Customer service bots are chatbots that act as automated customer support agents. Customers can communicate with the chatbot either by chatting with it (using text) or by speaking to it, normally over the phone.
The chatbot customer service models are enabled by recent advances in artificial intelligence specifically in natural language processing (NLP). This allows a business to implement a chatbot that can understand at least the simplest questions asked by customers.
Currently the most common use case for customer service is a text based bot that customers can chat with. Text based messaging, normally webchat, is very easy to implement and easy for customers to access.
Phone based chatbots for customer service are seen as a replacement for IVR however it takes more work to get a phone based chatbot to deliver a great customer experience. It will also be possible to get support in future over smart speakers such as Alexa whereby customers just need to speak to their device to get help with a product or service.
Advances in processing power and algorithms, and increases in the availability of data has lead to advances in speech recognition and natural language understanding (NLU) and this makes it feasible to use chatbots to answer certain types of questions asked in natural language. Companies can essentially set up a chatbot help desk as chatbots can easily respond to the simple, repetitive questions that make up the majority of the help desk queries. Essentially every customer gets greeted by a help bot.
If chatbots are able to handle a material percentage of inbound queries (normally in the range of 30% to 70%) this can lead to substantial headcount savings or more agent time dedicated to customers with more challenging queries. Not only can the return on investment be high, but customer satisfaction can be increased at the same time.
Call centers generally have a good idea of what the cost per incident is (or can easily figure out this number) so it is fairly easy to calculate the return on investment. This makes it much easier for sponsors of a support bot to get approval for the project.
While the economic advantages of customer service bots are obvious, what is less obvious is why this will result in better service to the end customer. One obvious answer is that there will be more time for human agents to deal with more complex queries if the bot takes some of the repetitive queries off their plate.
That however assumes that the human agents continue to increase as the number of queries grows as would normally happen. It is often the case however that once they have implemented a virtual chat assistant, businesses want to hold the headcount for human agents at the same level and meet the additional demand by scaling the chatbot. This is logical from an economic point of view, but it means that at some point the agents will reach their capacity again in terms of time they can spend on certain types of queries.
The real difference in customer satisfaction however comes via the following:
Of course, there have been cases where a chatbot is implemented poorly and this can lead to a poor customer experience. This normally happens when the chatbot is designed to be too broad i.e. cover too many topics, and where there is no quick escalation to a human if things go wrong.
Phone based chatbots are particularly vulnerable to poor customer experience because they tend to feel very slow in receiving information and responding. This problem is made worse by designers making them too verbose and making it difficult for the customer to reach to a human operator if things go wrong. If your customer wants you to bring back the IVR you know you have gone badly wrong.
It is important to differentiate between the different types of customer service bots. Broadly speaking there are bots that provide information and bots that allow customers to do things. Both type of bots normally require some integration with internal and external systems.
The simplest class of chatbot is the FAQ chatbot which normally does not require any integration with internal or third party systems. This chatbot only provides information, and the simplest version of it provides only static information. These chatbots become slightly more complex if the information they provide is dynamic and especially if they need to retrieve information from other systems. In general, FAQ bots are very simple to set up.
There can be simple and advanced versions of bots that provide information or do things. The following are the types of features that a given bot might have and to some extent determine how advanced it is.**
Human in the loop: The ability to escalate a query to a human agent in real time if the bot is not able to respond to the query. While this is useful functionality in itself, it also allows the chatbot designer to overcome a major problem of chatbots and that how the end users discover what the bot can do and what it knows. By having a human fall back the end user can submit any query they can think of confident that it will be dealt with by the bot or the human. There is an added benefit in that the bot can learn from the cases where the query is escalated to the human so that in future it can potentially answer the same question without escalating to a human.
Custom conversation flows: It’s often the case the bot needs more information from the user before it can help the user with their query. In this case conversation flows are designed to take the user through more questions before the bot provides a solution.
Learning: The bot learns from interactions with the users over time to improve its performance.
Integration: The bot sends or receives information to and from multiple internal or third party systems when processing a query from a customer. The bot essentially provides a user friendly interface to multiple systems.
Context: The bot can use context information to decide what to do next. For example this information could be information like user details, the screen in the system the user is using, some information provided by another system or what the user typed previously. This information can be used by the bot designer in manually designing conversation flows, or used by the AI to learn what to do or say in the appropriate context.
Actions: The bot can take actions based on the conversation at hand. It can serve up digital widgets, graphics on a screen or any other programmatic operation including interacting with internal or third party systems.
Having access to the best technology platform for developing bots is important, but the design of the bot is just as important. Like any software, a chatbot that is simple, succinct and intuitive will be much better to use than one that is poorly designed.
The best customer service bots will have some of the features above and typically operate in narrow topic domains where there is a high probability that they will be able to respond usefully to the query at hand. This way they can create an almost magical experience for the end user.
Companies considering implementing a chatbot need to first consider the economics of the use case in terms of economic impact the chatbot will have. They also obviously need to consider how the chatbot will be integrated with their existing systems and processes. It is normally the case that a chatbot will have some human in the loop capability and therefore the human agents will need access to a back end system where they can respond to escalated queries.
Companies also need to think beyond the way that customer service works now towards AI customer service models. Unlike a human, the chatbot can take action in any system instantly. This allows for new processes to be introduced that would not be possible with a purely human workforce. In fact, in many cases the user of the chatbot is not just the customers, but the human agents themselves.
Human agents not only train the bot both implicitly and explicitly, but also use the bot to do actions. For example, imagine an agent for an airline company working with a customer over the phone. Instead of loading up the booking system and going to the relevant screen to find the booking, the agent could just ask the bot, “bring me to last booking” and the bot would either automatically look the correct screen or could just give the agent that information. It is even possible in the future that both the customer and agent could interact with the booking system on the chat where they could both see the results.
No doubt that chatbots will first be implemented with the goal of answering repetitive questions but there is no doubt that in the future they will offer new possibilities for how companies deliver services.
Even now they are starting to be used more and more for customer enablement whereby customers can receive services from the company just by having a conversation with the bot. While this type of software interface, known as a conversational UI, is already ubiquitous in the home (for example with Alexa, Google Home and various mobile phone based assistants), more customized versions of these assistants are being developed by businesses for providing narrow services. Expect to see customer enablement use cases for chatbots eclipse customer service use case in future as businesses discover the benefits of the conversational UI.