Roughly 1-2% of the American labor force is employed by call centers, and dozens of firms pay over $300M+ in wages to these workers annually. At this point, most AI pundits have made offhand remarks about how call center work will be affected imminently. But for good reason; the profit potential for automation here is massive. So where will the human call center workers go if a breakthrough in voice models makes them redundant?

The median tenure for call center positions is under a year, and workers complain of burnout and stressful working conditions^[1]. Optimists gesture at how automation has historically increased living standards and society has eventually found ways to absorb displaced workers. Pessimists argue that these workers will struggle to find comparable work since further automation will likely hit adjacent roles in short order. I found youtube videos and articles exploring these ideas, but none took a dispassionate look at the situation.

Motivation

If white collar work becomes drastically more competitive and we see soaring unemployment, it seems valuable for people to make predictions and suggest policy interventions ahead of time. We’re lacking in visions for a positive economic equilibrium beyond suggestions of a drastic UBI program. Painful labor impacts in the next 3 years seem plausible regardless of AI timelines, we should approach the possibility with a clear-eyed view.

Call center work is an ideal case study for labor automation in general

Arguments about augmentation vs automation and human comparative advantage make it hard to predict how new tools will impact a given profession’s labor market. The lines get blurred when a role involves resolving ambiguity and integrating artifacts from separate work streams. But many call center employees have a role completely described as logging on, taking calls/chats, and rarely participating in training/group logistics. There really is a lump of calls to answer; humans charge $2-10 per call while AI agents will cost pennies.

The financial motive is there, and there are dozens of emerging companies aiming to tackle it, having collected billions in funding and with aggregate valuations of >$30B. Some American giants (e.g. AT&T, Wells Fargo) employ over 30 000 call center workers and manage sophisticated hierarchies of offshoring and local handoff to drive support costs down. Many of these massive companies have their own R&D and are running pilots with AI labs directly to explore solutions.

So why hasn’t automation happened already?

My short answer is “the tech isn’t there yet”. There have been a few high profile cases of companies letting go of their call center workers, and then rehiring them shortly after. Consumers have a strong preference for dealing with humans, and present deployments are pretty limited in capability (would you trust an LLM to provide refunds on behalf of your business?), while also being pretty blatantly robotic and delayed. Looking forward in time, I argue that the actual consumer preference is to have fast resolutions and without needing to navigate the company’s internal bureaucracy through phone menus.

“Duplex’’ models^[2] are in their infancy, but they deliver audio responses without turn taking latency and are posed to cross the uncanny valley of voice interaction. Interruptions and subtle interjections in human dialogue are an important part of natural sounding exchanges

Today, the customer service startups aren’t quite attempting the wholesale replacement of a human agent. They have product demos and some early pilots of voice agents, but they aren’t a flagship product. Based on customer highlight blog posts and external communication, the main strategy is to collaborate with businesses to offer workflows to circumvent the need for calls and use AI agents to direct the flow of repeated interactions. There are some voice models in use, but in limited scenarios (e.g. after hours, Q&A use cases). Typical examples have AI agents tackling chat backlogs, guiding users through fixed multi-step processes, and other product specific work. The biggest players don’t offer an off-the-shelf product; if you’re interested they want to investigate your workflows and try to deliver some bespoke agent solution.

Contact Center Career Trajectories and The National Longitudinal Survey of Youth

To understand what may happen to displaced workers, we’re essentially asking what jobs are they likely to pursue next, and what kind of patterns we see in the career trajectories for this workers.

The best data I could find on the topic is a dataset by the Bureau of Labor Statistics that surveyed a cohort of ~9000 people born between 1980-1984 had them fill out detailed questionnaires about their employment ~yearly. It goes into weekly detail about which industry/role they worked, including reasons for leaving and other metadata. The median call center agent is in their early-mid 30s, this seems to coincide well with our survey period (ended in 2022), where respondents were 38-42 years old. Labor markets and career trajectories are bound to have evolved over this window, but there’s some useful information here still. 

This limits us to American call center workers. While offshoring is an important part of the industry, America is the single country with the most call center workers by far. We may see impacts on offshored call center workers first, but American call center career trajectories are critically relevant in the event the role becomes redundant.

High turnover rate

Median tenure in our sample is <1 year, and yet at any given point in time ~3 million Americans are working in these jobs. This seems surprising at first glance, but it’s an artifact of just how many people eventually work such jobs. Roughly 8% of the NLSY97 workers held such a role at some point in their careers, with only 1.3-1.7% of the sample holding the position at any given time.

Career Trajectories

Most CC workers have short stints, then never come back (~25% of call center workers had two or more stints). This is good news, it means call center employment is more of a pit stop than a cornerstone for the workers’ professional careers. Automation in contact center work may look like reduced hiring and increased competition for adjacent roles; the question becomes “can the other job categories absorb this much demand for work?”. Jobs categorized as miscellaneous white collar work and sales make up about half of the “next-job after call center" destinations. Walking through the most common professions following call center work:

• Retail Clerk
• Receptionist
• Sales Agent
• Administrative Assistant/Secretary
• Bookkeeping Clerk

Which is less inspiring, since many of these roles are also threatened by AI automation.

Conclusion

Since there’s already so much turnover, I expect the development of this tool and the impact on millions of workers to be subtle. Assuming a firm has a sequence of layoffs, many of these workers will be already looking for ways out or would have voluntarily left of their own will shortly after. We don’t have the data to rule out a blossoming of AI powered roles or an unemployment cascade affecting many classes of white collar work, but it’s clear that many people are likely to feel some pain here. I think the pieces of information to watch here moving forward are:

• Step changes in how convincing the voice agent models are
• Dramatically reduced contracts for outsourced CC work
• Reduced hiring volumes in the destination occupations (admin, sales, food service).

1. ^{^}

   https://www.reddit.com/r/callcentres/comments/lv7p1m/is_working_at_a_call_center_that_hard_and/

   If they were all laid off tomorrow, I don’t expect them to all thank you for it a year later. But it seems important to note how dissatisfied many workers are in this profession

2. ^{^}

   Some new developments, and a benchmark for capturing how skilled voice models are at duplex-like behavior

   https://research.nvidia.com/labs/adlr/personaplex/

   https://arxiv.org/abs/2510.07838