AI in marketing: three cases, and the line runs where the numbers stop

Ask how AI is going in marketing and you get either a revolution or a scandal, depending on who is asked. Three deployments with published numbers answer more usefully, and they split along a line nobody planned.

The cases, in their own figures:

• Meta: over 9 million small businesses use at least one AI creative tool, and the end-to-end AI ad suite passed a $60bn annual run rate.
• Klarna: the assistant did the work of 700 agents in month one, handled 2.3 million conversations and cut resolution from 11 minutes to under 2.
• Coca-Cola: positive sentiment around the AI Christmas campaign fell from 23.8% before launch to 10.2% after.

Three deployments, three different answers

Meta is the clean success, and it is not a creative one. Its Q2 2026 ad revenue was $59.4bn, up 27%, and the AI-powered suite's $60bn run rate works out at roughly 25.3% of that business annualised. What the machine does there is targeting, bidding and variant generation, all of it judged by a click or a conversion. Nothing about it asks whether the work is good.

Klarna is the instructive failure, because the part that worked never stopped working. Its AI agent handled the volume it was given and the speed gain was never in dispute. Resolution time fell by at least 81.8%, and that number held. What degraded was the thing nobody had a column for. By May 2025 the chief executive was rehiring, and by 2026 the company had settled on AI for routine contact and humans for the complicated and the valuable, which it now frames as a premium service.

“As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.”

— Sebastian Siemiatkowski, Klarna, quoted by Forbes

Sebastian Siemiatkowski, chief executive of Klarna, on what the deployment produced

Coca-Cola ran an AI-made Christmas film for a second year despite the first year's reception, and the sentiment reading is sharper than the shouting suggested. Positive mentions fell from 23.8% before launch to 10.2% after

The line nobody planned

Put the three side by side and the line is clear. AI has taken the parts of marketing that carry a number: impressions, bids, handling time, variants per hour. It has stalled at the parts judged rather than counted, and both reversals happened exactly there. Klarna optimised the metric it had and lost the one it did not. Coca-Cola bought production savings and paid in the only currency a holiday campaign trades in.

This is the same shape we keep finding this week one layer up. An AI answer in search cuts the click rate from 15% to 8% because a summary is measurable and a reader's return visit is not, and the argument over whether AI agents will do the buying is really an argument about which end of that line the purchase sits on. The practical test for any marketing team is one question: if this task were done badly, would a number tell you? Where the answer is yes, the tooling is already there. Where it is no, someone still has to look.

Informational material, not investment advice. Company figures are as published by each firm and by Forbes; the ratios drawn from them are ours.