The CEO of Allianz just announced that the travel division of the German insurance giant will cut up to 1,800 jobs over the next 12 to 18 months. The business unit employs around 22,600 people. Roughly 14,000 of them answer customer inquiries and settle claims by phone; all of these positions are ripe for AI-driven transformation.
And Allianz is far from alone in cutting these types of jobs. In February, Munich Re announced job cuts of around 1,000 positions because the repetitive work could be done more efficiently by AI models.
Jobs like these are the most obvious place to begin with AI transformation—cut the entry-level roles and close the call centers. When companies follow this path, they do so on the basis of calculations about what these workers cost. But few companies stop to ask what these workers knew.
If frontline automation is anywhere on your road map, that unasked question will decide whether you come out of the process leaner—or more detached from reality. Handled carelessly, automation leaves you deaf to your own customers. Handled well, however, AI can make your organization a better listener than it ever was with humans on the phones. The difference is a matter of design, not technology—and every leader making these cuts is standing at that fork.
What the business case can’t see
A typical business case for call center automation captures metrics like handle time, cost per contact, and resolution rates. But what it doesn’t capture—in part because it’s hard to quantify—is the fact that those 14,000 people also serve as core contributors to the company’s sensory system. These human workers stand at the exact point where the organization touches the external world.
Every day, frontline workers absorb unfiltered reality. They hear the confusion about a policy clause before it becomes a regulatory complaint, and they feel the anger about a product change before it shows up in churn data. Communicating these experiences rarely forms part of their job description. Nobody pays them to do it, and few companies even try to measure the effect. The result is that a business case will almost never be able to show with any precision what is being lost.
As I argued when examining the unbossing wave, overzealous cuts often destroy institutional knowledge that took decades to accumulate. But frontline automation can also destroy something even more fundamental: not just what the organization has learned, but its capacity to learn at all.
Truth lives at the edge
Andy Grove, the longtime CEO of Intel, made the same point without taking the final step. He argued that strategic inflection points are detected first by the people at the edges—“snow melts first at the periphery, because that’s where it’s most exposed.” For Grove, this meant middle managers. But this principle applies even more to frontline workers, because frontline staff experience the snow melting long before that information reaches the management layer.
This isn’t a new idea. At Toyota, Taiichi Ohno—the father of the Toyota Production System—famously drew a circle on the factory floor (in chalk or metaphorically, depending on the account) and made young managers stand inside it for hours. The goal was to force them to watch what was happening around them until they could understand for themselves how the manufacturing process worked.
This doctrine was called genchi genbutsu: “go and see for yourself.” The reason was simple. Truth doesn’t live in reports. It lives at the gemba—“the real place”—where the work meets the world.
You can make the same point from the negative angle. In the 1980s, IBM had 12 layers of management between the people doing the work and the CEO, each layer polishing reality a little before they offered it up to the one above. The result: The man at the top was running the company on “compounding lies.” IBM even had an internal name for the entourage of gray-suited executives who kept the CEO from ever speaking to anyone doing actual work: the big gray cloud.
AI doesn’t filter the source. It deletes it.
And this is where AI-driven automation does something categorically new.
Ohno’s chalk circle, Grove’s periphery, the frontline workers obscured by IBM’s gray cloud: All of this assumes that there is a human being standing at the edge of the company, touching reality. Executives drift from that edge, goes the logic, but an effective leader should find ways to close the distance.
The problem with frontline automation is that it removes the destination itself. When the calls are answered by AI, there is no line engineer to sit with. When robots perform all factory processes, there is no gemba to walk to. Instead, the chalk circle gets drawn around a dashboard.
In this sense, Allianz and all the other companies seeking to replace their frontline workers with AI aren’t simply adding another layer between their leadership and reality—they are removing a key source of truth about that reality.
Some companies are discovering what that costs. Bloomberg reported in June that Ford brought back around 350 veteran engineers after finding that its automated design and quality systems could not replicate the expertise of the staff it had let go.
The work could be automated. The knowing, it turned out, could not.
The two futures
Leaders now face a fork in the road.
One path leads to the numb company: Automate the interactions, bank the savings, and discover 18 months later that the organization has become deaf to the exact frequencies where trouble announces itself.
The other path leads somewhere genuinely new. Historically, an organization’s contact with reality has been a piecemeal and unsystematic affair. A thousand human agents carried their knowledge in a thousand different heads, and when that knowledge surfaced it was often by accident—in hallway conversations and escalations.
But AI-handled interactions are fully transcribed, structured, and analyzable. The machine’s interactions are all data.
McKinsey’s research on agentic AI shows what happens when that data is deliberately turned into a sensing system. A European insurer that rebuilt its commercial model around AI agents went from manually reviewing 3% of its sales calls to automatically analyzing 95% of them—a thirtyfold expansion of what the organization could hear, feeding learning loops that no manual review process could sustain.
The frontrunners in McKinsey’s study aren’t merely automating interactions; they’re routing what the machine hears back into pricing, product, and strategy.
Even the cautionary tales point this way. The fintech Klarna famously replaced the work of 700 customer service agents with AI, then reversed course and began hiring humans again when service quality slipped. The company describes the new role it is piloting as one that “blends frontline excellence with real-time product feedback.”
The difference between the two futures is not the technology. It’s whether leadership understands that organizational sensing must now be consciously designed.
The sensing audit: Four moves before you automate
Here are four moves to build a learning loop that leaves your company better at sensing than it was before automation.
- Map what the function teaches, not just what it costs. Before automating any frontline role, inventory the signals that surface there: product failures, fraud patterns, emotional temperature, emerging customer needs.
- Treat the AI as a sensing system. Build the systematic layer: Route anomaly flags, emerging themes, and spikes in confusion or anger to named owners in product, risk, and strategy, with a weekly signal digest a decision-maker must read.
- Draw a modern chalk circle. Institute a monthly ritual in which senior executives read unfiltered transcripts of AI-handled interactions—no dashboard, no digest. As Ohno knew, there is no substitute for standing where the work meets the world, even when the world now arrives as text.
- Redeploy the veterans. Your best frontline people are trained sensors with years of pattern recognition that no model possesses. Move them into roles interpreting the machine’s contact with customers—escalation design, signal analysis, voice-of-the-customer intelligence—rather than out the door.
What you stop hearing
The call center was never just a cost center. It was also the place where the company listened. As AI takes over that work—and it will—the question for leaders is not whether to automate the front line. It’s whether you will notice what you stopped noticing.
Companies don’t die from what they automate. They die from what they stop hearing.
