Anyone familiar with hybrid AI knows that the goal is to combine symbolic AI (human intelligence) with non-symbolic AI (machine intelligence) to achieve better results. An example where hybrid AI is being used by companies is updating standard chatbots to improve customer service.
More and more organizations are using chatbots to provide customers and employees with faster answers to frequently asked questions. Until now, end users would often prefer to talk to a live agent who can quickly understand the situation and provide assistance. The limitations of chatbots have become all too obvious, like asking for the same clarification over and over.
“You can never take all the knowledge out of someone’s head and put it into an expert system, but you can extract a lot of useful guidelines from it,” said Michael Berthold, CEO of data science and analytics platform provider KNIME. “Make that the basis and let the chatbot learn from it, so that you have an 80% solution based on expertise.”
Moving from post hoc to ad hoc troubleshooting
Using hybrid chatbots, organizations can take advantage of AI’s fast answers to frequently asked questions, while maximizing the use of human agents. To achieve that, the chatbot must be based on human knowledge. Conversely, the stakeholders building and training the chatbot must be open-minded enough to accept when the AI discovers patterns that were not recognized before and resolve those limitations.
Traditionally, the limitations of chatbots are acknowledged afterwards or ‘post hoc’. By then, customers are likely to be frustrated. For deterministically programmed chatbots where given input yields given output, logs are mined for questions that have not yet been answered because they tend to be outliers. Over time, the chatbot wordt able to answer more questionsbut initially, the lack of answers to some questions can lead to customer churn.
“Make sure you include as much human expertise as possible and code the typical journey of the people having a conversation with your job chatbot. Also make sure you can catch the outliers very quickly and transfer them to people,” says Berthold van KNIME. †
People outside the organization can help with model training by providing feedback directly to the chatbot. For example, the chatbot might say, “I found this information that seems to match your search. Is this what you were looking for?” If so, the answer will be labeled as relevant to that particular question. If not, the chatbot can pull out more data or transfer the customer to a live agent.
Hybrid chatbots not only help close the gap between human knowledge and the types of questions a chatbot can answer, but also ensure that questions related to an emergency or other emotional issue are treated with empathy. This requires adjustments to the words the chatbot uses and a quick handover to a human agent who understands the customer’s dilemma and the best way to handle it.
Hybrid chatbots are used in the hospitality industry
Hospitality organizations use AI in different ways, including providing more personalized search results, answering guest questions, verifying guests, and granting instant access approval. However, the gap between that vision and current reality is that hosts have a huge amount of expertise and existing processes that are not or cannot be coded into a machine learning model.
“It’s tempting to try and implement a one-size-fits-all solution over which property managers and hospitality operators have complete control, but this severely limits the scope of questions that can be answered,” said Peter Sorbo, co-founder and CTO of digital transformation services company Enso Connect.
To get around this, Sorbo said hosts should be able to define their own “formulas” for responses. In this way, the specific answer to a question is guaranteed to be corrected by a host, and AI is only responsible for matching a question with the most correct answer it can provide.
In the hospitality industry, quality of customer service is a competitive advantage, so automated responses to guest inquiries are not enough. Therefore, hybrid chatbots combine both chatbots and manually operated live chats to provide 24/7 availability, instant responses, scalability and consistency across channels. A hybrid chatbot, if implemented properly, can also respond to complex requests without saying it doesn’t understand.
“If the hybrid AI is successfully trained and set up, customers won’t feel the frustration usually caused by communicating with bots. In fact, they may not even notice they’re talking to a robot,” Sorbo says.
Being successful with a hybrid chatbot is an iterative journey of continuous improvement. According to Sorbo, it starts with a long-term vision that prepares a company for the use of AI chatbots in the future. Based on that vision, the company should solid data infrastructure, adopt the necessary software, implement the required operational procedures and find the right talent. That way, there is a foundation for capturing the data needed to systematically train the chatbot AI platform.