AI in Customer Support will not Eliminate Agents, but Change How They Work

Artificial intelligence in customer support is often described as a technology that will replace agents, but Sandra Kujundžić Drašković, CIO of Transcom, argues that this interpretation is too simplistic. The industry, she says, heard similar claims around two decades ago, when RPA robots and process automation began to enter business operations. At the time, automation was also expected to replace large parts of agent work, yet that did not happen. “What we know for sure is that agents will not disappear,” Kujundžić Drašković says.

Transcom does not see AI as a replacement for people, but as a tool that can reduce stress, accelerate work, and improve the quality of customer interactions. The clearest results, she explains, are visible in two areas: training new agents and supporting agents during live conversations with customers. The first use case is particularly important because new employees often arrive without experience in telephone communication, while initial stress and fear can lead to early attrition. “One of our main indicators is how many newly hired agents leave the company quickly because the work is stressful,” she says.

For that reason, Transcom developed an internal solution for simulating customer conversations. The system can imitate different types of customers, from difficult users reporting a network problem to users involved in a sales interaction. Agents can therefore practise different scenarios before entering real customer communication. “With this solution, we simulate the customer and allow agents to practise telephone calls on their own in different situations,” Kujundžić Drašković explains.

The advantage of this approach is that the conversations are not rigidly scripted. The system’s response depends on what the agent says, making the exercise much closer to a real interaction. “The AI’s answer largely depends on what the agent has said, so it looks very much like a real-life example of customer interaction,” she says. In one intensive campaign, she adds, the use of this solution reduced early agent attrition by as much as 80 percent.

The second important use case supports agents who are already working with customers. When an agent is in a live conversation, there is no time to search extensively through knowledge bases, documents, and procedures. What is needed is a fast, understandable, and immediately usable answer. “It is important for the agent to be able to ask the knowledge base in a human way and receive an answer that can practically be repeated to the customer immediately,” Kujundžić Drašković says. In Transcom’s experience, this improves agent satisfaction and reduces the average handling time of customer interactions.

However, such AI use depends heavily on data. Kujundžić Drašković warns that current discussions often underestimate how important structured and high-quality data is for AI implementation. Transcom has a certain advantage because the customer support industry has relied for years on interaction analytics, sentiment analysis, conversation quality monitoring, and compliance with defined procedures. “Our BI system and overall business intelligence are at quite a high level, so we had a good starting point,” she says.

For companies without organised data, the path is much more difficult. In her assessment, organisations that lack a well-structured data foundation need at least six months of work to bring their data to a level where it can be used for AI. She uses a practical comparison: if a process can be described as a recipe that allows a junior employee to work independently from day one, the organisation is ready for AI. “If that person still needs a mentor sitting next to them, then you are not ready yet,” she says.

Looking ahead, Kujundžić Drašković does not believe all industries will implement AI at the same speed or depth. Some may never see the large-scale implementations currently being predicted. The most successful organisations, however, will be those that keep humans at the centre while preparing processes for automation and AI. “Technology can solve everything, but technology is never what actually makes the change,” she concludes. In her experience, technology itself accounts for around 20 percent of success, while 80 percent depends on change management, operating models, and processes ready for the next phase of work.