Nowadays we are used to interacting and communicating within websites, customer care or booking portals, in the knowledge that we are not in fact in conversation with another human but talking to a chatbot.
A chatbot is a software application that can conduct an online conversation, using a number of pre-determined responses depending on the information provided to it. They have found great success in modern businesses, particularly in customer care or HR. Here they can deal with low level complaints or queries quickly, effectively and 24 hours a day, freeing up human employees to deal with more complex issues.
There have however been challenges with the deployment of chatbots. For starters, everyone has their own unique use of language, and this will be reflected in how they type. Some people write formally in a chat window, some will use short sentences, some might use abbreviations or slang. This different use of language can prove confusing to a chatbot and can lead it to provide incorrect responses or to simply reply that it doesn’t understand.
When a chatbot is unable to find the correct information a customer or employee is looking for, it can hand off the conversation to a human, another communication channel or knowledge source. This however has also created a challenge for programmers. At which point should the chatbot hand off the conversation? Doing so too soon risks taking up a human’s time with a low-level matter, whilst doing it too late risks leaving the customer frustrated and damaging the customer relationship. Improper programming of chatbots can have a further damaging effect on the customer relationship if the chatbot is misinterpreting requests from a customer and offering irrelevant information. This can result in customers closing the chat session and damaging a customer’s relationship with that business.
The good news however is that chatbots can be integrated with powerful tools to help them understand questions and provide more relevant information. AI tools like Watson Discovery can work with chatbots to allow them to process and analyse high volumes of data. The technology within AI attemopts to mimic the key cognitive processes of a human brain, meaning it can reason with and apply existing knowledge to problems like a human does. In the context of chatbots, the technology allows the bot to analyse the question and draw on past experiences to access the correct data. It can also learn from user input and be guided to provide more accurate information each time. AI can also be used to intelligently retrieve information from datasets a business already has. For example, it can comb through a customers contract almost instantly and return the relevant information to them.
AI can also improve customer retention, by calculating and presenting the best offer to a customer at the best time. Consider an AI-infused chatbot in telecom that could comb a customer’s contract and see they are entering the final month. It can then look through the best current offering on new models for the customer and an improved contract and send it to them almost instantaneously. This is a dramatic reduction on the time this would have taken to calculate should a human agent have been on the other end of the chat.
Workflows can also be contrasted to guide the actions a chatbot can take when faced with certain tasks. Workflows can be particularly important when a chatbot doesn’t understand the question being posed to it. Often it might not be that your chatbot doesn’t have the knowledge it needs, rather the user requesting it has asked for the data with a name or request unknown to the bot. This is where a workflow can be so important.
If we wanted to look at a real example of this, we could think back to the telecom company using a chatbot. In this case, the chatbot is dealing with contract extensions. A user might refer to the length of their contract as the “contract duration”, a term the chatbot isn’t familiar with, it can however bring up a number of options of which “contract length” is one. The user can now select this option and the contract renewal can continue without human intervention.
Consumers are now demanding a more human interaction from the companies and brands they buy from. This could be a problem for chatbots, as they are inherently not human, however processes and systems can be put in place to make them more so. Natural Language Processing (NLP) is one way in which chatbots can appear more human. NLP is a part of artificial intelligence that enables the program or application to understand human language and interpret messages. It does this through a set of algorithms that explore, recognise and identify insights from text-based language information. NLP can provide the chatbot with context for a content of a message and identify opportunities within the message, for example, the opportunity to upsell to a customer. It can also be trained to identify the tone of a message and identify which are more angry or urgent and deal with them as a priority.
Although there is a lot that can be done to make sure a chatbot appears more human, there is a lot to be said for acknowledging to a customer they are speaking to a chatbot. This is due to the fact many people still hold a negative impression that by talking to a machine they will not be able to fulfil their goals. This is due to early iterations of the technology not being anywhere near as successful or powerful as today’s machines. People may still be reluctant to talk to a machine or chatbot and will most likely view it as a negative experience if the person they think they were talking to turns out to be a bot. Rather than presenting them as a machine however, companies have had success in naming the bot and presenting them as a member of their team. This could be for an internal HR bot or an external customer service bot. The reason they are so successful is that although they are not pretending to be a human, they still retain that human touch and give the user the same experience as speaking to another human.
The last thing to think about when making a chatbot successful is to keep a constant eye on the analytics a chatbot can collect. The data a chatbot can collect is critical not only for optimising their performance but to steer wider practices within your business. This can help you identify where users are entering your site, and if this event is happening quickly enough. You can also pinpoint where users are dropping out of the chat, how many of them are dropping out and why. It could be a lack of relevant information present for the chatbot to draw from or it could be a wider gap in your business strategy that needs addressing.
To find out more about effectively deploying chatbots within your business get in touch with a member of the ABP Consultancy team.