We've all been exposed to generative artificial intelligence, a set of technologies that leverage massive corpuses of data, including large language models (LLMs). Generative AI is so powerful that it can carry out complex conversations in fluid natural language. And when connected to customer service applications, it's a match made in heaven.
Customer interactions produce large volumes of unstructured data, such as emails, call transcripts and chat logs. Generative AI unleashed on this content surfaces, summarizes, and interprets key insights. This focuses your agents on value-added tasks, empathy, and relationship-building.
Generative AI will offer new user experiences and reshape the way your agents engage with their customer service applications, but be aware that it will not erode the need for these applications, nor the processes that your agents follow to engage with customers. On the contrary, it will make your customer service and CRM data more complete and your processes more accessible. Generative AI will also not replace your agents entirely; rather, it will assist them by taking on the role of a super-powered collaborator.
Generative AI embedded in customer service applications is already delivering value. It's helping create first versions of knowledgebase content from internal company sources and interaction logs. It's helping create channel-specific variants of knowledge. It's being used to autogenerate email responses for agent review based on customer profiles, products, and case details. It's being used for real-time translation, reducing the need for multilingual content and translation services. It's being used for post-call summarization. It helps arm agents with key case details so they can better understand customers and their histories. These scenarios only scratch the surface of what generative AI will be able to do.
How to Get Started with Generative AI
Assess your AI foundations. Begin by assessing your organization's readiness for generative AI. For example, are your data assets in good shape? Do you have a data center of excellence or a machine learning operations function that can absorb LLM operations. Do you have a data governance framework in place that oversees strategy, policies, and procedures to mitigate risk and validate outcomes. If you're not already an insights-driven organization or well on the way to becoming one, a generative AI project is likely to be a waste of time and resources at best.
Prepare your data and manage outputs. Depending on your use case, your LLM will need to be trained on CRM data, transcripts of customer conversations, and internal data. This significantly improves the accuracy of responses to customer inquiries, sentiment analysis, or intent recognition. You will have to prepare your internal data to train models for specific use cases.
Investigate what vendors have productized. Although building a custom LLMs offers greater flexibility, it's time-consuming and resource-intensive to do so. Buying an existing application reduces the need for specialized in-house talent and makes it easier to keep up with fast-changing technology. Every customer service vendor is piloting generative AI features. Look at what they have to offer. Many vendors integrate AI into standard workflows so agents don't have to learn new concepts. They offer predefined libraries of prompts, such as account overview or generate a service report, to guide agents to the right actions.
Only adopt a human-in-the-loop approach. Getting AI-generated content right can take many iterations and could miss the mark on the first try. Often an agent must prompt the model to change the tone or length of the content. Generated content can also infringe on copyright or include bias. Make sure that all content is reviewed by the front office for accuracy and completeness before communicating content to customers.
Rethink your workforce. Generative AI will make your agents more efficient, and you ultimately might need less talent. But, before adjusting staffing levels, rethink how to evolve your workforce. New roles, such as prompt engineers or LLM managers, will be needed, and these are natural career paths for customer service generalists and tier-one agents.
Kate Leggett is a vice president and principal analyst at Forrester Research.