The premise that AI will revolutionise customer service has collided with operational reality. Major vendors—Salesforce, Verizon, and others—have aggressively deployed AI-first strategies, with Salesforce cutting 4,000 support roles and Klarna initially replacing human agents before reversing course and hiring them back. Yet the evidence suggests this approach is fundamentally misaligned with both customer preferences and what AI actually does well. A 2024 Gartner survey found 61 percent of customers would prefer companies didn't use AI for support at all, with 53 percent willing to switch providers if they did. The core problem is architectural: most implementations bolt AI onto legacy systems as a front-end layer—essentially a modernised phone tree—which fails at the tasks that matter most. When customers are frustrated or face complex issues requiring empathy, AI consistently underperforms. This raises a critical question for teams already managing hybrid support models: if your organisation has invested in AI-powered triage, are you measuring resolution rates by channel, or are you inadvertently pushing difficult cases toward human agents without the context or resources to handle them efficiently?
The evidence points toward a fundamentally different implementation pattern. Gartner's Brad Fager and MIT-Stanford research both converge on the same insight: AI performs best as a backend augmentation tool rather than a customer-facing layer. When AI provided real-time suggestions to less experienced call centre agents, resolution rates improved by 14 percent per hour. Intercom's Fin agent demonstrates what happens when AI is built from the ground up for resolution rather than triage—a 67 percent autonomous resolution rate with no phone tree friction. The distinction matters operationally: backend AI reduces cognitive load on agents and accelerates case resolution, whilst front-end AI creates friction and damages customer relationships. For CX leaders, this suggests the vendors winning in 2026 won't be those claiming highest automation rates, but those delivering measurable improvements in agent productivity and first-contact resolution. The uncomfortable truth is that companies pursuing cost reduction through headcount cuts are discovering that customer satisfaction and operational efficiency require human agents—just deployed differently, with AI handling the work rather than replacing the worker.
The transformation underway is real but uneven, creating a widening capability gap. Larger organisations with resources to build or licence sophisticated AI systems like Intercom's can deliver frictionless experiences; smaller companies and legacy-heavy organisations will continue delivering poor ones. This isn't a temporary implementation problem—it's a structural advantage for vendors who can afford to invest in evaluation engines and continuous refinement. For support teams, the immediate implication is clear: if your organisation is still treating AI as a cost-reduction play rather than a productivity multiplier for existing staff, you're competing against companies that have already reframed the problem. The question isn't whether to deploy AI, but whether your deployment is making your agents' jobs harder or easier.
AI’s ultimate test: Making it easier to complain to companies Vox
AI’s ultimate test: Making it easier to complain to companies vox.com