Somewhere tonight, a customer is on hold. Her card got declined at a hotel checkout in a country she doesn't usually visit, and the fraud model that flagged it did its job well. It caught something a human reviewer might have missed at three in the morning. But she's standing in a lobby with a bag over her shoulder, and the model can't talk her through it, so she's waiting for a person, and the person who eventually picks up has spent the whole shift on calls exactly like this one, because all the easy calls got handled before they reached a human at all.
That's the part of the AI-in-customer-service story that doesn't show up in the adoption numbers. The numbers are impressive on their own terms. Gartner surveyed 321 customer service and support leaders in October 2025 and found 91 percent of them under direct pressure from their own executives to implement AI in 2026. Eighty-four percent plan to add new skills to the agent role and adjust hiring profiles around it. Nearly 80 percent are planning to shift at least some agents into new positions entirely. Kim Hedlin, a director of research on Gartner's customer service and support practice, put it plainly: "Service organizations are entering a period where AI and human expertise must work in tandem. Leaders are not just deploying AI, they are redesigning service models to ensure that technology enhances the customer experience while humans provide context, empathy, and judgment."
That's the plan. Here's the gap between the plan and what's actually been built. Adobe's 2026 AI and Digital Trends report, drawn from surveys of 3,000 executives and CX practitioners and 4,000 customers, found fewer than a quarter of organizations running even limited pilots of agentic AI, the kind that doesn't just answer a question but acts on a customer's behalf. Only 16 percent have it embedded in customer support today. And yet 78 percent expect agentic AI to be handling at least half of customer support within eighteen months. That's not a rollout. That's a belief running well ahead of anything anyone has actually built.
I don't think the gap is a story about AI hype, though it's a little of that too. I think it's a clue about where the real work is heading. Adobe's respondents already credit generative AI with real wins, 70 percent report improved personalization, 59 percent report improved retention, and it's not hard to see why. AI is being pointed at exactly the interactions that were always easiest to systematize. Status checks. Password resets. Order tracking. The volume that used to fill a contact center's day without asking much of anyone's judgment.
So AI isn't shrinking customer service work. It's sorting it. And the sorting concentrates what's left.
The Residue Map
Picture every customer interaction on two axes. One is how much is at stake. The other is how standard the situation is, routine on one end, exception on the other. Four kinds of interaction fall out of that.
Low stakes and routine is the shipping update, the password reset, the order status check. This is where AI belongs outright, and it's where most of the adoption numbers above are coming true. Nobody measures a brand relationship by how a routine query got answered.
High stakes and routine is the large refund, the contract renewal, the account change with real money attached but a well-understood path to resolving it. What's emerging here looks like AI drafts and a person signs off, which is a fair description of what Hedlin meant by technology and human judgment working in tandem.
Low stakes and exception is the odd one-off nobody wrote a policy for. AI can try here and fail without much cost, because the stakes are low enough to absorb a bad attempt.
High stakes and exception is the one that matters. This is the customer whose flight got cancelled, whose claim got denied, whose situation nobody anticipated when they wrote the manual, and it's where the whole relationship with that customer gets decided in a few minutes. The research says two things about this quadrant that haven't been reconciled yet. Adobe found only 43 percent of customers say they'd even be willing to engage a brand's AI concierge in the first place, and the trust factor customers name most often is the ability to reach a human at any time. At the same time, this is where the emotional weight is landing hardest as everything easier gets automated around it. Genesys surveyed 5,811 consumers and 1,560 CX and business leaders for its 2026 State of Customer Experience report and found 90 percent of leaders now expect human-agent interactions to become more complex or emotionally charged, even as 82 percent expect autonomous agents to be orchestrating customer experience broadly within three years.
Put the two together. The bulk of the volume is being absorbed by machines, and what's left is disproportionately the hard core, arriving on the desks of whichever humans are still there, right as those roles are being redefined out from under them.
Why the concentration goes unmanaged
If the residue quadrant is where trust actually gets won or lost, it should be the best-resourced, most closely measured part of the operation. It isn't. Only 31 percent of organizations have a measurement framework for agentic AI, and Adobe found less than half say their underlying customer data is even clean enough to support one. The investment is flowing toward the visible, easily-funded part of the shift. Fifty-six percent of leaders are prioritizing more personalized experiences. Forty-five percent are prioritizing automating repetitive tasks. Comparatively little is going toward training the smaller group of people who will handle whatever AI can't.
There's a communication problem sitting under the operational one, too. Nearly a third of organizations report real misalignment between executives and practitioners on AI strategy, and 61 percent of those leaders point to executive misunderstanding of what the technology actually does as the top driver. An executive under pressure to show AI progress has a reason to count deployments. A frontline leader watching the hard calls pile up is looking at a different set of facts. Neither one is wrong. They're describing different parts of the same organization.
What this changes
None of this is an argument against putting AI into customer service. The case for automating the low-stakes, routine work isn't really in dispute, and the numbers on it are real. The argument is narrower. The interactions AI removes and the interactions that remain aren't the same job, they don't get measured the same way, and they can't be staffed by people trained for the volume era. Treating "AI now handles most tickets, so we need fewer people" and "the tickets that are left are harder" as one fact instead of two is how a company ends up hollowing out exactly the part of its operation that decides whether a customer stays.
The organizations that get this right won't be the ones with the biggest AI rollout numbers to report next quarter. They'll be the ones who can say, specifically, who is handling the residue now, how that person was trained differently than the agent who sat in that chair two years ago, and how they know, with an actual number in hand, whether the handoff is working.
Sources
- Gartner Survey Finds 91% of Customer Service Leaders Under Pressure to Implement AI in 2026: gartner.com
- 2026 AI and Digital Trends Report (Adobe), via CMSWire coverage: cmswire.com
- 2026 State of Customer Experience Report (Genesys): genesys.com