A chatbot’s job is to make it easy for a customer and a brand to have a conversation. Therefore its design must reflect customer needs and expectations – and it should be informed by the type of experience that a business is attempting to deliver. No matter how cool or clever it could be, your chatbot is not going to exist in isolation. You need to understand where in the customer journey is it going to sit and what is it going to do;
Chatbot use cases need to be clearly defined. What exactly is it going to do and how is it going to do it? They can be classified by four broad chatbot use cases or families. They can provide a specific service; enable a commercial transaction; create or enhance an experience; or provide entertainment in the form of fun and diversion.
With the use case identified you need to decide if can meet your objectives with a Basic chatbot using a decision tree and scripts, or if you’re going to have to build an Intelligent chatbot capable of natural language processing (NLP) and machine learning.
This is crucial because if, for example, the goal is to automate the provision of a service, your chatbot won’t need to engage in immersive conversations with users. So building a chatbot supported by artificial intelligence (AI) will only lead to unnecessary complication and huge development costs.
For the same reason, you must conduct an ecosystem audit to understand your existing IT infrastructure and how the chatbot is going to fit within it. In the case of an Intelligent chatbot, every potential AI solution will need to be benchmarked against legacy systems to find the one that’s the best fit.
Every chatbot needs a clear identity and it has to reflect your brand and your approach to CX. Therefore you need to decide if it is going to use formal or informal English, if it’s going to have a sense of humor, if it will be masculine, feminine or robotic in personality and whether or not it’s going to have some form of visual representation.
Now it’s time to start planning and developing the conversational architecture. Regardless of chatbot type, it will mean considering every potential scenario that it could be expected to handle. That means performing FAQ analysis, web and archival research, and collecting and collating all the data needed to construct conversational paths and to understand the reasons why a customer contacts your customer service.
Once you have this information you can start writing the chatbot’s dialogue and storyboarding the journeys customer are going to take when communicating with it.
Once built, it’s time to test to make sure the chatbot can deliver the desired CX. If it’s a Basic chatbot following a decision tree, it will mean trying every single variant of every question for every single predetermined response. For an Intelligent chatbot it means having rich datasets of potential questions or intention structures and suitable answers and using that data to train the bot. The bot must understand intentions and words that may have different meanings depending on context. And it must be able to leverage its own conversational history to prompt or steer a user.
A project leader or IT developer can be tasked with training a bot. However, the best teachers are those that are already experts in your company’s CX – your customer service representatives. No one has a better understanding of customers, common contact drivers and your brand and it’s for this reason that Sitel Group created Bot Trainer.
Bot Trainer is an interface that allows customer service advisors to train and calibrate chatbots to ensure that they recognize customer intent and answer in line with the brand’s values. Regardless of chatbot use cases, training one to perform is a labor intensive task and one that can only be expedited by using teams of genuine domain experts.
Bot Trainer can be industrialized and leverages the fact that no one is better qualified than a customer service representative to tell a bot what is and isn’t relevant in the realm of customer expectations. What’s more, once a chatbot goes live, the Bot Trainer interface enables advisors to constantly monitor and improve chatbot performance as they begin to learn from customer interaction.
Once it goes live you will need to set benchmarks for performance and constantly review data to look for areas for improvement. With a Basic chatbot, there may be sets of questions and answers that are irrelevant in the real world and others that require greater work or more detail in the responses.
An Intelligent chatbot will become better at answering common questions but will need help to continue responding to new requests for answers it is capable of retrieving. For the best continued user experience, every update or change to an Intelligent chatbot should bring a noticeable improvement in performance.
But, no matter how well designed your Basic chatbot or how well trained your Intelligent bot, their performance should be measured in terms of how useful customers find it. Developing a chatbot is one thing, training and fine-tuning it so that it meets or surpasses user expectations is another entirely.
From developing a use case and a brand vocabulary to design, development training and deployment, Sitel Group can help you realize a chatbot that is firmly in line with both your brand and your customer expectations. To find out more about how we’ve helped other organizations automate and enhance their CX, or to learn how Bot Trainer is poised to revolutionize chatbot development, speak to one of our Digital Solutions experts.
If you missed the first article in our chatbot series, you can read it here: An Introduction to Chatbots.
Are you interested in learning more about the types of chatbots? Read the second article in our series: Which Chatbot is Right for Your Business.