The cost line and the AI line
Two instructions land on the same desk in the same week: invest aggressively in AI, and cut costs now. Most teams treat that as a contradiction to be managed. It is closer to a scheduling problem. The spending and the saving sit on different clocks, and the people who do the actual journey work are the ones who feel the gap first.
What follows is the old way of running a cost program next to the way it runs now, and the point where one turns into the other.
The way it worked last year
A cost program used to have a rhythm everyone in the building recognised. A hiring freeze, attrition left unfilled, a vendor renegotiation, a project pushed to the next fiscal year. The savings arrived slowly and could be reversed if demand came back. Service levels moved a little. Customers noticed almost nothing.
That rhythm gave you time. You could redesign a journey over three quarters and the cost line would wait for you.
What Klarna did
Klarna put one assistant in front of its customers and it did the work of 700 people. Within weeks it was handling two thirds of all conversations. Headcount went from roughly 5,500 to 3,400. Then the company moved back to a hybrid model, with people returning to the work the assistant had absorbed.
Both halves of that story matter. The volume shifted in weeks. The correction took longer.
Capacity can move faster than the decisions about who owns quality.
Where it tips
The tipping point is the moment your volume moves before your operating model does. In the old rhythm, work left the organisation at the pace of a headcount plan. Now it leaves at the pace of a deployment, and the questions that used to be answered along the way stay open: who reviews the assistant's answers, what triggers an escalation, which customer segments never meet a machine first.
Klarna answered those questions after the shift, not before. That is the expensive order.
What the numbers carry
AI gets named in layoff announcements far more often than it shows up in operating reality. Roughly 18,000 job cuts were attributed to AI in 2024, more than 100,000 in 2025, and more than 150,000 in the first six months of 2026. Challenger found AI as the single most cited reason four months running, and in one month it accounted for around 40 percent of all American layoffs.
Set that against two other findings:
- Gartner surveyed 321 customer service leaders. Only about 20 percent had genuinely reduced headcount because of AI.
- The NBER asked executives privately. About nine in ten said AI had done nothing to employment in their organisation.
Block attributed some four thousand roles to intelligence tools. The share price rose 24 percent that day. The same set of companies is spending toward seven hundred billion dollars on AI infrastructure in 2026.
The capability nobody bought
You reconcile the two instructions by hiring a capability you have not built yet. It sits between CX and operations, and it owns three things at once: where the agent works in the journey, what good looks like at each handover, and how fast you can move work back to people when quality drops.
That role makes the spending defensible and the saving durable. It also makes the correction cheap, which is the part Klarna paid for.
What you do on Monday
- Pick one journey and mark the three points where an agent could take volume today, with the escalation rule written next to each one.
- Name one person who owns quality at those handovers, by Friday.
- Put a reversal plan on paper: what you measure, what threshold sends work back to people, who decides.
- Separate your AI spend from your cost target in the reporting, so the boardroom sees two lines instead of one.
The gap between announced savings and realised savings is where most transformation programs quietly stall. The Agentic Journey Map gives you the handover points and escalation rules on one sheet, so you can place an agent in a journey without losing sight of who owns the outcome. Bekijk de materialen