The co-pilot that agrees with everything
In early 2024 Klarna told the world its AI assistant was doing the work of 700 customer service agents. The company said the assistant handled two thirds of its customer chats, and by May 2025 CEO Sebastian Siemiatkowski told Bloomberg that Klarna was investing in human support again. His explanation was that cost "seems to have been a too predominant evaluation factor", and quality dropped as a result.
That is not a technology story. The assistant did exactly what it was asked to do. The question behind it was written wrong, and no model fixes a wrong question. The CX+ Method runs through five phases for that reason: Exploration, Problem, Solution, Proof of Concept, MVP. AI makes each phase faster and pushes harder on the exact weakness that phase exists to catch.
Write the question down before you open the tool
Start Monday by writing one sentence: what does the customer get out of this. Then open the tool. The Problem phase asks you to listen, and a language model would rather talk. Ask it what customers struggle with and you get a tidy list within seconds, built from patterns in text rather than from anyone's actual week.
Use AI on what customers said. Decide yourself what they feel. Feed it your transcripts, your complaints, your cancellation reasons, and let it cluster them. The ranking stays with the people who spoke to those customers.
Cut the list before you grow it
Exploration works by subtraction, so set the cap first. Twenty options are not twenty times better than three, they are one decision postponed. Give yourself seven by the end of the session and name out loud why each of the others goes.
AI adds. That is its nature, and it is useful at the start of a morning and dangerous by the end of it.
The phase AI cannot speed up is the one where you decide what the customer actually feels.
Make the co-pilot argue against your concept
In the Solution phase, give AI the job of attacking the idea, not polishing it. Ask it for the three reasons this concept fails at week six, written from the customer's side. A model will make your favourite idea look better every time you ask it to help, and the Solution phase exists to kill favourites.
Then test the concept on the proof of concept with a real constraint: your actual data, your actual handover to a human, your actual peak hour.
Expect the rollback and build for it
The rollback is now the normal outcome, so plan the exit before the launch. Sinch, publishing on 13 May 2026, surveyed 2,527 senior decision makers in ten countries and found that 74 percent of enterprises with a live AI customer communications agent had rolled it back or shut it down after deployment, rising to 81 percent among the organisations with the most mature AI governance.
Mature teams roll back more because they see the damage sooner. Three things make that possible:
- one named owner for the agent after the project team leaves
- a human reviewing the high stakes conversations from day one
- a measure of resolution, not deflection
The work changes, or nothing does
The gains sit with the companies that rebuilt the work around the model. McKinsey's State of AI survey from August 2026 found that about 6 percent of respondents see real earnings impact, and nearly three quarters of that group say they fundamentally redesigned workflows around AI, against roughly a quarter of everyone else.
Redesign means your team leads give up the weekly status call and start reviewing what the agent decided. That is the hard part, and it is the whole part.
What you do on Monday
- Write the customer sentence for your current project in one line and put it at the top of the brief.
- Cap your option list at seven before the session starts.
- Ask your co-pilot for three reasons the concept fails, from the customer's side.
- Name one owner for every AI agent that talks to customers.
- Pick a resolution measure and report it next to cost per contact.
You just read one pillar of a five phase method. CX+ The Method sets out all five, with the trap each one catches and the checks that keep AI in the passenger seat. The book CX+AI puts it in the context of the six building blocks that carry a transformation. Bekijk het boek