Global employment figures in call centres are falling for the first time. It remains to be seen whether this will lead to a permanent restructuring. But it is worth thinking this trend through: how would a contact centre operate in 2031 if companies were to hire fewer agents over the years and AI were to handle an increasing number of processes autonomously?
The trend has taken a turn
Automation has been part of everyday life in contact centres for years. It is now a familiar expectation that bots will take over routine tasks, whilst more complex interactions remain the preserve of human agents. However, the latest employment figures provide concrete grounds for this discussion.
According to Revelio Labs, global call centre employment is 5.3 per cent below its peak in December 2023. Eight consecutive end-of-quarter figures were below the respective year-on-year figures – the only eight out of the 70 end-of-quarter figures analysed since 2009. This does not indicate a consistent quarter-on-quarter decline, but it does reveal a striking change.
The analysis does not allow us to determine how much of this is attributable to AI. Revelio cites several possible causes. Furthermore, this is an analysis of professional profile data, not a comprehensive global survey.
It is revealing to look at the twelve middle-income countries examined: in these, both entry and exit rates fell. Because the entry rate fell more sharply and eventually fell below the exit rate, employment is shrinking primarily due to fewer new entrants.
This makes a difference for workforce planning. A company does not need to announce a major wave of redundancies to significantly reduce its human service capacity over several years.
Fewer new hires today. Fewer jobs tomorrow?
Let us assume this trend continues until 2031. Companies will only partially replace agents who leave. At the same time, AI systems can handle an increasing number of enquiries, including the necessary steps in the connected systems.
No one knows today how many people will be needed then. In March 2025, Gartner predicted that Agentic AI would resolve 80 per cent of common service enquiries without human intervention by 2029, resulting in a 30 per cent reduction in operational costs. This forecast does not cover all customer contacts, nor can a corresponding reduction in staff numbers be calculated from it.
For the purposes of our thought experiment, the exact figure is of secondary importance anyway. Let’s look at how such a contact centre would need to function in day-to-day operations.
By the time a human takes over, the AI has already done its work
A customer wishes to amend their contract. The AI has already initiated the process, but during the conversation a specific issue arises that requires a human decision. An agent takes over an ongoing process in which information has already been checked and individual steps completed.
They must be able to understand the basis on which the AI acted and what remains to be done. A call summary is not necessarily sufficient for this. They need insight into the steps carried out and access to the ongoing process.
If they have to search through five systems to do this, they lose time whilst the customer waits. Agent Assist and automatic documentation can ease their workload. The prerequisite remains that the relevant information is collated and that they can actually continue processing the case.
Then comes the decision. If the agent is to allow an exception or adapt a pre-prepared solution, they need the appropriate authorisation. Multiple approval loops would quickly eat up any time saved.
Anyone who leaves the handling of exceptions predominantly to people must give them appropriate scope for decision-making.
What does a whole working day look like in this context?
Our agent resolves the contract issue. Afterwards, however, it is not a simple standard procedure that awaits them, but perhaps a complaint that requires special attention.
Such a role can be professionally stimulating. However, it can also be stressful if difficult decisions and emotionally charged conversations pile up throughout the day. After all, as routine contacts disappear, so too do the less demanding aspects of the job.
It would therefore be questionable to simply carry over the previous workload unchanged. Companies must take the new mix of cases into account when planning staff deployment and assess the actual relief that Agent Assist provides.
Performance assessment must also be adapted accordingly. Someone who finally resolves a case after several unsuccessful attempts may well take longer. A processing time viewed in isolation would give a poor representation of this performance.
Better human service does not, therefore, arise solely from AI taking over routine tasks. The company must structure the remaining work in such a way that people can perform it well in the long term.
Where will the people who can do this come from?
Up to this point, we have assumed that there will be enough experienced agents in 2031. However, this is by no means a given if companies continue to hire fewer people over the years.
Many employees initially gain confidence by handling simple tasks. They familiarise themselves with products and learn which questions crop up time and again. If this learning curve is largely eliminated, companies must create alternative pathways to gaining professional experience.
Simulations and supervised work on real cases can help. This requires the time of experienced colleagues, who are also needed in day-to-day operations. A more demanding job profile does not replace this training.
For our 2031 contact centre, this means it must build up skills in a targeted manner, even though it requires fewer staff overall. Anyone wishing to hire only fully-fledged experts is relying on others to take on this task.
How flexible will the division of labour between humans and AI remain?
Now let’s change one condition of our thought experiment. The company introduces a new contract model and wishes to have certain enquiries handled in person initially, before automating these processes as well. Or it may decide to deliberately offer more human contact to selected customers.
How flexibly can the contact centre respond to this?
The technical adjustment of the routing is likely to be the easier part. However, personal support also requires people who can explain the new service and handle the relevant processes. This capacity must be organised whilst the existing service continues to run.
This makes the division of labour between humans and AI an ongoing task of design. What makes sense to automate today may look different tomorrow due to a new offering or changing customer expectations. A company should be able to push this boundary without having to rebuild its entire service organisation.
This is the practical idea behind reversibility: tasks can be deliberately reallocated. Not every process needs to always remain fully handleable by humans. The key is knowing where flexibility is needed and what skills it requires.
A company can automate extensively whilst remaining capable of taking action. To do so, it needs a realistic view of its alternatives – and of the people who make these alternatives possible.
Involving people in the thinking from the outset
Global employment figures cannot be used to derive staffing forecasts for an insurer or a local authority customer service department. However, our thought experiment shows how closely technology decisions are linked to the development of the organisation.
When AI takes on more tasks, it opens up scope: for faster processes and for human support in areas where experience and personal attention make a particularly significant difference. How well we utilise this scope also depends on how we prepare agents for their future work.
That is why an AI project must address, at an early stage, the question of how people’s work will change. What do they need to learn? What decisions are they authorised to make? And what support do they need to do so?
Let us harness the potential of AI whilst keeping people at the heart of the process from the very beginning. We are shaping the work in the contact centre of 2031 today.
FAQ: Key terms relating to AI and the contact centre agent of the future
-
What is Agentic AI?
Agentic AI refers to AI systems that not only provide information or generate text but can also independently handle tasks and multi-stage processes within defined parameters. In a contact centre, this can mean that an AI understands an enquiry, processes information and carries out individual process steps before a human agent is even involved.
-
What is Agent Assist?
Agent Assist supports the human agent during or after a customer interaction. This may include, for example, relevant knowledge base information, summaries or recommendations for action. In the scenario described in this article, Agent Assist becomes particularly relevant because human agents are faced with increasingly complex situations and need to quickly grasp the existing context.
-
What does ‘human oversight’ mean?
Human oversight means that humans could monitor the work of an AI system appropriately and intervene where necessary. This becomes particularly relevant for contact centres when AI no longer merely formulates responses but carries out process steps itself or prepares decisions.
-
What does a handover between AI and a human mean?
A handover is the transfer of a customer enquiry from a virtual agent to a human agent. A good handover should convey more than just a summary of the conversation: the human agent must be able to understand what has already happened, what information is available and what steps still need to be taken. Otherwise, the customer effectively must start from scratch again when speaking to the human agent.
-
What does ‘reversibility’ mean in a contact centre?
In this article, I use the term ‘reversibility’ to refer to a company’s ability to subsequently adjust the division of labour between AI and humans. A process that is automated today should not necessarily mean that the company will be technically or organisationally unable tomorrow to have certain tasks handled by humans again.