Because of the complexity and diversity of ways in which customer intent can be expressed , even humans sometimes have difficulty understanding it . So , you can imagine how hard it must be to ingrain such skills in chatbots .
For example , a customer wanting to return a pair of shoes could type the request to a chatbot in a number of ways :
• How can I get a refund on my shoes ?
• Shoes are too big and want $ back .
• I need a refund on these shoes .
Even though none of these requests use the word “ return ”, the chatbot must understand that the customer intent is to return an item .
This is where chatbots often stumble , particularly as the need becomes more complex or emotionally-driven . This is why consumers are commonly disappointed by their chatbot experiences . Despite the existence of quite capable technology , the bots don ’ t always fulfill their intended purpose .
Chatbots typically use natural language understanding , a branch of AI , to interpret typed customer inputs .
But chatbots are only as good as the knowledge that powers them . And due to the complexity and variety of ways that humans can convey their intent , it takes large volumes of data to effectively train a chatbot to recognize the wide range of inputs they must interpret .
CHATGPT AND CHATBOTS
ChatGPT ( GPT stands for generative pre-trained transformer ) is a new chatbot technology developed by OpenAI . First released in November 2022 , it has gone through a few iterations since then ; at presstime it is on GPT-4 .
We asked Christoph Börner of Cyara “ What are the impacts and implications of ChatGPT , including the new GPT-4 , on contact center chatbots ?”
Here is his reply :
“ ChatGPT has created one of these rare disruptive moments in technology that can change everything . Technically , it is indeed a significant improvement in natural language processing .
Such training data is a set of examples expressing each intent that a chatbot uses to turn into a model for recognizing each intent . By emphasizing target use cases , the training data can teach chatbots to successfully recognize and handle each one .
... IT TAKES LARGE VOLUMES OF DATA TO EFFECTIVELY TRAIN A CHATBOT TO RECOGNIZE THE WIDE RANGE OF INPUTS THEY MUST INTERPRET .
Q : PLEASE OUTLINE THE SOLUTIONS
A : Human communication is constantly evolving , requiring continuous learning and continuous adapting for the chatbot to assure quality . The only way to accomplish this at scale is to automate the ongoing testing process .
Effective QA throughout a bot ’ s lifecycle is essential to delivering exceptional customer experiences ( CXs ). Here are the key testing considerations for organizations seeking to improve the quality and efficiency of their chatbots :
• Create training data from sources such as your call center , or third-party sources .
• Test target use cases , as well as non-target use cases .
• Test the chatbot in the context of the whole omnichannel customer journey .
• Test the chatbot ’ s understanding .
• Test chatbot escalations to a live chat agent .
• Test chatbot and live chat agent performance under peak load conditions .
• Test the security of the chatbot .
• Monitor the chatbot in production .
“ ChatGPT ’ s ability to understand complex user inquiries , as well as its power to make human-computer interactions feel more human-like , is pointing the way for other players . Together , these things could make it substantial for all kinds of business communication and create a new era of how companies engage with their customers .
“ More than that , Generative AI , with ChatGPT , could also be leveraged to assist human agents in contact centers . For 32 example CONTACT , summarizing CENTER PIPELINE conversations with clients , personalizing content , or doing a more intelligent routing .”
CONTINUED ON PAGE 34