Match human involvement to the moment
Choose human involvement according to the emotional needs of the interaction, rather than assuming AI should always act like a person. Set the boundary as a brand decision, not just a technical choice.
In a case study shared by Peat, Klarna used clearly non-human, transactional AI to follow up on missed payments. Customers had a better brand experience than when a human called: the machine handled the task without the embarrassment of being chased by a person. Fake empathy and small talk were not needed.
Telstra drew a different boundary within customer service. AI handled broadband issues, but a person always handled calls to cancel a deceased partner's phone contract. The distinction makes sensitive support a deliberate commitment rather than a decision left to automation.
Define those boundaries before deployment, including where friction or frustration calls for genuine human interaction. Treat AI as a way to provide responsive, helpful service, not simply to cut costs. Gartner reported in 2026 that 87% of B2B and B2C customers say companies using GenAI for customer service must provide an option to reach a human agent.
Sources: The Prompt: Are your brand guidelines ready to deliver behaviour?, The Prompt: Why brand matters more than ever in an AI world