, there are plenty of examples of bots that don’t match the brand of the product. My friend tried to use MeditateBot on Facebook messenger. Instead of replying to the chats, the bot persistently marketed the app. It asked him to download the app after every reply. This experience made my friend turn away from the product altogether.
8. Consider both voice-enabled virtual assistants and text-based chatbots
Headless, voice-enabled interfaces like Amazon’s Alexa and Google Home are great companions to text-based chatbots. We need to better understand which use cases are the best match for which form of interaction and optimize for that. For example, hands-free interaction using Alexa is great at home. However, I cannot check my flight status on the go. I am also not sure I want to check my bank balance using a voice interface, especially in a public place. I would prefer a chatbot for that. As designers, we need to refocus on human needs and be better matchmakers of technology to use cases.
9. Security and trust are paramount
Bots know a lot more about us than other forms of interactions. Businesses need to implement measures to safeguard customer data in order to earn and maintain our trust.
Here is a personal anecdote. I saw an ad for a product on Facebook and clicked on it. This took me to the product website. There, I watched a product video but did not proceed. A few days later, I decided to buy the product, and I went back to the website and paid for it. Immediately, I received a Facebook chatbot message that thanked me for the order. This caught me by surprise. I had forgotten that I had clicked on the Facebook ad to begin with. I wasn’t sure if I wanted this vendor to have my Facebook data. What data did Facebook share with this vendor? How long will this information be stored on their server? Could the vendor repackage and sell my data to other vendors? I don’t know, and not knowing makes me uncomfortable.
Imagine if I bought the product in a store and, when I went to check out, the cashier immediately asked me several personal questions like, “What is my relationship status? Do I have children?” How would I feel? I am pretty sure I would leave the product on the counter and walk away from the store.
Interestingly, customers are also creating bots to interact with businesses. Let me tell you the story of “sneaker bots”. My friend is a sneaker enthusiast. He is part of is a community that is into buying new and rare sneakers. A couple of weeks ago, a set of ten highly sought-after sneakers were released by Nike, and he tried to purchase a pair but it was all but impossible due to these “sneaker-bots”. Essentially a group of people have figured out a way to automate the purchase process for sneakers so that the sneakers are purchased much faster than a human being can click through several screens on a retailer’s website or app and as a result, the sneakers sell out in milliseconds. This example comes from the user side rather than the retailer side but I am curious how businesses will respond to address this issue and make the purchase of coveted sneakers fairer for the average consumer who does not utilize a bot.