When it comes to justifying investment in conversational AI, the primary factor for many companies is cost savings. Thus, from the outset, humans have been pitted against machines in a battle for survival. Natural language understanding and other machine learning techniques have created value by predicting intent and generating responses. However, the shortcomings of those technologies has made it clear that tapping into the knowledge and skills of customer support employees is critical, as currently only humans can deliver the intuition and empathy that customers have come to expect during difficult or exceptional circumstances. Companies are now realizing that for AI to successfully improve customer experience, they must find a delicate balance between technology and human capabilities.
Bot—Agent Interactions Are Currently Either/Or
Most consumers are by now familiar with the “tiered support” model of bot-agent coordination. A chatbot or speech bot (or IVR) serves as the first touchpoint for customer care: it greets customers, prompts them to describe their problem in their own words, determines the purpose of the call, answers basic questions and may even resolve routine issues. The bot automates the conversation as much as possible, but then escalates to an agent if it needs to. The escalation may be triggered by AI limitations — the bot cannot understand what the customer wants — or business rules, such as the policy to route high-stakes or highly-emotional interactions to human agents.
For an acceptable customer experience, any escalation should be offered in the same channel as the bot: forcing a web or mobile user to call into the contact center for additional help is like teaching the customer to bypass the chatbot the next time. To make it worthwhile for customers to engage with the bot, any escalation should also pass the context of the conversation, so the agent does not start from scratch but rather is aware of the problem and knows what the customer has already tried.
This type of bot-agent coordination is an “either/or” approach where either the bot is clearly in charge or the human agent is in charge. The hand-off from one to the other is "one-time" and "one-way": the human agent is not involved before the transfer, and the bot is not involved afterwards. A higher-value approach is to make the collaboration more fluid and pervasive, where the bot and human agent can jump in and out at any point in the conversation based on when their skills are best utilized.
Related Article: Why Your Approach to Chatbot and IVR Projects Is All Wrong
4 Methods to Blend AI and Human in the Contact Center
Forward-thinking companies are already testing and adopting a number of these advanced AI-human blending scenarios, including innovative techniques such as the following:
The bot remains the first point of contact for the customer. However, if the AI gets stuck or confused, a human supervisor helps the bot get back on course. For example, the customer’s utterance may be vague and map to a number of possible intents. Or the bot may misunderstand the intent and become more confused as the customer attempts to clarify the request. In these situations, a human agent can review the transcript, disambiguate the intent, and set the conversation on the right track. The bot then continues without the customer knowing that a human was in the loop.
Human Process Automation
This form of blending deals with a frequent roadblock in many enterprises: the difficulty in accessing backend systems. When this occurs, conversational AI outstrips the capabilities of the company’s APIs – that is, the bot understands the intent but cannot access the enterprise system needed to automate that intent. In this situation, the bot can ask a human agent to fetch the information and execute the transaction, which the agent can do by logging onto the various enterprise applications and cutting and pasting data between them. This technique can be a stop-gap measure while the company deploys RPA (robotic process automation) or APIs. The benefit is that customers learn to trust automation without being exposed to the work behind the scenes to overcome backend limitations.
Related Article: Why Case Management Needs to Be Part of Your Automation Plans
Some brands may prefer a high-touch support model that focuses on human agents rather than bots as the primary interface for consumers. Under this approach, the human agent can benefit by delegating certain routine tasks to the bot, such as collecting structured input (account registration, address change, credit card details) or presenting uniform content (product details, terms and conditions, regulatory disclosures). A benefit is that sensitive information (such as social security or credit card numbers) can be transmitted directly to backend systems without passing through the agent. Another benefit is to ensure compliance by presenting information that is curated, consistent and auditable. In all these cases, the agent invokes a bot that drives the conversation for the specified task. The agent remains in control, however, and can take over the interaction at any point.
When a customer interacts directly with a human agent, AI can be used to enhance the effectiveness and productivity of the agent. A bot listens to the conversation (which could be in text or voice) and feeds the transcript to a machine learning model that outputs a suggested response to the agent in real-time. The suggestion can come from a knowledge base, or can be generated from a deep neural network trained from other agent responses in similar situations. To remove noise and avoid misguided training, the training set should be filtered to select conversations from the best agents, as measured by resolution rates and customer satisfaction ratings. The neural network technology is similar to that used in some experimental open-domain chatbots that can converse with users on any topic for fun and entertainment. In customer service, neural conversational models are not ready for direct customer interactions due to the risks of uncertified answers and lack of explainability or transparency. However, the technology can be extremely effective as an agent-facing tool, providing best-practice responses that the human agent can accept as-is or edit slightly, thereby saving agent time while improving quality.
Related Article: What Makes a Chatbot Tick?
Finding the Right Balance
No longer is orchestrating interactions that involve both AI and human agents an “either/or” proposition. Companies now have the opportunity to blend the best of both in ways that take advantage of what each does best. AI supports intent determination and the ability to offer consistently accurate answers or actions at large scale. Human agents bring the personal touch: intuition, empathy and experience, which AI-based systems can only emulate.
Human agents can step in to move things forward based on their best understanding of what has been said or typed. They can also be called into the conversation when an individual raises a new topic or category that the AI has not yet been trained to recognize. On the reverse side, AI can relieve human agents of mundane tasks, and can provide advice based on the collective knowledge of the entire agent population.
Ultimately, the adoption of approaches that combine artificial with human intelligence will be driven by business and customer experience priorities, but a compelling number of high-impact use cases have already emerged. Executives in charge of customer care contact centers and digital commerce would be wise to understand and embrace those use cases.