You might have trained people at work. You’ve possibly trained, or know how to train, a pet. But do you know how to train a chatbot?
The consequences of getting it wrong are similar to the consequences of any poor training. A chatbot’s not going to bite a postal worker. But it could say the wrong thing. And with chatbots becoming more central to the customer experience – saying the wrong thing could cause you real problems. Microsoft discovered the consequences of rogue chatbot chat when ‘Tay’ turned racist in 2016 and ‘Zo’ insulted Windows 10.
That said, many global brands are already realizing the benefits of chatbots; for instance, Facebook Messenger users can now hail an Uber using a chatbot within the messaging app. Rides can be requested by starting a conversation with the Uber chatbot, which will also provide instant status updates.
As Head of Connected Customer Experience for Americas at Wipro Ltd., part of my role is to build and train chatbots that provide customers with valuable experiences like Uber's use case. Here's an overview of how that works.
Natural language processing (NLP)
First you need to make sure your chatbot is speaking the right language, and the right national version of that language. We know that pants, chips and queen mean different things to American and British English audiences. Your chatbot needs to master all required languages and regional variations.
Then there are cultural nuances. People attribute personalities to those they interact with. So, you have to decide whether you want your chatbot to be formal, informal… or even ‘street’. Plus, every sector has its own specialist terms. These need to be included.
Once it’s ready to communicate, your chatbot has to learn how to act. (It’s like teaching a child – but without the years of tantrums.)
There are two parts to this training – understanding the intent and extracting the entities. Intent is the objective of the human interaction. Once it understands this, it must extract the relevant information from that conversation (entities).
Efficient communications thrive on empathy and intuition. These abilities strengthen relationships and help us quickly identify what people want.
A chatbot needs training to do things that human beings do naturally. It needs to learn to understand the intent of a message. A customer might get annoyed providing too much information. If that happens, they could drop out of the communication funnel. So chatbots need to learn ways of quickly grasping the intent of a message.
Once the chatbot has worked out what you want to achieve, it needs to extract the pertinent information. This is what you do when a mechanic is overloading you with details about your car’s fault – you detune from the detail and listen out for the entities: when it can be done, and how much it will cost.
The entities the chatbot extracts are labeled for use. Effective extraction and labeling solves customer problems, improves the customer experience and drives sales.
An employee might be tasked with moving products around a warehouse. But you’d also expect them to close the door behind them to prevent theft. A chatbot has no idea what else might be expected of it. You have to train and equip it.
Being a piece of software, it can go beyond providing regular expected behaviors. It can be supercharged. Your chatbot could additionally be a security expert or pre-disposed to a particular sentiment – responding to bereavement, for example. A chatbot could even identify with a customer’s sentiment and, if necessary, trigger a human call rather than try to resolve a sensitive issue itself.
Correct and retrain
Like any new employee, your chatbot needs to learn about your business activities. Fortunately, you don’t need to spend forever writing new training materials. Reference data can be provided in the form of existing communication assets – such as chat samples or voice transcripts written for human customer contact agents.
Once these assets are loaded, your chatbot is technically in its probationary period. The next phase, like with human employees, is to correct its mistakes. This is an iterative process. But, over time, your chatbot gets increasingly better at representing your business.
Artificial intelligence (AI) integration
The final phase of training the perfect chatbot is AI integration. Here’s where it becomes a senior employee – gaining knowledge and perfecting its business practices. During AI integration, your chatbot uses machine learning to improve its communication and business skills. From the customer viewpoint, it learns how to better identify and respond to their needs.
There are two parts to this process. Firstly, natural language processing is fine-tuned to get to the heart of customers’ problems. Do they want more information, product support or to make a purchase?
The second element is the AI engine. Using this, the chatbot interacts with other AI systems to deliver outcomes. If a customer wants to meet a mortgage advisor, the chatbot can connect with the advisor’s booking system to facilitate this. Or, if the customer is simply enquiring about mortgage rates, the chatbot can provide the relevant information. Once the customer’s request has been actioned, the chatbot can also determine whether this was the most appropriate action by analyzing the customer’s response, enabling the chatbot to learn on the go and optimize future interactions.
The focus of your chatbot training regime should be to deliver the same (or higher) standards of customer satisfaction than human customer contact agents. When these standards are achieved, your chatbot has the additional benefits of being able to respond at almost limitless scale, 24/7 and in a predictable way. This level of service meets the customer need for support to be available ‘anywhere and everywhere’. Plus, there’s the added benefit that you don’t need to keep hiring and training new customer contact recruits.
Overall, chatbots are a powerful tool to enhance brand to customer communications, especially as instant messenger apps and social media are replacing emails and phone calls. The novelty of chatbots allow consumers to experience businesses in a completely new way – and, once set, chatbots are likely to become one of your most trusted and reliable employees.
About the Author
Anubhav leads Wipro’s Connected Customer Experience business delivering services across marketing, sales, commerce, digitization, automation, API, modernization, mobility and user experience in the Americas. He also manages the mobility and UX engineering practice globally. With over 20 years of experience in building software, Anubhav has contributed to innovative and award-winning programs around digital transformation and customer experience. He is passionate about conversational bots and supports clients in adopting them as part of their transformation initiatives.More Content by Anubhav Mishra