AI in customer service is often perceived as synonymous with chatbots. And pre-generative AI or rule-based chatbots tend to have a bad reputation – deservedly! Their poor natural language understanding and lack of context can lead to frustrating customer experiences.
The latest large language models (LLMs) have dramatically improved language understanding creating endless opportunities for AI in customer service. This generation’s genAI-powered chatbots are technically lightyears ahead of rule-based chatbots and can really help solve customers' problems with their capabilities.
Despite the significant technological leaps, studies show that customers’ trust in AI in service settings might be tarnished by poor chatbot experiences. The 2024 Consumer Voice Report by ServiceNow found 77% searching for tips and tricks trying to avoid automated customer service solutions. In the research, one in two even admitted to having posted advice online on how to sidestep automated responses.
Companies might not be ready to leave their customer service at the hands of AI either. Many worry about hallucinations and the issues they might bring up with customers. Moreover, LLM-based chatbots still require significant work to tweak and integrate into backend systems for good enough context for the bot to work on. Still, I think that customers’ hesitations or technical complexities are not enough to throw the baby out with the bath water. Generative AI has massive potential especially in the customer service context.
At Reaktor, we have found three powerful ways to enhance customer service with generative AI while keeping customer satisfaction high and technical complexities manageable. Here are three concrete ways to get started.
1. Start internally, scale externally
Customers’ determination to only interact with human service agents doesn’t mean that AI can’t be used to significantly improve both the efficiency and quality of your customer service. A great place to start is to introduce genAI-powered tools internally to customer service agents, making their daily work easier and faster.
This is a great way for companies to educate themselves on the strengths and weaknesses of GenAI solutions. You’ll have first-hand information how badly and how often AI hallucinates in your specific context and an opportunity to tweak and improve those weaknesses without compromising service quality. As AI maturity of your organization grows, you’ll be more confident about which AI-automated services and features you can offer your customers directly.
In addition to just providing information and answers, the latest LLM models are also capable of suggesting actions or even executing those after human approval. Agentic AI could for example reset a user’s password or make a reservation for a hotel room.
2. Enhance search
Another great opportunity for AI automation is user-friendly semantic search on ecommerce sites. In addition to keyword search, or advanced search with multiple options, you can provide AI-powered semantic search for products. This means that your customers can describe what they want in natural language, and AI will map their description into advanced search query parameters. The user can even continue the dialogue with AI search and narrow down or adjust the search and results.
This is usually easy to implement if your ecommerce site already has an advanced search feature. What’s left to do is to map natural language query from user to machine understanable query parameters. LLMs shine on this human-computer interface (HCI), when mapping natural language to machine, and back.
3. Improve email communications
A third effective use case for AI in customer service has been mapping emails coming to customer service into pre-created template answers. Typically about 80% of customer questions relate to small number of frequently asked questions. AI can categorize the customer’s question and select pre-generated template answer for a human customer service agent to edit and approve. This will free customer service agents time for that 10% of unique questions while the bulk of the work gets handled by AI.
