Technology for automating certain types of customer inquiries using voice and text-based chatbots is becoming more and more common in contact centers.
Although in 2018, artificial intelligence (AI) handled only 5% of all customer service interactions, by 2022, that is expected to increase fourfold to 20%.
Yet even now, 74% of customers want more human interaction in customer support, so the only way self-service can be introduced successfully is by augmenting and supporting—but not replacing—agent-based contact. AI-powered text and voice chatbots can be incredibly effective in triaging high volumes of routine customer inquiries, but they must be executed realistically, with a clear operational goal, and sensitively.
The real hallmark of success for an automated self-service system is its ability to understand cases where an automated response is not appropriate. It needs to correctly recognize the need for human intervention, and seamlessly put the customer through to an agent where necessary. When situations escalate or get emotional, this is when swift access to a live agent is required.
What AI Requires to Do Well
An AI-based system learns how to manage situations based on data and experience. This means chatbots are only as good as both the data they are given to learn from and the process steps that have been pre-programmed to automate customer service requests.
A lot of background data is required for an AI chatbot to be able to answer any given customer query. There may be several hundred different ways that a customer may phrase a query, and numerous response options available to resolve the issue. For instance, a leading credit card company analyzed how many different ways customers asked for their current account balance. The outcome was 2,100 distinct ways—a surprising amount of variation for what seems a pretty prescriptive request at first glance.
It’s true that machine learning allows AI to incrementally improve performance over time. But at the outset, initial training is needed to customize the algorithms—the rules—for specific use cases. Live chat and email logs can provide real-world data to capture customer intents. Once these are analyzed and understood, organizations can then move towards a machine-learning chatbot implementation.
When deciding where to apply chatbot technology, focusing on processing only a small number of very high-volume customer requests can often create bigger business wins. For example, 50% of all customer inquiries may in fact relate to just three very common, frequently seen issues or faults. Directing chatbot learning and training to recognize and triage these inquiries successfully will quickly improve the customer experience, while reducing agent workloads.
Natural language processing (NLP) in an integrated platform that balances human and AI-assisted service can identify whether an interaction would be better resolved by a human agent or AI. If human, it will route it to the best-skilled advisor; if AI, it will use Robotic Process Automation (RPA) to process the requests and respond to the customer.
NLP technology can also extend beyond a chatbot—playing an integral role in email customer service support or online chat, automating and accelerating response times.
Self-service technology is not here to replace or diminish communication with agents—it is instead a tool to manage interactions more effectively. By judiciously applying intelligent routing with AI, those customers who need or expect agent intervention receive it seamlessly with a better experience—while those who have a routine query can rapidly access the information they need.