We All Saw This Coming

Gartner published a prediction in February 2026: by 2027, 50% of companies that reduced support headcount because of AI will reverse those cuts and start rehiring. This is presented as a forecast. It's already happening.

Klarna was the poster child for AI replacing support staff. Their CEO said their AI handled two-thirds of customer service chats in the first month, doing the work of 700 agents. The stock went up and the LinkedIn posts wrote themselves. Then customer satisfaction dropped 22%, resolution quality declined on complex issues, and CEO Sebastian Siemiatkowski admitted "we went too far." The company started hiring support staff again by mid-2025.

An Orgvue workforce survey of over 1,100 C-suite and senior leaders (conducted February-March 2025) found that 55% of companies that made employees redundant to implement AI admitted they made the wrong decision. Not "mixed feelings." Regret.

Why AI-Only Support Fails

AI handles volume well. It handles complexity poorly. That's not a temporary limitation of current models. It's a structural feature of how support actually works.

Tier-one support is pattern matching. Customer says X, the answer is Y. Password resets. Order status checks. Return policy questions. Billing cycle explanations. These map cleanly to classified intents with known responses. AI does this better than humans: faster, more consistent, available 24/7, no bad days.

Tier-two and tier-three support is judgment. A customer is upset about a billing error that's technically their fault but they've been a subscriber for four years. Do you enforce the policy or make an exception? An LLM will either always enforce or always make exceptions, depending on its prompt. A human support agent reads the situation. They check the customer's history, assess the tone, calculate the lifetime value, and make a call that balances the relationship against the precedent.

Companies that fired their tier-one staff and then asked AI to handle tier-two work discovered that AI makes tier-two decisions with tier-one reasoning. It follows rules. It doesn't exercise judgment. And customers can tell the difference immediately.

The Actual Numbers

Let's look at what happens when a 20-person support team tries to go AI-only versus AI-assisted.

AI-only approach: fire 15 agents, keep 5 for "oversight." The AI handles all incoming volume. Monthly savings in salary: roughly $60,000-75,000 (assuming $48,000-60,000/year per agent). Monthly AI cost: if you're processing 10,000 tickets at $0.99/resolution with 70% automation, that's $6,930/month. Net savings look great on paper.

Then CSAT drops 15-20 points over three months. Complex tickets pile up. The 5 remaining agents burn out handling the hardest cases with no support. Escalation queues grow. One-star reviews mention "I couldn't get a real person." Customer churn increases 5-8%. Revenue impact dwarfs the salary savings.

AI-assisted approach: keep 15 agents, reduce to 10 through natural attrition over 12 months. AI handles tier-one (60-70% of volume). Humans handle everything else, now with smaller queues and more time per ticket. Monthly AI cost at $0.20-0.30/resolution: $1,400-2,100 for the same 10,000 tickets at 70% automation. Salary savings from 5 fewer agents: $20,000-25,000/month.

CSAT stays flat or improves because humans have more time for complex cases. No burnout spiral. No escalation queue. Customer churn holds steady.

The AI-assisted approach saves less per month but doesn't generate the hidden costs that eat the AI-only savings.

What the Smart Teams Are Doing

The companies getting AI right in support aren't using it to replace people. They're using it to sort, classify, and handle the repetitive stuff so their human agents can focus on work that actually requires a human.

A classification model that sorts incoming messages into 315 intents at 92% accuracy in under 200 milliseconds does something no human can: triage every single message instantly, with no queue time, no context switching, and no cognitive load. The agent gets a pre-classified ticket with the customer's intent already identified and the relevant account information pulled up.

That's a 30-40% productivity gain per agent. Not because the AI replaced the agent, but because it eliminated the 2-3 minutes of "let me read this, figure out what they want, look up their account" that happens at the start of every interaction.

The Gartner Prediction Understates It

Gartner says 50% will rehire. I think it's higher. The 50% number probably captures companies that made visible, publicized cuts. It doesn't capture the ones that quietly stopped hiring, let attrition shrink the team, and then realized six months later that response times were climbing and CSAT was dropping.

The lesson isn't that AI doesn't work in support. It works extremely well for classification, routing, and handling known-answer questions. The lesson is that "AI works well" and "AI can replace your team" are two completely different statements.

Use AI to make your team faster. Don't use it to make your team smaller. The companies that understand the difference are the ones that won't be in Gartner's "rehiring" statistic next year.