Generative AI shifts CX into high gear
Although still in their infancy, generative AI and large language models are poised to change many areas of business. Nowhere is this more apparent than in customer experience.
As consumers, we’ve all done it: frantically pressing zero to skip automated systems to reach live support, or typing “agent” into a chatbot prompt to bypass the inevitable list of links to online help pages that don’t solve the problem.
Although the current generation of AI-powered tools can handle simple queries, they often struggle to answer complex questions and provide a personalised touch. These shortcomings can lead to frustrated customers and agent burnout. But this will change in coming years, thanks to a new generation of customer experience (CX) tools that incorporate generative AI trained on large language models (LLMs) to provide better, faster and more personalised responses even to complex problems.
“The beauty of conversational automation with generative AI is providing better AI-powered support to much more advanced customer needs,” says Adrian McDermott, chief technology officer at Zendesk. “A significantly greater number of incoming support needs will be effectively and efficiently answered automatically using more advanced AI, like generative AI.”
“
The beauty of conversational automation with generative AI is providing better AI-powered support to much more advanced customer needs.”
Adrian McDermott
Chief technology officer, Zendesk
For businesses, generative AI has already opened the door to more sophisticated features, such as detecting sentiment and intent. For instance, a customer showing signs of frustration can be escalated to a human agent, while someone with a very specific technical issue can be more quickly connected to the right solution.
Generative AI also helps agents by providing concise summaries and response options. Crucially, such tools enable staff to spend more time with customers who require a higher level of service.
“The first step in implementing AI is deciding what to automate versus what not to automate,” says Cristina Fonseca, head of AI at Zendesk. “We know very well that not everything should be automated – for example, interactions that are time-sensitive, edge cases and key revenue drivers.”
A major concern for any company evaluating an exciting new technology is how it is built, deployed and integrated into existing operations.
Until recently, AI modelling for call centres or customer service operations entailed a laborious training process. In addition to determining exactly what the AI was supposed to understand, the company would need to collect vast amounts of data to train the model. This in turn required quality control and optimisation steps, usually overseen by data scientists and expensive consultants.
Cristina Fonseca
Head of AI, Zendesk
“Until now, only companies with big budgets, time, and either in-house expertise or the ability to hire outside experts were able to deploy and implement AI,” explains Ms Fonseca. “It was not accessible to small and medium-sized businesses.”
The new generation of LLMs are different. Zendesk’s approach is to leverage its extensive CX dataset to build proprietary models capable of addressing specific industry use cases, as well as features such as sentiment detection.
Through OpenAI’s application programming interfaces (APIs), Zendesk has incorporated ChatGPT and other generative AI tools to build out features that can help systems more readily understand customers’ issues, expand bullet points into complete responses for human agents to use, and create easy-to-understand summaries of conversations that can help with agent handoffs and re-opened cases. Zendesk is also utilising companies' existing datasets, such as product lists, transaction histories and knowledge base articles, to enhance the AI experience.
“Companies don't want to train the model,” Ms Fonseca points out. “They just want to use it and expect it to work immediately, out of the box.”
Integrating AI
From the Economist
“The recent success of LLMs in generating convincing text, as well as their startling emergent abilities, is due to the coalescence of three things: gobsmacking quantities of data, algorithms capable of learning from them and the computational power to do so.1”
This convergence of LLMs, corporate data and generative AI will soon lead to sophisticated CX tools that do not require small and medium-sized businesses to hire data scientists or an army of contractors to deploy. Companies will be able to quickly incorporate AI into support and other CX functions, and it will also open the door to new lines of business – for instance, manufacturers selling directly to consumers,2 or companies launching paid services with more robust support from automated and human agents. It is an opportunity for business leaders to build durable competitive advantages and better understand their customers.
Adrian McDermott
Chief technology officer, Zendesk
However, although the potential of generative AI is coming into focus, there are risks. Recognising that brand equity is precious and that not every interaction should be automated are basic considerations.
“Companies have high bars for trust, safety and quality of interaction,” Mr McDermott says. “Understand how you can ground a model in facts and fine-tune it. Understand how you can keep humans in the loop as needed, and supervise AI to ensure it’s responsible, secure and ethical.”
Learn more from thought leaders and experts at Zendesk’s digital launch event.
References
1 The Economist, “Large, creative AI models will transform lives and labour markets,” 22 April 2023
2 Zendesk Blog, “Why manufacturers must shift gears from product to CX, powered by AI,” 1 August 2023
Transforming customer service: the impact of AI
We recently brought together industry experts to examine how AI will change customer service for industries and businesses in the coming years.
