The blended human+AI model

After a wave of ambitious "AI will handle all our support" experiments — and some very public reversals — the industry has settled on a clear consensus: the winning model is blended human + AI, not AI alone. Understanding why (and the cautionary tale behind it) sets the strategic frame for deploying AI in support.

The cautionary tale — Klarna. Klarna famously deployed an AI assistant that handled millions of chats, doing the work of hundreds of agents, and publicly celebrated the cost savings. Then it walked it back — resuming human hiring after quality and customer-satisfaction complaints on nuanced cases, and settling into a hybrid model with an "always able to reach a human" promise. It's the single best illustration of the lesson: AI-only support looks great on cost and falls apart on the cases that need a human — and the resulting customer dissatisfaction isn't worth the savings. The all-AI dream met reality, and reality won.

Why blended beats AI-only:

• AI is great at some things, bad at others. It excels at high-volume, routine, well-documented, low-emotion issues (order status, password resets, simple policy questions). It fails at complex, novel, multi-system, emotionally-charged, or high-stakes cases — exactly where customers most need real help and where a bad AI experience does the most damage.
• Customers want a human available (the backlash data). Even customers happy to use AI for simple things want a path to a person for the hard or emotional stuff. Blocking that path is what generates resentment.
• **The economics only work on eligible volume.** AI genuinely saves money on the routine tickets it handles well — but pushing it into cases it can't handle costs you in dissatisfaction, churn, and brand damage that dwarfs the savings.

What the blended model looks like:

• AI handles the routine volume — the high-frequency, simple, well-documented tickets — instantly and at scale, freeing humans.
• Humans handle the complex, emotional, and high-stakes — the cases needing empathy, judgment, and real problem-solving.
• Humans supervise the AI — reviewing its performance, catching its mistakes, improving it (the evolving agent role).
• A graceful path to a human always exists — customers are never trapped with the AI; escalation is easy and well-designed (Module 2).
• Agent assist bridges both — even on human-handled tickets, AI helps the agent (drafting, summarizing, suggesting) without facing the customer directly.

The strategic principle: deploy AI to augment your support operation — handling routine volume and helping agents — not to replace human support entirely. The goal isn't maximum automation; it's the right work to AI (routine, at scale) and the right work to humans (complex, emotional, high-stakes), with AI helping humans throughout and customers never blocked from a person.

The mindset for the whole course: every ambitious "fully automate support" plan should be tempered by the Klarna lesson — AI-only fails on the cases that matter most, and the customer trust you lose costs more than the humans you saved. Blend AI and humans deliberately: AI for scale on the routine, humans for the hard and emotional, AI assisting humans throughout, and always a path to a person. That blended model — not AI alone — is what actually improves support while keeping customers satisfied, and it's the consensus every subsequent lesson builds on.