Labor Market Polarization and the Rise of the Virtual AI Agent for Sales Calls: An Application of Routine-Biased Technological Change

In a time when labor economics is constantly evolving, the routine-biased technological change (RBTC) hypothesis serves as a potent lens to view the implications of automation and artificial intelligence on employment. This idea—one that gets taught in most undergraduate and graduate economics courses—posits that technological advancements has a negative side: it disproportionately automates routine, middle-skill tasks, giving rise to labor market polarization. As such, there’s a simultaneous increase in the demand for high-skill and low-skill jobs and a hollowing out of the middle.

There’s no better example to illustrate this trend than in the call center industry, a field that, in the past, had been a source of employment for millions of middle-skilled workers around the world. This job generally needed a mix of problem-solving but within a narrow script, clear communication, and some level of technical training. These skills placed them in the middle of the wage and complexity spectrum. But today, a new type of virtual AI agents for sales calls is challenging the notion that only low-skill, repetitive tasks will be taken over by machine automation.

Economic Forces at Play

RBTC's theory is that non-routine analytical tasks (say strategic management or software engineering) and non-routine manual tasks (say caregiving or janitorial work) are difficult to automate, while routine tasks, both cognitive and manual, are easier to automate. Call-center work—which is structured, rule-based, and easily scripted—falls neatly into this vulnerable category.

With the exponential progress of natural language understanding (NLU) and machine learning (ML) algorithms, it is possible now that AI can not only answer customer queries but also start a sales conversation. Companies like Salesforce and Gong.io are developing systems that score leads, analyze voice tone, and simulate natural, human conversation with artificial intelligence. Some companies have even claimed that AI sales reps are more effective at converting than their human counterparts in certain situations.

There are a number of economic reasons for this change:

Cost reduction: Aside from salaries, commissions, and other benefits, human sales reps incur additional expenses through the cost of training and managing them. A scalable AI agent, on the other hand, has high initial costs but low marginal costs, which means the business can be more profitable over time as volume increases.

24/7 availability: By using AI agents that never tire, become distracted, or experience a drop in quality, services can be provided around the clock, improving global reach.

Consistency and compliance: AI can be programmed to adhere to regulatory scripts to the letter, decreasing legal risks in highly-regulated industries such as insurance or finance.

The Numbers Behind the Shift

The global call center market alone was worth $340 billion in 2020, and is still growing! That’s according to Statista. In India and the Philippines, the world’s two largest call center hubs, more than 2 million people work in customer service. In 2023, Gartner says more than 30% of call centers in North America have piloted AI agents for inbound or outbound contact, and that will double by 2027.

According to proponents, AI will “augment” human workers instead of displacing them. However, actual deployment paints a different picture. Take for instance, California startup Replicant.ai, which has implemented AI sales agents in everything from telecom to healthcare, and claim businesses can cover up to 80% of Tier 1 calls without having a human-in-charge.

Ramifications for the Labor Market

Labor market polarization worsens income inequality by concentrating wage gains on highly-skilled positions and causing an oversupply of middle-skill workers who are pushed into lower-wage jobs. This was already happening with manufacturing automation; and now, it’s surfacing in white-collar service positions.

Policymakers and educators will have to rethink job retraining and educational pathways. The traditional advice to “learn to code,” is neither realistic nor scalable for hundreds of millions of people. Rather, the focus might have to be on skills that don’t succumb to automation: adaptability, emotional intelligence, jobs that require human empathy, and ethical judgments.

Conclusion

The rise of the virtual AI agent for sales calls is yet another demonstration of the long reach of routine-biased technological change, a key concept in labor economics. As AI becomes more capable, the line between routine and non-routine labor will blur, inviting us to re-evaluate assumptions about which jobs are safe. If hollowing out in the middle keeps up, economic policy will have to address one critical question: how do we create a labor market that is both efficient and inclusive in the age of intelligent machines?