For companies, there is no doubt that the integration of AI has revolutionised customer management.
More and more sectors are including agents in CX. An example of a solution we have at Telefónica Empresas is in the healthcare sector, where we are introducing AI agents so that doctor-patient interactions during consultations can take place without the barrier of a PC. The doctor focuses all their attention on the patient, and the AI agent transcribes the diagnosis and procedure carried out by the doctor into the medical record. But we have many sectors; in e-commerce, AI agents are being introduced to resolve queries about stock, the banking sector uses them for possible queries about transactions and fraud, and so on.
In any case, we must bear in mind that adding an AI agent is not simply a matter of creating an AI chatbot that only answers questions, but rather of generating useful and fluid experiences. And these AI agents do not replace humans, but rather their job is to empower human experts to provide an optimal customer experience.
One of the key elements for an AI agent to be useful is access to information and the correct selection of this information. There is nothing more frustrating for a customer than being transferred from agent to agent and having to repeat the same thing over and over again because the context is not retained.
AI agents work with variables: containers that store different types of information relevant to the interaction.
We can distinguish between three types:
- User variables: with specific data about the user in question
- Session variables: with temporary information that is only useful for that session
- Global variables: These apply to all interactions with the user and must be used in every contact with the customer.
For an optimal flow of AI-human support integration, several factors must be taken into account:
Clearly define the criteria for switching from AI support to human support. What should they be based on?
On the one hand, on the complexity of the task and the AI’s ability to carry it out and, on the other hand, on the emotional load: it must be possible to identify signs of a need for escalation on the part of customers: repetition of the query, bad language, critical emergencies, appeal to a human, etc.
- Have a ticketing tool for this AI-human support escalation: define the processes that will involve escalating to a human and how that transition will take place.
- Update metrics: in addition to the usual CX metrics (abandonment rate, satisfaction, first contact resolution rate, etc.), new KPIs should be added, such as:
- The transfer rate to a human agent: how much autonomy does the agent have for the tasks assigned
- Intent accuracy: How reliable is the agent in understanding the customer’s intent and tailoring responses to be accurate to what they need?
In conclusion, it is important to mention that it is about striking a balance between cost reduction and improving the customer experience. Automation should never be forced, as not all problems can be solved with AI.
