Revisión de la información de tu cuenta

Tried to reach my bank's support line this week. Got a bot. Bot couldn't help. Bot wouldn't transfer me. Turns out the data agrees with my Tuesday morning.

Gartner analysed 432 AI customer service deployments and found that only 25% produced a positive return on investment. Another 25% produced negative returns. The remaining 42% had outcomes so unclear that the leaders running them couldn't say which way it went. Eleven percent broke even. Three-quarters of the investment, by Gartner's own accounting, is either losing money or unproven — and yet more than three-quarters of service leaders plan to increase AI spending this year anyway.

The mechanism behind that contradiction isn't hard to spot. Customer service operates under relentless cost pressure. AI demonstrably reduces cost-per-contact on the tickets it handles. Finance sees that number, boards get excited, and the investment case writes itself. What doesn't get measured is what happens to the customers who didn't get their problem solved — the ones who called back, churned quietly, or just stopped bothering. The American Customer Satisfaction Index fell to 76.1 in Q2 2026, the sharpest drop since the pandemic. Complaint rates hit an all-time high, roughly double where they were a decade ago. Those two data sets are not unrelated.

The gap between what vendors promise and what actually ships is also worth naming. Vendors market 90% deflection rates in demos, but production data across thousands of real implementations consistently lands at 55–70%. The remaining 30–45% of contacts tend to be the highest-stakes ones — billing disputes, complaints, the conversations that determine whether a customer stays. That's the slice that gets routed to a bot that wasn't built for it. You can prove the savings in one quarter. The damage shows up in three.

That's the American Customer Satisfaction Index score for Q2 2026, down from 76.7 in Q1 — the sharpest single-quarter drop since COVID. Corporate profits are at record levels. Customer complaints are also at a record high. ACSI's own analysis describes it plainly: companies are charging more while supplying less. That dynamic doesn't tend to hold indefinitely.

EU AI Act Article 50 became enforceable on August 2, with fines up to €15 million or 3% of global turnover. The rule is simple: if a customer is talking to an AI, they have to be told. At the start. Clearly. Not in a footer. Not in a terms update. UK businesses serving EU users are in scope even without an equivalent UK statute.

A lot of companies missed it. The EU deferred high-risk AI deadlines in May, headlines announced an 'AI Act delay', and a chunk of compliance teams read that as a general reprieve. It wasn't. The transparency chapter was deliberately excluded from the deferral. It landed on August 2 as scheduled, with enforcement authority attached. The companies most exposed aren't the ones that tried and failed — they're the ones that were watching the wrong deadline entirely.

Rereading Rory Sutherland's Alchemy this week. His argument that 'logic is the enemy of good ideas' hits differently when you're looking at a Gartner chart showing 75% of AI spending delivering nothing or worse.