In every contact center there were calls that nobody considered particularly complex. A change of address, verification of an order, an explanation of a bill, a simple complaint, a first objection to an offer. From a manager’s perspective, this was routine. From a new agent’s perspective, however, such calls were an important school. They learned to listen, ask questions, find their way around the system, tell the essential from the nonessential and gain confidence in a conversation with a person who reacts in their own way. Today this school is gradually disappearing. Chatbots, voicebots and self-service processes handle a growing share of simpler requests. Operations gain capacity, and customers often a faster answer. Yet a question arises that is easy to forget when calculating savings: where will our seniors learn tomorrow?
Routine was not just cheap work. It was a hidden educational infrastructure.
When a new hire started on simple contacts, they did not have to master all the variables of customer communication at once. The content of the inquiry was predictable, the next step relatively clear and the risk of mistakes lower. They could therefore devote attention to tone of voice, verification of data, working with the CRM, recording the request or the final summary. Each repetition freed up some mental capacity for the next skill. After weeks, the basic steps became automatic without the person ceasing to understand their purpose.
Experience does not arise from simply counting hours. It arises from encountering deviations: a customer asks an unexpected question, an internal procedure collides with reality, data in the system contradict each other. It is from these small exceptions that an agent builds a mental map of operations. They understand not only what to say but also why they say it, what can go wrong and when to ask for help. For a long time, routine work thus functioned as an informal apprenticeship workshop of customer care.
How experience debt arises
By the term experience debt, we here mean a situation in which an organization takes away people’s opportunities to learn the work gradually without replacing them with another learning system. In the short term, everything may be fine: automation handles routine requests and operational numbers look better. In the long term, however, there may be a missing layer of employees who built the basic skills needed for more complex cases. It is not an established accounting indicator but a useful managerial metaphor.
The debt is inconspicuous. At first it does not show up in the budget. It appears later in longer training, more frequent escalations, uncertain decisions, overloaded seniors and hard-to-fill specialist roles. When the company then needs a person who can handle an atypical dispute or crisis communication, it discovers that the classic path from a simple call to a demanding case no longer exists. The question should therefore not be whether to stop automation. It should be: which learning function did the original work serve and what will we replace it with?
The automation paradox: the human steps in only at the hardest moment
The problem was pointed out in the professional literature long before the arrival of generative AI. In 1983, Lisanne Bainbridge described the “ironies of automation”: an automated system can take a person out of day-to-day control but expects them to intervene in a nonstandard and demanding situation. Yet it is precisely then that the person may lack recent practical experience. This principle cannot simply be transferred mechanically to every chatbot, but as a warning for work design it is very relevant.
In a customer center, it may look like this: a bot handles a simple shipment status check. It hands over to the agent only a customer with a complicated complaint, contradictory information and growing frustration. The person must take over the context, assess the options, stay calm and decide within the limits of their authority. The more successful the automation of routine work, the more demanding the average contact left for people may be. That is not a flaw of the technology in itself. It is a change in the job profile, to which leadership and training must respond.
Why a new hire cannot simply skip to senior work
Seniority is not just knowledge of a wider range of products or the ability to click the right button. It includes recognizing patterns, a feel for context, working with incomplete information and estimating the consequences of individual steps. An experienced agent often knows when to ask the customer further questions, when to pay attention to a seemingly small detail and when a quick solution would create another problem. These abilities are built up gradually.
A new hire can be familiarized with the rules during training. A presentation alone, however, cannot pass on the full experience of how the rules work in diverse situations. If an organization is content that an employee passed a knowledge test, it may mistake memorizing information for the ability to act well. The difference becomes apparent only when the customer interrupts the prepared procedure, starts to object or presents an unexpected combination of problems. That is exactly why the new path of development must be built on action, decision-making and feedback, not just on consuming content.
AI can transfer knowledge. It must not replace the learning itself.
Technology is not only a cause of the risk; it can also be part of the solution. A study by Brynjolfsson, Li and Raymond published in the Quarterly Journal of Economics followed the deployment of a generative assistant among more than five thousand customer support workers. The authors recorded an average increase in productivity and more pronounced benefits for less experienced workers. The results suggest that suitably designed AI can convey the proven practices of more experienced colleagues. The specific percentages, however, come from that particular environment and cannot be transferred as an expected result for WOXA.
This leads to a fundamental difference between two ways of using AI. An assistant that merely writes a reply, which the employee reads out without understanding it, may improve immediate output without developing their judgment equally fast. A training tool that lets a person respond on their own and then explains the strong and weak points creates room for active learning. An organization needs to distinguish both: operational assistance for efficiency and training for a lasting ability to perform even without prompts.
A new apprenticeship model: from a safe attempt to an independent decision
The replacement for lost routine should not be one long course before starting. It should be a deliberately built series of experiences in which difficulty increases and the employee has a safe space to make mistakes. First, simple scenarios with a single goal, for example correctly verifying the customer’s need. Then a combination of two skills: identifying the need and explaining the procedure. Next, a realistic objection, a lack of information, a conflict of priorities or an emotionally charged conversation.
After each practice session, feedback must follow, ideally on specific lines, not just a general score. The next attempt should target precisely the area that did not succeed. The supervisor thereby gets a structured development path and the agent the feeling that they can grow without being exposed to unnecessary risk when dealing with a real customer. Only then does it make sense to gradually add real contacts with mentoring and quality control.
What WOXA brings to this model
WOXA is a web platform for training voice and text conversations with an AI counterpart, developed within the Konecta group. Its essence is not to handle customer calls automatically on behalf of the agent. On the contrary, it allows the agent to actively lead the conversation, respond to objections, uncertainty or emotional pressure and then receive feedback. This is exactly why it naturally fits the space that remains after part of the simple operational workload is automated.
The WOXA website describes voice and chat simulations, weighted scenario goals, an annotated replay with an AI coach, a Scenario Studio and team overviews. A trainer can design a series of situations from a basic call to a demanding complaint handover. A supervisor tracks progress over time and works with specific areas for improvement. The result of a simulation, however, is not automatic proof that the agent will perform equally well in live operations. That must be verified separately.
Example: three levels of the same customer problem
Let us imagine an inquiry about an undelivered order. At the first level, the customer is only checking the status. The new hire must verify identity, find the correct data and clearly communicate the next step. At the second level, the customer is impatient because they have already called once. The agent additionally summarizes the history, acknowledges the complication and works with expectations. At the third level, the shipment is important because of an upcoming event, information from different systems differs and the customer immediately wants a supervisor.
Seemingly it is still one category of request. In reality, the demands on memory, judgment, communication and decision-making change step by step. If self-service resolves the first two levels, the third level does not disappear anyway. A simulation makes it possible to recreate the first two as learning steps and gradually add complexity. What matters is not to create the most difficult scenario on day one but to ensure a comprehensible path from the basics to confidence.
Experience is not just a score: what to actually evaluate
In training, it makes sense to track primarily observable behavior. Did the agent understand the problem in their own words? Did they verify the necessary facts? Did they avoid making an unauthorized promise? Did they present an option and clearly summarize the next step? Could they recognize the moment when they must escalate? Did they manage to set boundaries during personal attacks? These are questions that relate more closely to real work than a single summary number.
A score can help track repeated attempts, but it must be readable: the employee should know what they are evaluated on and how they can improve. Within a pilot, it is advisable to use the same scenarios and scoring rules at the beginning and after several weeks. Alongside simulation results, then measure the quality of real calls, repeat contacts, the safety of escalation and employee feedback. Improvement in a simulation is important, but higher customer satisfaction or lower turnover do not follow from it automatically.
The mentor and the AI coach have different roles
An experienced trainer brings knowledge of unwritten connections, company culture and sensitive decisions. They can explain why a chosen phrasing helps in one situation and may sound cold in another. An AI coach offers availability, repetition and a comparable structure of feedback. They are not rivals. A meaningful design combines both: the agent completes routine attempts on their own and goes through error patterns, disputed situations and higher-impact decisions with the supervisor.
Mentoring can moreover shift from constantly explaining basic scripts to developing judgment. An experienced colleague is not tied up replaying the identical conversation to dozens of new hires; they can invest time in working with complex cases. This is not an argument for limiting human support but for using it more meaningfully. People must remain available when a situation exceeds the prepared scenario or requires accountability.
Recommended four-week pilot
Devote the first week to mapping skills and a simple introductory test. Choose three typical situations on which new hires used to learn. For each, describe the specific behavior the agent should master. In the second week, train the basic scenarios, change only one variable and use annotated feedback after each attempt. In the third week, raise the difficulty: incomplete information, emotional reactions, a repeated complaint and handover. In the fourth week, repeat the baseline test and add controlled verification in live operations.
For each team, determine in advance a sponsor, the minimum time for training and the rules for working with data. One group can start the program earlier, another later; the comparison will help separate the effect of training from ordinary operational changes. There is no need to promise a specific percentage increase. More important is to find out whether employees really decide better, handle demanding situations safely and gain independence faster. The pilot must be a learning project, not yet another way of sanctioning new hires.
Skills map: what a junior must be able to do before moving on
Arrange competencies into four areas. The operational basics include orientation in systems, working with history and accurate recording. Communication competencies consist of active listening, structured questions and a clear summary. Decision-making competencies include distinguishing the verified from the assumed, choosing the next step and the ability to admit uncertainty. The fourth area is safety and respect: boundaries, escalation, handling sensitive data and working with emotions.
A junior does not have to be perfect at everything, but should demonstrate basic competence across the areas before being given a more complex workload independently. The criteria should be explained in advance and supplemented with specific examples. One flawless practice call is not enough. Better is repetition in several scenarios with a slightly different context, to verify understanding and transfer of the skill, not memory of an exact phrasing.
Why simulations must not become another automated routine
Poorly designed training can repeat the same problem as the automation of operations. If a participant merely clicks the right answer from a menu but does not have to formulate it independently, they are not practicing a real conversation. If they learn to collect points by repeating a single memorized sentence, they may acquire a trick for the test instead of a competence for the customer. And if AI always advises before the employee starts thinking, they get another prompt, not experience.
That is why you should include in the design free-form answers, different customer reactions, ambiguity, harder combinations of requests and moments when the correct procedure is to hand over the case. Before the debrief, first ask the agent what they consider their strong point and what they would do differently. Only then confront their reflection with the AI feedback. Learning should require some effort. Without it, a trainer may become just another automated aid.
Developing seniority is also management’s responsibility
No software will fix a situation in which the team has no time to learn, the supervisor has no capacity to give feedback and all metrics reward solely immediate speed. If performance is measured only by the number of contacts processed, employees will logically limit the activities that slow the number down: reflection, consultation, verification and training. A company may then buy a sophisticated platform but in fact give it no place in operations.
Developing seniority requires capacity planning just like shifts, technology integrations or marketing. Dedicated time, clear expectations, quality scenarios, access to mentors and the ability to make a mistake in a safe environment are management decisions. The result should be an employee who can cope even at the moment when the scenario does not cover everything and AI is not available. Knowledge without independent judgment creates dependence, not an expert.
Automation and respect: an employee should not be the last line of defense without support
In the human layer of customer care, the frustration that the bot did not resolve may concentrate. The agent often takes over the customer only after they have unsuccessfully gone through self-service. A good design therefore does not start with the question of how to teach people to endure more aggression. It starts with checking the handover of context, understandable escalation options and real competencies to resolve the issue. Training does not replace fair working conditions or functioning processes.
Respect applies on both ends of the line. The customer has the right to clear information and effective help, the agent to a safe and dignified working environment. Simulations can help with setting boundaries and responding calmly, but the employer must also ensure rules for threats, supervisor support and adequate recovery after difficult calls. Expertise cannot be built on exhaustion. If AI is to be a real benefit, not only the output but also the conditions in which people work must improve.
What will change in the role of the senior agent
Alongside handling complicated cases, an experienced agent will become increasingly important as a carrier of quality decision-making patterns. They can help formulate training scenarios, identify exceptions, approve rules and provide feedback on cases in which AI does not give a reliable answer. Their experience can thus be multiplied: it should not remain only in the heads of a few of the best people but turn into systematically transferred know-how.
This shift, however, requires recognition and time. A senior who, in addition to full production, is to design scenarios and mentor colleagues needs a clearly defined role and corresponding evaluation. Otherwise the organization merely shifts the hidden costs of automation onto its most experienced employees. The point is not to produce more content but better decision-making in operations. This is exactly where the combination of senior know-how, AI simulations and progress measurement has the greatest potential.
How to tell that an organization is truly paying off experience debt
Ask about five practical signals. Is the time to independence for a new hire getting shorter without quality dropping? Can a junior explain their decision, not just repeat a script? Are escalations caused by uncertainty declining, as opposed to legitimate escalations? Are supervisors less burdened by repeatedly explaining the basics? And can people also handle cases they have literally not yet seen in training? The last question is the crucial one: good learning shows itself in the transfer of skill.
Ideally, compare entry and exit simulations, calibrated evaluation of real contacts and employee feedback. Also watch for side effects, such as overload of senior mentors, a rise in frustration or excessive dependence on prompts. When AI removes simple calls, it does not mean it has to remove the opportunity to learn. It means we must stop leaving it to chance.
Conclusion: let us automate the routine, not the path to expertise
The future of customer centers does not have to rest on the romantic idea that people should keep handling by hand everything software can handle. Automating routine can be advantageous for both the customer and the company. We only have to create, alongside the operational design, an educational design too: how a new hire gets from a first safe attempt to independently handling a demanding call.
WOXA brings to this change a practical tool for training voice and chat situations, annotated feedback and tracking the development of skills. It is not a miraculous replacement for an experienced trainer or a guarantee of certain business results. It can, however, be an important part of a new system in which a junior does not have to wait until a difficult situation catches them off guard with a real customer.
Technologies today help us do more work automatically. All the more consciously we must create the experience that no automation can live through on a person’s behalf. The more AI, the more it matters who the person is, their judgment and their preparation. We train with AI. We communicate as humans.
FAQ: most common questions
What does experience debt mean in a call center?
It is a managerial term for the risk that a company automates simple work that previously served juniors for gaining practice but does not replace it with targeted learning. The consequence may be a smaller pool of people prepared for complex cases.
Why did new hires previously learn precisely on routine calls?
Simple contacts have a limited number of variables and make it possible to gradually master working with systems, questions, explanations and closing a call. As confidence grows, further complexity can be added.
Does this mean we should not automate simple calls?
No. Automation can bring faster service and free up capacity. At the same time, however, an alternative learning environment must be created for skills that were previously developed during routine work.
How can AI simulations replace part of the lost practice?
They allow a new hire to repeatedly hold a conversation with a model customer, receive feedback and gradually increase difficulty. The best results presuppose mentoring and verification of skills in real contacts as well.
What is the difference between an AI assistant and an AI trainer?
An assistant supports a person during real work, for example by suggesting a reply. A trainer creates a safe environment in which the person creates the answer themselves, makes mistakes and learns. Both roles are useful, but they are not identical.
What should agent training track besides the score?
The quality of questions, the correctness of fact verification, the clarity of explanations, respect for the customer and for one’s own boundaries, the safety of decisions, next steps and the transfer of skill into real practice.
Can WOXA replace a senior trainer?
Not in full. WOXA can extend the possibilities of simulation, repetition and feedback. An experienced trainer is important for context, disputed cases, mentoring and accountable decisions.
How do we start without a major reorganization?
Choose three common situations, define measurable criteria, carry out an entry practice, gradually increase difficulty over four weeks and finally evaluate simulation results together with the quality of real calls.
