Nearly half of customer service reps are fixing AI mistakes
While close to 50 per cent of customer service reps say they have to address errors artificial intelligence (AI) tools make, 10 per cent admit they don’t even realize something’s wrong until customers inform them, according to research published by Typewise.
Based in Zurich, where it offers a customer service software platform, Typewise surveyed more than 200 customer service professionals to produce its 2026 Agentic AI in Customer Service Index.
The need to fix AI’s mistakes could explain where customer service professionals are most willing to use the technology. Nearly 50 per cent were comfortable with AI handling simple FAQs, for example, followed by 38 per cent who trusted it in managing order updates.
For higher-stakes areas like complaints, however, trust fell below 20 per cent, and only 13 per cent said they’d just agentic AI to manage a product return.
Despite many vendors’ claims that AI will take manual and repetitive tasks off their plate so they can focus on higher-value work, only 42 per cent said it saves them time and effort.
“AI creates new work, shifts effort rather than removing it, or in some cases, introduces problems,” the report’s authors wrote. “Drafting suggestions must be reviewed. Automated actions must be monitored. Multiple systems require verification. Escalations must be evaluated. AI increases throughput in some dimensions while expanding oversight responsibilities in others.”
360 Magazine Insight
Typewise’s research offers a refreshingly granular look at how customer service teams are using AI today, and why a significant proportion aren’t realizing the potential vendors have promised.
The report attributes much of the problem to a relatively ad-hoc approach to AI adoption, with 81 per cent using a set of disconnected tools. This obviously creates difficulties in terms of orchestrating data and agentic actions to better serve customers.
What the report doesn’t mention is the high degree of legacy technologies within many contact centers, which make it harder to integrate AI or to deploy a holistic platform that allows AI agents to work together and access data more readily.
There’s a chicken-and-egg problem here, in that brands may be reluctant to move away from point tools until they’re more confident AI will deliver mostly error-free results. At the very least, reps may need a different set of tools to double-check AI outputs so they’re not hearing about issues from customers first. The latter scenario damages their trust and could affect their overall relationship and loyalty.
This gated 12-page report goes into the specifics of how customer service teams are using AI on a daily basis, and addresses one area that many brands might not have considered: how owns the outcome in multi-agent interactions? A one-liner near the end of the report sums this up very well: “Agentic AI does not eliminate human responsibility. It redistributes it.”
