AI
History of Artificial Intelligence
In one of the key paradigms of computing, a system would pass the “Turing Test” if an evaluator could not tell whether they were interacting with a human or a program. The test is largely symbolic because it highlights that interaction is such an innate human ability that it represents the final hurdle in distinguishing software from human beings. We are social beings; from infancy, we learn—consciously or unconsciously—to converse with others. If artificial intelligence applications are capable of passing the “Turing Test,” can we then incorporate AI agents into customer service?
The interface between customers and companies has many layers, which vary from industry to industry. In general, there is often an advertising layer, which includes social media profiles, blogs, and other content. In the retail sector, there is also the website, the interface through which conversions take place. In the financial and banking sectors, we typically interact with the institution through an app.
Across all these layers, the interaction is limited to the screen and its features. However, these features do not always resolve the customer’s questions or issues, so the customer ultimately turns to a human: a representative who will resolve the issue so the customer can continue their shopping journey.
If artificial intelligence applications can pass the “Turing Test,” can we therefore deploy AI agents in customer service to interact with end consumers at one of these interface levels? Read on to find the answer to this question and learn what factors to consider when planning to configure AI agents to converse with prospects or customers.
Given the ability of AI chatbots to interact using natural language, we might easily be led to believe that they can handle interactions in a variety of contexts. However, whether on a website, in an app, or through another interface, when considering the use of chatbots in customer interactions, it’s natural to ask the following question: Is AI actually talking to customers? If so, what are they talking about?
We recently published a post here on the Inmetrics blog about how a delicate balance is needed in the autonomy of artificial intelligence agents to ensure that the interactions they facilitate are of high quality: so that they demonstrate their “social skills” and present reliable data, all without “hallucinating.” To achieve this level of refinement, many quality practices must be applied to AI.
These are the mechanisms that ensure agents provide accurate information throughout their interactions with users. Through a variety of testing strategies—some of which are widely used in the market, such as experience tests or unit tests, and others that are specific to artificial intelligence applications, such as bias tests or metamorphic tests —we have strengthened the criteria that ensure agents deliver reliable information to users.
In addition, to minimize the likelihood of “hallucinations” occurring in AI agents used for customer service, it is necessary to monitor them continuously. To this end, we rely on quality-as-a-service approaches.
Integrating AI agents into customer service is essentially a way to promote process automation, at least for some of the tasks involved in interacting with prospects and customers. Therefore, as with any automation process, it is necessary to thoroughly map out the processes and their workflows to determine which stages of customer service can be delegated to agents and which will still require human intervention.
Once the sequences of activities have been mapped out, the AI agents are connected to both the information that will form part of the context for their decisions and the information that may be included in the responses sent to customers. Many of the quality practices applied to AI that we have mentioned throughout this text begin to take effect during this phase—the training phase. The LLMs that will control the AI agents are linked to data repositories, always undergoing tests to verify their level of autonomy—that is, which data they are authorized to access.
Next, we check whether the agents are able to complete the tasks requested by the user, whether they are able to understand (and learn from) the contextual information, and whether they are retrieving the correct data from the connected databases to respond to the request. These checks are already part of the user experience tests, in which we assess the user’s overall satisfaction with the application.
Data from a recently published article in the University of California, Berkeley’s business school journal show that chatbots powered by large language models (LLMs) can save up to $0.70 per interaction. However, this figure can backfire, and instead of generating savings, it can result in losses for the company if it does not pay close attention to the quality of its AI agents. This is exactly what happened to Air Canada in February 2024.
In a Canadian court ruling, the airline was ordered to reimburse a passenger. The passenger had purchased a ticket at a fare that entitled him to a refund in the event of a family bereavement. When he requested the refund, his claim was denied on the grounds that such a provision did not exist in Air Canada’s policies; he was told this because of a mistake made by an AI agent.
Therefore, rather than simply checking whether AI agents are interacting with customers, it is essential to examine what they are saying and whether they are addressing customers’ needs. A Gartner study indicates that only 14% of self-service sessions—including those conducted by AI-powered chatbots—resolve the issues raised by prospects or consumers.
Here at Inmetrics, we have two units that work together to implement AI agents in customer service. The Digital Acceleration unit implements artificial intelligence with a focus on transforming and optimizing companies’ business processes, thereby supporting the entire journey of solutions that generate real business impact. And the Digital Experience unit, which has, among its objectives, ensuring that AI solutions are performing as expected and, thus, safeguarding both the company’s results and its reputation.
If you’re planning to replace part of your customer service with AI agents but want the assurance that service quality will remain high, click here to get in touch! Our experts will show you how you can not only maintain but also increase customer satisfaction with your service by using properly trained AI agents!
