Cut staff first, hire them back later: Why the positions saved by AI customer service were filled again
In a November 5, 2026 survey, Gartner found that 21% of service and support leaders had reduced their human agent headcount because of AI. Organizations that had already cut staff had reduced headcount by an average of 16.7% over the previous 18 months. The same survey included two figures that are rarely cited: 7% had already rehired people they had laid off, and 11% planned to do so within 18 months.
This is where two ways of accounting diverge. One simply enters the number of people cut in the quarterly results column; the other first measures how many tasks in a given role can be independently completed in a week, and the rework rate when tasks are handled incorrectly.
The latter approach is less flattering, but it helps prevent mistakes. At least four new expenses arise after AI goes live: dedicated AI support staff, knowledge and script maintenance, handling exceptions and complaints, and quality checks and review. The first two are visible; the latter two are often recorded as costs that already existed.
There is another, quieter cost. If permissions, approvals, and responsibility boundaries are left until the operation scales up, the price will show up two years later. Gartner predicts that in 2027, 40% of enterprises will downgrade or shut down autonomous agents already in operation because of governance gaps, forcing them to redo the process changes they have already made.
This article is adapted from a post on our company’s official WeChat account. The draft is not publicly available. This is not investment advice.
In a November 5, 2026 survey, Gartner found that 21% of service and support leaders had reduced their human agent headcount because of AI. Organizations that had already cut staff had reduced headcount by an average of 16.7% over the previous 18 months. The same survey included two figures that are rarely cited: 7% had already rehired people they had laid off, and 11% planned to do so within 18 months.
This is where two ways of accounting diverge. One simply enters the number of people cut in the quarterly results column; the other first measures how many tasks in a given role can be independently completed in a week, and the rework rate when tasks are handled incorrectly.
The latter approach is less flattering, but it helps prevent mistakes. At least four new expenses arise after AI goes live: dedicated AI support staff, knowledge and script maintenance, handling exceptions and complaints, and quality checks and review. The first two are visible; the latter two are often recorded as costs that already existed.
There is another, quieter cost. If permissions, approvals, and responsibility boundaries are left until the operation scales up, the price will show up two years later. Gartner predicts that in 2027, 40% of enterprises will downgrade or shut down autonomous agents already in operation because of governance gaps, forcing them to redo the process changes they have already made.
This article is adapted from a post on our company’s official WeChat account. The draft is not publicly available. This is not investment advice.