I get asked a lot of questions.
I’m a CTO. I expect people to ask me questions. Crazy, geeky, technical questions, because they know I’ll answer them without spin, bias, or agenda. Arguably the best part of being an engineer is getting to understand how things work.
How do you build a better bot?
By far the question I am asked most often right now is how to create conversational experiences for customers. Specifically, conversational experiences that directly impact a business’ bottom line — decrease costs, increase revenues, and equally importantly, delight customers. No one just wants to “build a bot.” Everyone wants to build a better bot, literally the best bot to help them to meet their business’ specific needs and objectives.
Full disclosure: I work for Conversational AI and Cognitive IoT platform company UIB (which is also a WhatsApp Business solution partner) which means I build a lot of bots.
Why is everyone suddenly bursting over bots?
“Conversational experiences” can be any type of bot, chatbot, or virtual customer assistant (VCA). What they all refer to is using a conversational interface (CUI or CUX) to create a better way for your customers to talk directly to your business.
Bots have become increasingly popular over the last two years and companies’ interest in them is only increasing. The short answer is that bots are incredibly cost-effective and scalable. Bots can automate your customers’ conversations with your business in a way that both saves you money and differentiates your business. But the real silver bullet is that due to the inherent “intimacy” of conversation, companies are discovering what their customers want and need at a level of granularity (and in real time!) that has never before been possible.
Here’s what I see in our industry right now:
- Beyond offering companies the ability to deliver affordable intimacy on a global scale, the newest type of cloud-based, subscription SaaS, “Bots-as-a-Service” or BaaS, is fast, easy, and secure. I want to add a note here about the industry squabbling you may have heard between voice and texting/messaging advocates. Some smart people were passionately positing that voice or messaging was the future of conversational interfaces. The answer was decided quickly. Customers want and need both voice and text. Here’s why — when people are by themselves in a quiet environment, they choose voice. Likewise, when people are in crowded, noisy environments, they choose messaging.
- While it’s easy to joke that certain companies don’t have any customer service, there isn’t a company in business that doesn’t need to provide customer service and support. Human to machine (bot) communication dramatically improves the quality, cost, and efficiency of providing that customer service. Human call center agents, no matter how large your budget and how many people you have working your phones and chat screens, cannot instantly respond to all customers’ inquiries in their preferred languages.
- And finally, use cases for bots aren’t just about answering customers’ Frequently Asked Questions (FAQs). Smart bots are being used to generate leads, close sales, train employees, report problems, generate helpdesk tickets, book hotel rooms, purchase airline tickets, and many other use cases. Where there’s a need for customers to talk to your business, there’s a conversational solution.
What language(s) should I use?
While there are always budget limitations, look at your bot’s target user audience. What languages do they prefer to speak? Are they comfortable with conversing in English? If not, it’s well worth the one-time investment to add additional languages so that all of your customers can easily communicate in their preferred language.
What channel(s) should I use?
The answer is what you’d expect, use your customers’ preferred channels. That said, the choice of social media platforms, messaging apps, smart speakers, virtual assistants, and other communications channels can truly be overwhelming. With over 2 billion users, the world’s most popular communications channel is WhatsApp. Supporting voice and text, WhatsApp is not only fast, secure, and reliable, it provides support for rich media (e.g., images and videos) and flawlessly handles multiple languages.
When you’re working with WhatsApp, remember that it does not support graphical, menu-driven chat as Facebook Messenger does. Users can say whatever they want, which is why you should use a good NLP/AI engine and take the time to train and test your bot so it meets your objectives for it.
What NLP/AI engine should I use?
New mainstream and niche engines seem to be popping up every day. Ask yourself whether an “all-purpose” or a niche engine will work best for your specific use case. If, for example, your bot is going to be used to support cancer patients as they go through their treatments, an engine with an extensive cancer-specific library will be your best choice. Currently, the four most popular engines are IBM Watson, Google Dialog Flow, Wit.ai, and Amazon Lex. If you’re not building your bot yourself, check that your provider can support the engine that you want to use. Watch out that you don’t get locked in. With technology changing quickly in a competitive market, look for a provider who can use any engine to save you stress, time, and money should you ever want to switch engines.
Know that when your users first start using your bot, they will be curious and ask very simple questions. What you will soon discover, however, is that they very quickly move from “gently” testing your bot to directly and aggressively challenging it. It’s human nature to try to “break the bot.” While this is great for your testing, planning for this natural shift in human behavior will help you to design the bot’s NLP/AI layer to survive these “stress tests.”
How do I build my bot?
Congratulations! You’ve decided to build (or commission) a bot, and you’ve selected your language(s), channel(s), and NLP/AI engine. Now comes the fun part — actually building your bot. Based on nine million users and counting, this is my own personal checklist for how to create a compelling conversational experience with your bot:
- Create your conversations. What will people ask your bot? And more importantly, how will the bot answer their questions? If this is your first bot, remember that your bot’s replies define the character of your business just as a live call center agent’s responses do. Engage your marketing team to define the bot’s personality based on your company’s mission, values, and culture.
- Create and test your variants. Variants are the different ways that your customers will ask the bot their questions. You’ve likely heard the joke that AI isn’t very smart but it can learn fast. This is 100% true. Involve multiple people in generating your variants, ideally matching the demo- and psychographics of your targeted audience. For example, instead of asking 10 people to generate 30 variants each, it’s much better to ask 30 people to generate 10 variants each. The greater the diversity of the people you ask, the better your bot will be able to answer your customers’ questions. Keep in mind that while you’re able to add new variants at any time but everyone wants to launch a bot that can answer everyone’s questions no matter how they ask them. When finished, be sure test all of your variants on your NLP.
- Make your bot conversational. How do you do this? Always ask the users for their feedback after each conversation. A thorough beta test will really polish your bot.
- Implement live agent handover (if needed).
- Actively track the conversations. For your bot’s first week of life, invest the time to closely monitor ALL conversations so that any problems can be immediately addressed as they happen. Use people’s feedback, good and bad, to constantly improve your NLP engine. Users will give invaluable insights into what they expect from your business. Make sure that this goes — directly and unfiltered — to your top management. Remember that this is the business model behind Amazon’s Alexa Voice Service. Understanding your customers is key to delivering the products and services they want and need.
I can’t wait to talk to your bot!
About the Author
Aby Varghese is UIB Holdings Pte. Ltd.’s (UIB’s) Chief Technology Officer. He’s based in India.
Aby’s 20+ years of expertise in rapid prototyping and lean development methodologies have allowed him to explore incorporating technologies such as machine learning, natural language processing, and unified communications to unleash the potential of the Internet of Things (IoT).
He spent his early years as a software engineer with the multi-billion-dollar conglomerate KGISL Group, where he worked with Patni Computers for Hitachi Consulting. Aby earned his Bachelor of Engineering in Electronics and Communication from Bharathiar University. He is also an angel investor for startups in India and founder of the Bangalore Node School. Aby can be reached on aby.varghese@uib.ai.
